Advances in Knowledge Discovery and Data Mining

Part II 14th edition by Mohammed Zaki , Jeffrey

Xu Yu, Ravindran, Vikram Pudi ISBN 3642136710
978-3642136719 download
https://ebookball.com/product/advances-in-knowledge-discovery-
and-data-mining-part-ii-14th-edition-by-mohammed-zaki-jeffrey-xu-
yu-ravindran-vikram-pudi-isbn-3642136710-978-3642136719-19674/

Download more ebook instantly today at https://ebookball.com

 Get Your Digital Files Instantly: PDF, ePub, MOBI and More
Quick Digital Downloads: PDF, ePub, MOBI and Other Formats

Data Mining and Analysis Fundamental Concepts and Algorithms 1st

Edition by Mohammed Zaki

https://ebookball.com/product/data-mining-and-analysis-
fundamental-concepts-and-algorithms-1st-edition-by-mohammed-
zaki-15688/

Large Scale Parallel Data Mining 1759 Lecture Notes in Computer

Science 1st edition by Mohammed Zaki, Ching Tien Ho ISBN 3540671943
978-3540671947

https://ebookball.com/product/large-scale-parallel-data-
mining-1759-lecture-notes-in-computer-science-1st-edition-by-
mohammed-zaki-ching-tien-ho-isbn-3540671943-978-3540671947-19650/

Knowledge Discovery Practices and Emerging Applications of Data Mining

Trends and New Domains 1st edition by Senthil Kumar
160960069XÂ 9781609600693

https://ebookball.com/product/knowledge-discovery-practices-and-
emerging-applications-of-data-mining-trends-and-new-domains-1st-
edition-by-senthil-kumar-160960069x-9781609600693-14490/

LNAI 2682 A Data Mining Query Language for Knowledge Discovery in a

Geographical Information System 1st Edition by Donato Malerba,
Annalisa Appice, Michelangelo Ceci ISBN 9783540224792 354022479X

https://ebookball.com/product/lnai-2682-a-data-mining-query-
language-for-knowledge-discovery-in-a-geographical-information-
system-1st-edition-by-donato-malerba-annalisa-appice-
michelangelo-ceci-isbn-9783540224792-354022479x-11018/
Managing and Mining Uncertain Data Advances in Database Systems 1st
Edition by Charu Aggarwal ISBN 0387096892 9780387096896

https://ebookball.com/product/managing-and-mining-uncertain-data-
advances-in-database-systems-1st-edition-by-charu-aggarwal-
isbn-0387096892-9780387096896-12012/

Knowledge Discovery from Sensor Data Industrial Innovation 1st Edition

by Auroop R Ganguly ISBN 0367386232 9780367386238

https://ebookball.com/product/knowledge-discovery-from-sensor-
data-industrial-innovation-1st-edition-by-auroop-r-ganguly-
isbn-0367386232-9780367386238-13442/

Knowledge Discovery in Databases PKDD 2007 1st Edition by Anique ISBN

https://ebookball.com/product/knowledge-discovery-in-databases-
pkdd-2007-1st-edition-by-anique-isbn-13694/

Machine Learning in Cyber Trust Security Privacy and Reliability 1st

edition by Jeffrey Tsai, Philip Yu ISBN 0387887342 978-0387887340

https://ebookball.com/product/machine-learning-in-cyber-trust-
security-privacy-and-reliability-1st-edition-by-jeffrey-tsai-
philip-yu-isbn-0387887342-978-0387887340-16710/

Mining the Web Discovering Knowledge from Hypertext Data 1st Edition
by Soumen Chakrabarti ISBN 9780080511726 0080511724

https://ebookball.com/product/mining-the-web-discovering-
knowledge-from-hypertext-data-1st-edition-by-soumen-chakrabarti-
isbn-9780080511726-0080511724-10336/
Lecture Notes in Artificial Intelligence 6119
Edited by R. Goebel, J. Siekmann, and W. Wahlster

Subseries of Lecture Notes in Computer Science

Mohammed J. Zaki Jeffrey Xu Yu
B. Ravindran Vikram Pudi (Eds.)

Advances in
Knowledge Discovery
and Data Mining

14th Pacific-Asia Conference, PAKDD 2010

Hyderabad, India, June 21-24, 2010
Proceedings
Part II

13
Series Editors
Randy Goebel, University of Alberta, Edmonton, Canada
Jörg Siekmann, University of Saarland, Saarbrücken, Germany
Wolfgang Wahlster, DFKI and University of Saarland, Saarbrücken, Germany

Volume Editors
Mohammed J. Zaki
Rensselaer Polytechnic Institute
Troy, NY, USA
E-mail: [email protected]
Jeffrey Xu Yu
The Chinese University of Hong Kong
Hong Kong, China
E-mail: [email protected]
B. Ravindran
IIT Madras, Chennai, India
E-mail: [email protected]
Vikram Pudi
IIIT, Hyderabad, India
E-mail: [email protected]

Library of Congress Control Number: 2010928262

CR Subject Classification (1998): I.2, H.3, H.4, H.2.8, I.4, C.2

LNCS Sublibrary: SL 7 – Artificial Intelligence

ISSN 0302-9743
ISBN-10 3-642-13671-0 Springer Berlin Heidelberg New York
ISBN-13 978-3-642-13671-9 Springer Berlin Heidelberg New York

This work is subject to copyright. All rights are reserved, whether the whole or part of the material is
concerned, specifically the rights of translation, reprinting, re-use of illustrations, recitation, broadcasting,
reproduction on microfilms or in any other way, and storage in data banks. Duplication of this publication
or parts thereof is permitted only under the provisions of the German Copyright Law of September 9, 1965,
in its current version, and permission for use must always be obtained from Springer. Violations are liable
to prosecution under the German Copyright Law.
springer.com
© Springer-Verlag Berlin Heidelberg 2010
Printed in Germany
Typesetting: Camera-ready by author, data conversion by Scientific Publishing Services, Chennai, India
Printed on acid-free paper 06/3180
 Preface

The 14th Pacific-Asia Conference on Knowledge Discovery and Data Mining was
held in Hyderabad, India during June 21–24, 2010; this was the first time the
conference was held in India.
PAKDD is a major international conference in the areas of data mining (DM)
and knowledge discovery in databases (KDD). It provides an international fo-
rum for researchers and industry practitioners to share their new ideas, original
research results and practical development experiences from all KDD-related
areas including data mining, data warehousing, machine learning, databases,
statistics, knowledge acquisition and automatic scientific discovery, data visual-
ization, causal induction and knowledge-based systems.
PAKDD-2010 received 412 research papers from over 34 countries includ-
ing: Australia, Austria, Belgium, Canada, China, Cuba, Egypt, Finland, France,
Germany, Greece, Hong Kong, India, Iran, Italy, Japan, S. Korea, Malaysia,
Mexico, The Netherlands, New Caledonia, New Zealand, San Marino, Singapore,
Slovenia, Spain, Switzerland, Taiwan, Thailand, Tunisia, Turkey, UK, USA, and
Vietnam. This clearly reflects the truly international stature of the PAKDD
conference.
After an initial screening of the papers by the Program Committee Chairs, for
papers that did not conform to the submission guidelines or that were deemed
not worthy of further reviews, 60 papers were rejected with a brief explana-
tion for the decision. The remaining 352 papers were rigorously reviewed by
at least three reviewers. The initial results were discussed among the reviewers
and finally judged by the Program Committee Chairs. In some cases of con-
flict additional reviews were sought. As a result of the deliberation process, only
42 papers (10.2%) were accepted as long presentations (25 mins), and an addi-
tional 55 papers (13.3%) were accepted as short presentations (15 mins). The
total acceptance rate was thus about 23.5% across both categories.
The PAKDD 2010 conference program also included seven workshops: Work-
shop on Data Mining for Healthcare Management (DMHM 2010), Pacific Asia
Workshop on Intelligence and Security Informatics (PAISI 2010), Workshop on
Feature Selection in Data Mining (FSDM 2010), Workshop on Emerging Re-
search Trends in Vehicle Health Management (VHM 2010), Workshop on Behav-
ior Informatics (BI 2010), Workshop on Data Mining and Knowledge Discovery
for e-Governance (DMEG 2010), Workshop on Knowledge Discovery for Rural
Systems (KDRS 2010).
The conference would not have been successful without the support of the
Program Committee members (164), external reviewers (195), Conference Orga-
nizing Committee members, invited speakers, authors, tutorial presenters, work-
shop organizers, reviewers, authors and the conference attendees. We highly
appreciate the conscientious reviews provided by the Program Committee
VI Preface

members, and external reviewers. The Program Committee members were

matched with the papers using the SubSift system (http://subsift.ilrt.bris.ac.uk/)
for bid matching; we thank Simon Price and Peter Flach, of Bristol University,
for developing this wonderful system. Thanks also to Andrei Voronkov for host-
ing the entire PAKDD reviewing process on the easychair.org site.
We are indebted to the members of the PAKDD Steering Committee for their
invaluable suggestions and support throughout the organization process. We
thank Vikram Pudi (Publication Chair), Pabitra Mitra (Workshops Chair), Ka-
mal Karlapalem (Tutorials Chair), and Arnab Bhattacharya (Publicity Chair).
Special thanks to the Local Arrangements Commitee and Chair R.K. Bagga, and
the General Chairs: Jaideep Srivastava, Masaru Kitsuregawa, and P. Krishna
Reddy. We would also like to thank all those who contributed to the success of
PAKDD 2010 but whose names may not be listed.
We greatly appreciate the support from various institutions. The conference
was organized by IIIT Hyderabad. It was sponsored by the Office of Naval Re-
search Global (ONRG) and the Air Force Office of Scientific Research/Asian
Office of Aerospace Research and Development (AFOSR/AOARD).
We hope you enjoy the proceedings of the PAKDD conference, which presents
cutting edge research in data mining and knowledge discovery. We also hope
all participants took this opportunity to share and exchange ideas with each
other and enjoyed the cultural and social attractions of the wonderful city of
Hyderabad!

June 2010 Mohammed J. Zaki

Jeffrey Xu Yu
B. Ravindran
 PAKDD 2010 Conference Organization

Honorary Chair
Rajeev Sangal IIIT Hyderabad, India

General Chairs
Jaideep Srivastava University of Minnesota, USA
Masaru Kitsuregawa University of Tokyo, Japan
P. Krishna Reddy IIIT Hyderabad, India

Program Committee Chairs

Mohammed J. Zaki Rensselaer Polytechnic Institute, USA
Jeffrey Xu Yu The Chinese University of Hong Kong
B. Ravindran IIT Madras, India

Workshop Chair
Pabitra Mitra IIT Kharagpur, India

Tutorial Chairs
Kamal Karlapalem IIIT Hyderabad, India

Publicity Chairs
Arnab Bhattacharya IIT Kanpur, India

Publication Chair
Vikram Pudi IIIT Hyderabad, India

Local Arrangements Committee

Raji Bagga (Chair) IIIT Hyderabad, India
K.S. Vijaya Sekhar IIIT Hyderabad, India
T. Ragunathan IIIT Hyderabad, India
P. Radhakrishna Infosys SET Labs, Hyderabad, India
A. Govardhan JNTU, Hyderabad, India
R.B.V. Subramanyam NIT, Warangal, India
VIII Conference Organization

Program Committee

Osman Abul Joao Gama

Muhammad Abulaish Jean-Gabriel Ganascia
Arun Agarwal Gemma Garriga
Hisham Al-Mubaid Ravi Gupta
Reda Alhajj Mohammad Hasan
Hiroki Arimura Tu Bao Ho
Hideo Bannai Vasant Honavar
Jayanta Basak Wynne Hsu
M.M. Sufyan Beg Xiaohua Hu
Bettina Berendt Akihiro Inokuchi
Fernando Berzal Shen Jialie
Raj Bhatnagar Hasan Jamil
Vasudha Bhatnagar Daxin Jiang
Arnab Bhattacharya Ruoming Jin
Pushpak Bhattacharyya Bo Jin
Chiranjib Bhattacharyya Hiroyuki Kawano
Vivek Borkar Tamer Kahveci
Keith C.C. Chan Toshihiro Kamishima
Longbing Cao Ben Kao
Doina Caragea Panagiotis Karras
Venkatesan Chakaravarthy Hisashi Kashima
Vineet Chaoji Yiping Ke
Sanjay Chawla Latifur Khan
Arbee Chen Hiroyuki Kitagawa
Phoebe Chen Ravi Kothari
Jake Yue Chen Mehmet Koyuturk
Zheng Chen Bharadwaja Kumar
Hong Cheng Wai Lam
James Cheng Chung-Hong Lee
Alok Choudhary Xue Li
Diane Cook Jinyan Li
Alfredo Cuzzocrea Tao Li
Sanmay Das Chun-hung Li
Anne Denton Ee-Peng Lim
Lipika Dey Chih-Jen Lin
Guozhu Dong Guimei Liu
Petros Drineas Tie-Yan Liu
Tina Eliassi-Rad Changtien Lu
Wei Fan Vasilis Megalooikonomou
Eibe Frank Tao Mei
Benjamin C.M. Fung Wagner Meira Jr.
Sachin Garg Rosa Meo
Mohamed Gaber Dunja Mladenic
 Conference Organization IX

Yang-Sae Moon Ambuj K. Singh

Yasuhiko Morimoto Mingli Song
Tsuyoshi Murata P. Sreenivasa Kumar
Atsuyoshi Nakamura Aixin Sun
J. Saketha Nath S. Sundararajan
Juggapong Natwichai Ashish Sureka
Richi Nayak Domenico Talia
Wilfred Ng Kay Chen Tan
Mitsunori Ogihara Ah-Hwee Tan
Salvatore Orlando Pang-Ning Tan
Satoshi Oyama David Taniar
Prabin Panigrahi Ashish Tendulkar
Spiros Papadimitriou Thanaruk Theeramunkong
Srinivasan Parthasarathy W. Ivor Tsang
Wen-Chih Peng Vincent S. Tseng
Bernhard Pfahringer Tomoyuki Uchida
Srinivasa Raghavan Lipo Wang
R. Rajesh Jason Wang
Naren Ramakrishnan Jianyong Wang
Ganesh Ramakrishnan Graham Williams
Jan Ramon Xintao Wu
Sanjay Ranka Xindong Wu
Rajeev Rastogi Meng Xiaofeng
Chandan Reddy Hui Xiong
Patricia Riddle Seiji Yamada
S. Raju Bapi Jiong Yang
Saeed Salem Dit-Yan Yeung
Sudeshna Sarkar Tetsuya Yoshida
Tamas Sarlos Aidong Zhang
C. Chandra Sekhar Zhongfei (Mark) Zhang
Srinivasan Sengamedu Zhi-Hua Zhou
Shirish Shevade Chengqi Zhang
M. Shimbo

External Reviewers
Abdul Nizar Atul Saroop
Abhinav Mishra Blaz Novak
Alessandra Raffaeta Brian Ruttenberg
Aminul Islam Bum-Soo Kim
Andrea Tagarelli Carlo Mastroianni
Anitha Varghese Carlos Ferreira
Ankit Agrawal Carmela Comito
Anuj Mahajan Cha Lun Li
Anupam Bhattacharjee Chandra Sekhar Chellu
X Conference Organization

Chao Luo Leting Wu

Chen-Yi Lin Li Zheng
Chih jui Lin Wang Li An
Chuancong Gao Li Da Jun
Chun Kit Chui Lili Jiang
Chun Wei Seah Ling Guo
Chun-Hao Chen Linhong Zhu
Chung An Yeh Lixin Duan
Claudio Silvestri Lorand Dali
Da Jun Li Luka Bradesko
David Uthus Luming Zhang
De-Chuan Zhan Manas Somaiya
Delia Rusu Mark Beltramo
Dhruv Mahajan Markus Ojala
Di Wang Masayuki Okabe
Dinesh Garg Miao Qiao
Elena Ikonomovska Michael Barnathan
En Tzu Wang Min-Ling Zhang
Eugenio Cesario Mingkui Tan
Feilong Chen Mitja Trampus
Feng Chen Mohammed Aziz
Ferhat Ay Mohammed Imran
Gokhan Yavas Muad Abu-Ata
Gongqing Wu Nagaraj Kota
Hea-Suk Kim Nagarajan Krishnamurthy
Hideyuki Kawashima Narasimha Murty Musti
Hui Zhu Su Narayan Bhamidipati
Ilija Subasic Ngoc Khanh Pham
Indranil Palit Ngoc Tu Le
Jan Rupnik Nicholas Larusso
Janez Brank Nidhi Raj
Jeyashanker Ramamirtham Nikolaj Tatti
Jitendra Ajmera Ning Ruan
Junjie Wu Nirmalya Bandyopadhyay
Kathy Macropol Nithi Gupta
Khalil Al-Hussaeni Noman Mohammed
Kong Wah Wan Palvali Teja
Krishna Prasad Chitrapura Pannagadatta Shivaswamy
Kunpeng Zhang Paolo Trunfio
Kyle Chipman Parthasarathy Krishnaswamy
L. Venkata Subramaniam Pedro P. Rodrigues
Lang Huo Peipei Li
Lei Liu Prahladavaradan Sampath
Lei Shi Prakash Mandayam Comare
Lei Yang Prasad Deshpande
 Conference Organization XI

Prithviraj Sen Subhajit Sanyal

Pruet Boonma Sufyan Beg
Qi Mao Sugato Chakrabarty
Qiang Wang Sundararajan Sellamanickam
Qingyan Yang Tadej Štajner
Quan Yuan Takehiko Sakamoto
Quang Khoat Than Thi Nhan Le
Rahul Chougule Tianrui Li
Ramanathan Narayanan Timothy DeVries
Raquel Sebastiao Toshiyuki Amagasa
Rashid Ali Venu Satuluri
Rui Chen Victor Lee
S. Sathiya Keerthi Vikram Pudi
Shailesh Kumar Vishwakarma Singh
Sai Sundarakrishna Viswanath G
Saikat Dey Wang Wen-Chi
J. Saketha Nath Wei Chu
SakethaNath Jagarlapudi Wei Jin
Salim Akhter Chowdhury Wei Peng
Samah Fodeh Wei Su
Sami Hanhijärvi Wendell Jordan-Brangman
Satnam Singh Wenjun Zhou
Sau Dan Lee Wenting Liu
Sayan Ranu Xiaofeng Zhu
Sergio Flesca Xiaogang Wu
Shafkat Amin Xin Liu
Shailesh Kumar Xing Jiang
Shalabh Bhatnagar Xintian Yang
Shantanu Godbole Xutong Liu
Sharanjit Kaur Yan Zhang
Shazzad Hosain Yanchang Zhao
Shenghua Gao Yang Xiang
Shirish Shevade Yang Zhou
Shirish Tatikonda Yasufumi Takama
Shirong Li Yezhou Yang
Shumo Chu Yilin Kang
Shuo Miao Yin Zhang
Sinan Erten Yong Ge
Sk. Mirajul Haque Yuan Liu
Soumen De Yukai He
Sourangshu Bhattacharya Yuko Itokawa
Sourav Dutta Zakia Sultana
Srinivasan Sengamedu Zubin Abraham
Srujana Merugu
XII Conference Organization

Organized by

IIIT Hyderabad, India

Sponsoring Institutions

AFOSR, USA

AOARD, Tokyo, Japan

ONRG, USA
 Table of Contents – Part II

Session 4B. Dimensionality Reduction/Parallelism

Subclass-Oriented Dimension Reduction with Constraint
Transformation and Manifold Regularization . . . . . . . . . . . . . . . . . . . . . . . . 1
Bin Tong and Einoshin Suzuki

Distributed Knowledge Discovery with Non Linear Dimensionality

Reduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
Panagis Magdalinos, Michalis Vazirgiannis, and Dialecti Valsamou

DPSP: Distributed Progressive Sequential Pattern Mining on the

Cloud . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
Jen-Wei Huang, Su-Chen Lin, and Ming-Syan Chen

An Approach for Fast Hierarchical Agglomerative Clustering Using

Graphics Processors with CUDA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
S.A. Arul Shalom, Manoranjan Dash, and Minh Tue

Session 5A. Novel Applications

Ontology-Based Mining of Brainwaves: A Sequence Similarity Technique
for Mapping Alternative Features in Event-Related Potentials (ERP)
Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
Haishan Liu, Gwen Frishkoff, Robert Frank, and Dejing Dou

Combining Support Vector Machines and the t-statistic for Gene

Selection in DNA Microarray Data Analysis . . . . . . . . . . . . . . . . . . . . . . . . . 55
Tao Yang, Vojislave Kecman, Longbing Cao, and Chengqi Zhang

Satrap: Data and Network Heterogeneity Aware P2P Data-Mining . . . . . 63

Hock Hee Ang, Vivekanand Gopalkrishnan, Anwitaman Datta,
Wee Keong Ng, and Steven C.H. Hoi

Player Performance Prediction in Massively Multiplayer Online

Role-Playing Games (MMORPGs) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
Kyong Jin Shim, Richa Sharan, and Jaideep Srivastava

Relevant Gene Selection Using Normalized Cut Clustering with

Maximal Compression Similarity Measure . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
Rajni Bala, R.K. Agrawal, and Manju Sardana
XIV Table of Contents – Part II

Session 5B. Feature Selection/Visualization

A Novel Prototype Reduction Method for the K-Nearest Neighbor
Algorithm with K ≥ 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
Tao Yang, Longbing Cao, and Chengqi Zhang

Generalized Two-Dimensional FLD Method for Feature Extraction: An

Application to Face Recognition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
Shiladitya Chowdhury, Jamuna Kanta Sing,
Dipak Kumar Basu, and Mita Nasipuri

Learning Gradients with Gaussian Processes . . . . . . . . . . . . . . . . . . . . . . . . . 113

Xinwei Jiang, Junbin Gao, Tianjiang Wang, and Paul W. Kwan

Analyzing the Role of Dimension Arrangement for Data Visualization

in Radviz . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
Luigi Di Caro, Vanessa Frias-Martinez, and Enrique Frias-Martinez

Session 6A. Graph Mining

Subgraph Mining on Directed and Weighted Graphs . . . . . . . . . . . . . . . . . . 133
Stephan Günnemann and Thomas Seidl

Finding Itemset-Sharing Patterns in a Large Itemset-Associated

Graph . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147
Mutsumi Fukuzaki, Mio Seki, Hisashi Kashima, and Jun Sese

A Framework for SQL-Based Mining of Large Graphs on Relational

Databases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
Sriganesh Srihari, Shruti Chandrashekar, and
Srinivasan Parthasarathy

Fast Discovery of Reliable k-terminal Subgraphs . . . . . . . . . . . . . . . . . . . . . 168

Melissa Kasari, Hannu Toivonen, and Petteri Hintsanen

GTRACE2: Improving Performance Using Labeled Union Graphs . . . . . . 178

Akihiro Inokuchi and Takashi Washio

Session 6B. Clustering II

Orthogonal Nonnegative Matrix Tri-factorization for Semi-supervised
Document Co-clustering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189
Huifang Ma, Weizhong Zhao, Qing Tan, and Zhongzhi Shi

Rule Synthesizing from Multiple Related Databases . . . . . . . . . . . . . . . . . . 201

Dan He, Xindong Wu, and Xingquan Zhu
 Table of Contents – Part II XV

Fast Orthogonal Nonnegative Matrix Tri-Factorization for Simultaneous

Clustering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214
Zhao Li, Xindong Wu, and Zhenyu Lu

Hierarchical Web-Page Clustering via In-Page and Cross-Page Link

Structures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 222
Cindy Xide Lin, Yintao Yu, Jiawei Han, and Bing Liu

Mining Numbers in Text Using Suffix Arrays and Clustering Based on

Dirichlet Process Mixture Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 230
Minoru Yoshida, Issei Sato, Hiroshi Nakagawa, and Akira Terada

Session 7A. Opinion/Sentiment Mining

Opinion-Based Imprecise Query Answering . . . . . . . . . . . . . . . . . . . . . . . . . . 238
Muhammad Abulaish, Tanvir Ahmad, Jahiruddin, and
Mohammad Najmud Doja

Blog Opinion Retrieval Based on Topic-Opinion Mixture Model . . . . . . . . 249

Peng Jiang, Chunxia Zhang, Qing Yang, and Zhendong Niu

Feature Subsumption for Sentiment Classification in Multiple

Languages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 261
Zhongwu Zhai, Hua Xu, Jun Li, and Peifa Jia

Decentralisation of ScoreFinder: A Framework for Credibility

Management on User-Generated Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . 272
Yang Liao, Aaron Harwood, and Kotagiri Ramamohanarao

Classification and Pattern Discovery of Mood in Weblogs . . . . . . . . . . . . . 283

Thin Nguyen, Dinh Phung, Brett Adams, Truyen Tran, and
Svetha Venkatesh

Capture of Evidence for Summarization: An Application of Enhanced

Subjective Logic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 291
Sukanya Manna, B. Sumudu U. Mendis, and Tom Gedeon

Session 7B. Stream Mining

Fast Perceptron Decision Tree Learning from Evolving Data Streams . . . 299
Albert Bifet, Geoff Holmes, Bernhard Pfahringer, and Eibe Frank

Classification and Novel Class Detection in Data Streams with Active

Mining . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 311
Mohammad M. Masud, Jing Gao, Latifur Khan, Jiawei Han, and
Bhavani Thuraisingham
XVI Table of Contents – Part II

Bulk Loading Hierarchical Mixture Models for Efficient Stream

Classification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 325
Philipp Kranen, Ralph Krieger, Stefan Denker, and Thomas Seidl

Summarizing Multidimensional Data Streams: A Hierarchy-Graph-

Based Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 335
Yoann Pitarch, Anne Laurent, and Pascal Poncelet

Efficient Trade-Off between Speed Processing and Accuracy in

Summarizing Data Streams . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 343
Nesrine Gabsi, Fabrice Clérot, and Georges Hébrail

Subsequence Matching of Stream Synopses under the Time Warping

Distance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 354
Su-Chen Lin, Mi-Yen Yeh, and Ming-Syan Chen

Session 8A. Similarity and Kernels

Normalized Kernels as Similarity Indices . . . . . . . . . . . . . . . . . . . . . . . . . . . . 362
Julien Ah-Pine

Adaptive Matching Based Kernels for Labelled Graphs . . . . . . . . . . . . . . . 374

Adam Woźnica, Alexandros Kalousis, and Melanie Hilario

A New Framework for Dissimilarity and Similarity Learning . . . . . . . . . . . 386

Adam Woźnica and Alexandros Kalousis

Semantic-Distance Based Clustering for XML Keyword Search . . . . . . . . . 398

Weidong Yang and Hao Zhu

Session 8B. Graph Analysis

OddBall: Spotting Anomalies in Weighted Graphs . . . . . . . . . . . . . . . . . . . 410
Leman Akoglu, Mary McGlohon, and Christos Faloutsos

Robust Outlier Detection Using Commute Time and Eigenspace

Embedding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 422
Nguyen Lu Dang Khoa and Sanjay Chawla

EigenSpokes: Surprising Patterns and Scalable Community Chipping

in Large Graphs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 435
B. Aditya Prakash, Ashwin Sridharan, Mukund Seshadri,
Sridhar Machiraju, and Christos Faloutsos

BASSET: Scalable Gateway Finder in Large Graphs . . . . . . . . . . . . . . . . . . 449

Hanghang Tong, Spiros Papadimitriou, Christos Faloutsos,
Philip S. Yu, and Tina Eliassi-Rad
 Table of Contents – Part II XVII

Session 8C. Classification II

Ensemble Learning Based on Multi-Task Class Labels . . . . . . . . . . . . . . . . 464
Qing Wang and Liang Zhang

Supervised Learning with Minimal Effort . . . . . . . . . . . . . . . . . . . . . . . . . . . 476

Eileen A. Ni and Charles X. Ling

Generating Diverse Ensembles to Counter the Problem of Class

Imbalance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 488
T. Ryan Hoens and Nitesh V. Chawla

Relationship between Diversity and Correlation in Multi-Classifier

Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 500
Kuo-Wei Hsu and Jaideep Srivastava

Compact Margin Machine . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 507

Bo Dai and Gang Niu

Author Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 515

 Subclass-Oriented Dimension Reduction with
Constraint Transformation and Manifold
Regularization

Bin Tong and Einoshin Suzuki

Graduate School of Systems Life Sciences, Kyushu University, Japan

{bintong,suzuki}@i.kyushu-u.ac.jp

Abstract. We propose a new method, called Subclass-oriented Dimension Re-

duction with Pairwise Constraints (SODRPaC), for dimension reduction on high
dimensional data. Current linear semi-supervised dimension reduction methods
using pairwise constraints, e.g., must-link constraints and cannot-link constraints,
can not handle appropriately the data of multiple subclasses where the points of a
class are separately distributed in different groups. To illustrate this problem, we
particularly classify the must-link constraint into two categories, which are the
inter-subclass must-link constraint and the intra-subclass must-link constraint,
respectively. We argue that handling the inter-subclass must-link constraint is
challenging for current discriminant criteria. Inspired by the above observation
and the cluster assumption that nearby points are possible in the same class, we
carefully transform must-link constraints into cannot-link constraints, and then
propose a new discriminant criterion by employing the cannot-link constraints
and the compactness of shared nearest neighbors. For the reason that the local
data structure is one of the most significant features for the data of multiple sub-
classes, manifold regularization is also incorporated in our dimension reduction
framework. Extensive experiments on both synthetic and practical data sets illus-
trate the effectiveness of our method.

1 Introduction

In various applications, such as gene expression, image retrieval, etc., one is often con-
fronted with high dimensional data [1]. Dimension reduction, which maps data points
in a high-dimensional space into those in a low-dimensional space, is thus viewed as
one of the most crucial preprocessing steps of data analysis. Dimension reduction meth-
ods can be divided into three categories, which are supervised ones [2], unsupervised
ones[3], and semi-supervised ones[4]. The input data in these three categories are la-
beled data, unlabeled data, and both of them, respectively. In a typical real-world ap-
plication, only a small number of labeled data points are available due to the high cost
to obtain them [4]. Hence the semi-supervised dimension reduction may be considered
to fit into the practical setting. Instead of labeled data points, some semi-supervised
methods assume pairwise constraints, for it is easier for experts to specify them than to
assign the class labels of data points. More specifically speaking, pairwise constraints
consist of must-link constraints and cannot-link constraints. The pair of data points in

M.J. Zaki et al. (Eds.): PAKDD 2010, Part II, LNAI 6119, pp. 1–13, 2010.

c Springer-Verlag Berlin Heidelberg 2010
2 B. Tong and E. Suzuki

a must-link constraint shares the same class label, while the pair of data points in a
cannot-link constraint is given different class labels.
From another viewpoint, dimension reduction methods can be divided into nonlinear
and linear ones. The former allows a nonlinear transformation in the mapping while
the latter restricts itself to linear transformation. We consider a complex distribution
of points that are distributed in multiple subclasses. In other words, the data points of
one class form several separated clusters. A nonlinear method has a higher degree of
freedom and hence can handle data with complex distribution effectively while a linear
method tends to be incompetent in such a case.
In this paper, we restrict our attention to the linear semi-supervised dimension re-
duction for the data of multiple subclasses with pairwise constraints. Previously rele-
vant methods [5] [6] [7] [8] implicitly assume that a class consists of a single cluster.
If the points are of multiple subclasses, handling the pairwise constraints to project
the points into multiple subclasses in the transformed space is challenging for linear di-
mension reduction. For a deep analysis, we particularly classify the must-link constraint
into two categories. If two points in a must-link constraint reside in a single subclass,
we define such a must-link constraint as an intra-subclass must-link constraint. On the
contrary, if two points in a must-link constraint come from different subclasses, we
define such kind of must-link constraint as an inter-subclass must-link constraint. We
attribute the reason of the improper behavior of current linear methods to the fact that
the inter-subclass must-link constraint most probably confuses the discriminant criteria
of existing methods. The problem resulted from the inter-subclass must-link constraint
is also encountered by the constraint transformation. For instance, a method in [9] trans-
forms multiple must-link constraints, which are connected via points in two different
classes, into a cannot-link constraint between the centroids of the points of two classes.
This method fails to give a comprehensible meaning if the points belong to different
subclasses because the centroids may fall into the region of another class.
To overcome above problems, we propose SODRPaC, which consists of two steps. In
the first step, must-link constraints which satisfy several conditions are transformed into
cannot-link constraints and the remaining must-link constraints are deleted. The idea
behind this step is to reduce the harmfulness of the inter-subclass must-link constraints
while exploiting the must-link constraint information as much as possible by respecting
the cluster assumption [10]: nearby points on the same manifold structure in the original
space are likely to belong to the same class. In the second step, we obtain a projection
mapping by inventing a new discriminant criterion for dimension reduction, which is
suitable for the data of multiple subclasses, and employing the manifold regularization
[11], which is helpful for discovering the local structure of data.

2 Problem Setting and Motivation

The problem setting is defined as follows. We are given a set of N points X = {x1 , x2 ,
. . . , xN }, where xi represents a point in a d-dimensional space, a set of must-link
constraints M = {m1 , m2 , . . . , mNML }, and a set of cannot-link constraints C =
{c1 , c2 , . . . , cNCL }. Here mi consists of a pair of points belonging to the same class
while ci consists of a pair of points belonging to different classes. The output is a
 Subclass-Oriented Dimension Reduction 3

d × l transformation matrix W which consists of l projective vectors {w1 , w2 , . . . , wl }

(l  d). W maps x1 ,x2 ,...,xN to a set of lower dimensional points Y = {y1 , ..., yN }.
Hence yi = W T xi where yi represents a point in a l-dimensional space. After mak-
ing data projection, we only consider the classification task in the transformed space.
For avoiding the bias caused by the choice of the classification method, the accuracy of
nearest neighborhood (1-NN) classifier is considered as a measurement for the good-
ness of dimension reduction with the 20×5-fold cross validation.

SODRPaC 15
10 NPSSDR
SSDR
CLPP 10
5 CMM
Discriminant Criterion
5
0
0

−5 −5

−10
−10
−2 −1 0 1 2 −2 0 2

(a) (b)

Fig. 1. Motivating examples. The data points are of Gaussian distribution. In (a), the blue and
red points are distributed in different clusters. In (b), the red points reside in different subclasses.
Must-link constraints and cannot-link constraints are denoted by black solid and dashed lines,
respectively.

Fig. 1 presents the motivating examples, where d = 2 and l = 1. The task for di-
mension reduction here is thus to project the two dimensional data onto a line, where
the points from different classes can be differentiated. A horizontal line is supposed to
be the best projection while a vertical one is the worst projection. To better illustrate
the motivation of our method, previously relevant methods are firstly retrospected. In
the aspect of pairwise constraints, SSDR [5] and CMM [6] are to maximize the aver-
age distance between the points in cannot-link constraints, and to minimize the average
distance between the points in must-link constraints simultaneously. We can see that
minimizing the average distance between the points in must-link constraints is reason-
able in the case shown in Fig. 1a, where all the must-link constraints are intra-subclass
must-link constraints. However, it disturbs to maximize the average distance between
the points in cannot-link constraints in the case shown in Fig. 1b, where all the must-
link constraints are inter-subclass must-link constraints. CLPP [7] builds an affinity
matrix, each entry of which indicates the similarity between two points. To utilize the
constraint information, the affinity matrix is revised by setting the similarity degree be-
tween non-neighboring points involved in pairwise constraints. For example, given a
must-link constraint, the similarity degree between two points is revised to be 1, indi-
cating two points are close (similar) to each other, no matter the two points are distant
(dissimilar) or not. Suppose that the must-link constraint is inter-subclass must-link
constraint, it implies that the two points are not geometrically nearby each other. This
4 B. Tong and E. Suzuki

arbitrary updating may damage the geometrical structure of data points. This problem
is also confronted by NPSSDR [8]. The above analysis explains the reason why CMM,
SSDR, CLPP and NPSSDR are capable of obtaining excellent performance as shown
in Fig. 1a, while they fail to reach the same fine performance in the multiple subclass
case shown in Fig. 1b.
In the light of observations, we argue that the inter-subclass must-link constraint is
probably harmful for the discriminant criteria of existing methods. For this reason, we
attempt to design a new discriminant criterion that is able to behave appropriately in the
case of multiple subclasses. The new discrimination criterion marked as ‘Discriminant
Criterion’ is able to obtain almost the same performance as others, as shown in Fig.
1a, and can even outperform previous methods, as shown in Fig. 1b. Moreover, the
manifold regularization is helpful for discovering the local structure of data which is
considered as one of the most principle characteristics of the data of multiple subclasses
[12]. We therefore consider to make the new discriminant criterion and the manifold
regularization work together in a collaborative way. Fig. 1b also demonstrates that our
method SODRPaC, which is the combination of the new discrimination criterion and
the manifold regularization, can obtain the best performance.

3 Subclass-Oriented Dimension Reduction with Pairwise

Constraints
The overview of our SODRPaC involves two steps described as follows:
(1) Transformation. This step transforms must-link constraints into cannot-link con-
straints under the cluster assumption.
(2) Dimension reduction. This step includes two components. The first component
is the new discriminant criterion suitable for the case of multiple subclasses. The
other one is the manifold regularization, which helps discovering the local structure
of data.

3.1 Transformation from Must-Link Constraints

Although a method that transforms must-link constraints into cannot-link constraints
is provided in [9], we would point out that its purpose that the plentiful amount of
constraints are reduced is substantially different from ours. Moreover, it becomes in-
effective due to the inter-subclass must-link constraint. In a high dimensional space,
the boundaries of subclasses and the number of subclasses within one class can not be
explicitly detected by using the unlabeled data and link constraints. Thus, it is difficult
to identify whether a must-link constraint is of inter-subclass must-link constraint or
not. To reduce the harmfulness of inter-subclass must-link constraints, removing all the
must-link constraints is, therefore, the most straightforward way. However, it can be
regarded as a waste of must-link constraint information. Preserving the useful must-
link constraints as much as possible in the form of cannot-link constraints is then the
fundamental idea behind the transformation.
In our method, the transformation from must-link constraints into cannot-link con-
straints basically occurs when a must-link constraint and a cannot-link constraint are
 Subclass-Oriented Dimension Reduction 5

connected. Under the cluster assumption, it is natural to consider two nearby points as
another form of must-link constraint, so that we have more opportunities to transform
must-link constraints into cannot-link constraints. In this paper, we employ shared near-
est neighbor (SNN) [13] to formulate the sense of ‘nearby’ points. A set of shared near-
est neighbors is denoted by NS = {NxS1 , NxS2 , . . . , NxSN } where NxSi = {{xi, xj }|xi ∈
N(xj ), xj ∈ N(xi )}. N(xi ) denotes the k nearest neighbors set of xi . Let |NS | be the
number of the pairs of shared nearest neighbors, where |·| denotes the cardinality of a
set. The value of SNN between xi and xj is defined as the number of points shared by
their neighbors SN N (i, j) = |N(xi ) ∩ N(xj )|. The larger the value of SNN between
two points is, the closer the two points are. It should be noted that we design a N × N
matrix L to specify a kind of reliability for cannot-link constraints, which could be also
deemed as the trustiness to them. Suppose that all the previously specified constraints
are correct, for the previously given cannot-link constraints and the generated cannot-
link constraints by using must-link constraints, their reliabilities are set to be 1. For the
generated cannot-link constraints by using shared nearest neighbors, their reliabilities
are equal to the similarities between the shared nearest neighbors. It is because that
transformation by employing shared nearest neighbors are considered to be less trustful
than that by using must-link constraints. We believe it is natural to take the similarity
between the shared nearest neighbors as a measurement for the trustiness. For exam-
ple, given a pair of shared nearest neighbors {xi , xj }, we represent the reliability of
a generated cannot-link constraint by using it as a Gaussian kernel, which is a simple
kernel and has been widely applied in research fields. The reliability is formulated as
θ(xi , xj ) = exp(− xi − xj 2 /γ), where · denotes the Euclidian norm and γ is the
kernel parameter. Note that, for the convenient access to the matrix L, given a cannot-
link constraint c = {xi , xj }, we use L(c) to denote the entries of Lij and Lji , thus L
is a symmetric matrix.

a c e a c e a c e a c e

b d f b d f b d f b d f

(a) (b) (c) (d)

Fig. 2. Four simple cases for the transformation. The previously specified must-link constraints
and cannot-link constraints are denoted by the black solid and dashed lines, respectively. The
shared nearest neighbor is presented as the blue dotted line. The red dash-dotted line specifies the
new cannot-link constraint.

Fig. 2 shows four fundamental scenarios of the transformation. The set {a, b, e, f},
and {c, d} represent different classes of data points. We explain these four scenarios in a
metaphorical way where the must-link constraint is taken as a friend relationship while
the cannot-link constraint is considered as an enemy one. Standing from the viewpoint
of point ‘a’, it is given a friend relationship, say {a, e}, as shown in Fig. 2a, which is
6 B. Tong and E. Suzuki

called as a basic form. If ‘d’ is my enemy, instead of keeping my friend ‘e’, consider
that ‘e’ is the enemy of my enemy ‘d’. Fig. 2b shows an extension of the basic form
with an enemy’s friend rule. If my enemy ‘d’ has a friend ‘c’, ‘c’ is the enemy of my
friend ‘e’ and me. In these two cases, the reliabilities for the new enemy relationships
are set to be 1. Fig. 2c presents an extension of the basic form, which is called as a
proximity form. If I have no enemy but my neighbor ‘b’ has an enemy ‘d’, ‘d’ is the
enemy of my friend ‘e’ and me. Fig. 2d shows an extension of the proximity form with
the enemy’s friend rule. Note that, in the latter two cases, the reliabilities for the new
enemy relationships are set to be the similarity between my neighbor ‘b’ and me. The
pseudo code for the summary of these four cases is illustrated in Algorithm 1.

Algorithm 1. Transformation from Must-link Constraints into Cannot-link Constraints

Input: M, C, k, γ.
Output: C, L.
1: create a N × N zero matrix L.
2: for each c ∈ C do
3: L(c) = 1.
4: end for
5: if ∃ c ∈ C, m ∈ M s.t. m ∩ c = ∅ then
6: define a ∈ m ∩ c, e ∈ m − m ∩ c, d ∈ c − m ∩ c.
7: create a new cannot-link constraint c = {d, e}; if c ∈ / C then C ← C∪{c }, L(c ) = 1.
  
8: if ∃ m ∈ M s.t. d ∈ m , m = {d, e} then
9: define c ∈ m − m ∩ c.
10: create two new cannot-link constraints c1 = {a, c}, c2 = {e, c}; for each ci , i = 1, 2,
if ci ∈/ C, then C ← C ∪ {ci }, L(ci ) = 1.
11: end if
12: end if
13: if ∃ m ∈ M, c ∈ C, ∀a ∈ m, ∀naS ∈ NaS s.t. c ∈ / NaS , c ∩ naS = ∅, a ∈
/ c ∩ naS then
14: define d ∈ c − c ∩ naS , e ∈ m − m ∩ c.
15: create two new cannot-link constraints c1 = {a, d}, c2 = {e, d} and r = θ(a, b); for
each ci , i = 1, 2, if ci ∈
/ C, then C ← C ∪ {ci }, L(ci ) = r.
16: if ∃ m ∈ M s.t. d ∈ m and m = {d, e} then
17: define c ∈ m − m ∪ c.
18: create two new cannot-link constraints c1 = {a, c}, c2 = {e, c} and r = θ(a, b); for
each ci , i = 1, 2, if ci ∈/ C, then C ← C ∪ {ci }, L(ci ) = r.
19: end if
20: end if

3.2 Dimension Reduction

In this section, we explain the dimension reduction which is based on a novel discrim-
inant criterion and the manifold regularization. As mentioned in section 2, minimizing
the average distance between the points in must-link constraints is inappropriate when
the must-link constraints are inter-subclass must-link constraints. Under the cluster as-
sumption, the shared nearest neighbors could be naturally deemed as another kind of
intra-subclass must-link constraints. Thus, minimizing the average distance between
the points in intra-subclass must-link constraints could be relaxed as making the shared
 Subclass-Oriented Dimension Reduction 7

nearest neighbors closer in the transformed space. Furthermore, the pair of points in
the shared nearest neighbors probably resides in the same subclass, such that this re-
laxation would not suffer from the harmfulness of inter-subclass must-link constraints.
Therefore, the discriminant criterion, which maximizes the average distance between
the points in cannot-link constraints and minimizes the average distance between the
shared nearest neighbors, is expected to be suitable for the data of multiple subclasses.
Suppose that xi and xj are projected to the image yik and yjk along the direction wk ,
the new discriminant criterion is defined as follows:
 k   k 
 y − y k 2  y − y k 2
i j i j
∂(wk ) = Lij − Hij (1)
2|C| 2|NS |
i,j:{xi ,xj }∈C i,j:{xi ,xj }∈NS

where the elements of H are given below:


SN N (i, j), {xi , xj } ∈ NxSi
Hij = (2)
0, otherwise

Inspired by the local scatter [14], the intuition behind the latter part of the right side
of Eq. 1 could be regarded as the compactness of shared nearest neighbors, since two
points are more likely to be close if the value of SNN between them is large. The differ-
ence from the local scatter lies in the fact that a weighted matrix H which handles the
similarity degree between shared nearest neighbors is employed. Since SNN provides
a robust property that the side effect caused by the noisy points could be reduced to
some degree, the compactness of shared nearest neighbors is more reliable than that of
local scatter. The compactness of shared nearest neighbors could be also re-written as
follows:
 k 
 y − y k 2 1 
i j
Hij = Hij (wkT xi − wkT xj )2
2|NS | 2|NS | i j
i,j:{xi, xj }∈NS
⎡ ⎤
1 
= wkT ⎣ Hij (xi − xj )(xi − xj )T ⎦ wk
2|NS |
i j

= wkT S1 wk (3)
1
where S1 = 2|NS | Hij (xi − xj )(xi − xj )T . S1 then could be computed as
i j
follows:
⎛ ⎞
1 ⎝   
S1 = Hij xi xTi + Hij xj xTj − 2 Hij xi xTj ⎠
2|NS |
i j i j i j
⎛ ⎞
1 ⎝ 
= Dii xi xTi − Hij xi xTj ⎠
|NS | i i j

1  
= XDX T − XHX T (4)
|NS |
8 B. Tong and E. Suzuki

where D is a diagonal matrix whose entries are column sums of H, Dii = Hij .
j
Similarly, the first part of right hand of Eq. 1 could be reformulated as:
 k 
 y − y k 2 1 
i j
Lij = Lij (wkT xi − wkT xj )2
2|C| 2|C| i j
i,j:{xi, xj }∈C

= wkT S2 wk (5)
1
 
where S2 = |C| XGX T − XLX T where G is a diagonal matrix whose entries are
column sums of L, Gii = Lij . Then, ∂(wk ) can be briefly written as:
j

∂(wk ) = wkT X(P − Q)X T wk (6)

where P = D − H, and Q = G − L. For all the wk , k = 1, ..., l, we can arrive at
 
∂ = tr W T X(P − Q)X T W (7)
where tr denotes the trace operator. As illustrated in Fig. 1b, the manifold regularization
[11] is helpful for enhancing the performance obtained by the new discriminant crite-
rion. We therefore incorporate it into our dimension reduction framework. The manifold
regularization is defined as:
 
ξ = tr W T XMX T W (8)
where M = I − K−1/2 UK−1/2 is defined as a normalized graph Laplacian. I is a unit
matrix, and K is a diagonal matrix whose entries are column sums of U, Kii = Uij ,
j
where U is defined as follows:
 2
exp(xi − xj  /γ), xi ∈ N(xj )
Uij = (9)
0, otherwise.
ξ is expected to be minimized in order to preserve the sub-manifold of data. At last,
the final objective function that combines Eq. 7 and Eq. 8 together is expected to be
maximized, and is derived as
 
arg max tr W T X(P − Q − λM)X T W (10)
W ∈Rd×l
s.t.W T W =I

where λ is a parameter to control the impact of manifold regularization. By introducing

the Lagrangian, the objective function is given by the maximum eigenvalue solution to
the following generalized eigenvector problem:
X(P − Q − λM)X T w = φw (11)
where φ is the eigenvalue of P − Q − λM , and w is the corresponding eigenvector.
One may argue that, when the graph of SNN is equal to the k-NN graph of the manifold
regularization, Q is almost equivalent to M on preserving the local structure. As shown
in [13], this situation would rarely happen since the two types of graph are dramatically
different from each other in the general case. Moreover, to minimize the average dis-
tance between the shared nearest neighbors, which are considered as another form of
must-link constraints, is conceptually distinct from preserving the local structure.
 Subclass-Oriented Dimension Reduction 9

4 Evaluation by Experiments
4.1 Experiments Setup
We use public data sets to evaluate the performance of SODRPaC. Table 1 summarizes
the characteristics of the data sets. All the data come from the UCI repository [15]
except for GCM [16] that is of very high dimensionality. For the ‘monks-1’, ‘monks-2’,
and ‘monks-3’ data, we combined the train and test sets into a whole one. For the ‘letter’
data, we chose ‘A’, ‘B’, ‘C’, and ‘D’ letters from the train and test sets respectively by
randomly picking up 100 samples for each letter, and then assembled them into a whole
set.

Table 1. Summary of the benchmark data sets

Data set Dimension Instance Class Data set Dimension Instance Class
monks-1 6 556 2 monks-2 6 601 2
monks-3 6 554 2 letter(ABCD) 16 800 4
heart 13 270 4 GCM 16063 198 14

As shown in Eq. 10, λ is the parameter that controls the balance between P − Q
and M. In this experiments setting, the parameter λ is searched from 2α , where α ∈
{α| − 5 ≤ α ≤ 10, α ∈ Z}. A weighted 5-nearest-neighbor graph is employed to
construct the manifold regularizer. In addition, the kernel parameter γ follows the sug-
δ2 δ 2 δ2 δ2 2
gestion in [17] that it is searched from the grid { 16 , 8 , 4 , 2 ,δ , 2δ 2 ,4δ 2 ,8δ 2 ,16δ 2 },
where δ is the mean norm of data. The parameter λ and the manifold regularizer are
then optimized by means of the 5-fold cross-validation. As to the parameter settings
of other competitive methods, we follow the parameters recommended by them, which
are considered to be optimal. Without specific explanation, the number of must-link
constraints is always set to be equal to that of cannot-link constraints, as the equal
equilibrium between must-link constraints and cannot-link constraints is favorable for
the existing methods. In addition, the value of k for searching shared nearest neigh-
bors is set to be 3. The reason of this setting is to guarantee that the pairs of points
in shared nearest neighbors reside in the same subclass, and to make the constraint
transformation have more opportunities to be performed. In our experiments, must-link
constraints and cannot-link constraints are selected according to the ground-truth of
data labels.

4.2 Analysis of Experiments

First, the effectiveness of SODRPaC is exhibited by changing the number of constraints.
Apart from the semi-supervised dimension reduction methods, we also take PCA as the
baseline. As illustrated in Fig. 3, SODRPaC always keeps the best performance when
the number of available constraints increases from 10 to 150. As seen in Fig. 3a, Fig. 3b,
Fig. 3d, and Fig.3f, CLPP is inferior to PCA even if the number of constraints is small.
The side effect of inter-subclass must-link constraints, in this case, can be neglected.
10 B. Tong and E. Suzuki

0.7 0.88 0.9

0.86 0.88
0.65
0.84
0.86

Accuracy
Accuracy

Accuracy
0.82
0.6 0.84
0.8
SODRPaC
NPSSDR 0.82
0.78
0.55 SSDR
CLPP 0.8
0.76
CMM
PCA
0.5 0.74 0.78
20 40 60 80 100 120 140 20 40 60 80 100 120 140 20 40 60 80 100 120 140
Number of Constraints Number of Constraints Number of Constraints

(a) monks-1 (d=2) (b) monks-2 (d=4) (c) monks-3 (d=5)

0.79 0.995 0.6

0.78 0.99
0.5
0.985
0.77
0.98 0.4
Accuracy

Accuracy
Accuracy

0.76
0.975
0.75
0.3
0.97
0.74
0.965
0.2
0.73
0.96

0.72 0.955 0.1

20 40 60 80 100 120 140 20 40 60 80 100 120 140 20 40 60 80 100 120 140
Number of Constraints Number of Constraints Number of Constraints

(d) heart (d=10) (e) letter(abcd) (d=10) (f) GCM (d=150)

Fig. 3. The performance with different numbers of constraints (d: reduced dimensionality)

The reason is probably that the feature of discovering the local structure of data points
could not help CLPP to outperform PCA. However, our SODRPaC, which also utilizes
the manifold regularization due to its property of discovering the local structure, ob-
tains the best performance. We can judge that the new discriminant criterion boosts the
performance. It is also presented in Fig. 3d that the performance of SSDR decreases
to some extent with the increase of the number of constraints. The possible reason is
that increasing the number of available constraints makes the opportunity higher that
inter-subclass must-link constraints exist, which deteriorates the optimization on the
fine dimension reduction. It should be also pointed out that SODRPaC does not sig-
nificantly outperform other methods. A possible reason is that the Euclidean distance,
which is employed to formulate the similarity between points in the original space, is
likely to be meaningless in the high dimensional space.
We then examine the relative impact between must-link constraints and cannot-link
constraints on the performance of SODRPaC. In this experiment, given 150 available
constraints, the ratio of must-link constraints to cannot-link constraints is varied from 0
to 1. Fig. 4 presents that SODRPaC has a much smoother behavior than others with the
change of ratio. It indicates that SODRPaC is more robust than other semi-supervised
methods in terms of the imbalance between must-link constraints and cannot-link con-
straints. As shown in Fig. 4b and Fig. 4f, SODRPaC presents an obvious degradation of
 Subclass-Oriented Dimension Reduction 11

0.7 0.9 1

0.68 0.85
0.95
0.66 0.8

0.64 0.75 0.9

Accuracy
Accuracy

Accuracy
0.62 0.7

0.6 SODRPaC 0.65 0.85

NPSSDR
0.58 SSDR 0.6
CLPP 0.8
0.56 CMM 0.55
PCA
0.54 0.5 0.75
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
Rate of Must−link Set Rate of Must−link Set Rate of Must−link Set

(a) monks-1 (d=2) (b) monks-2 (d=4) (c) monks-3 (d=5)

0.79 1 0.8

0.78 0.7
0.77 0.95
0.6
0.76
Accuracy

Accuracy

Accuracy
0.5
0.75 0.9
0.4
0.74
0.3
0.73 0.85

0.72 0.2

0.71 0.8 0.1

0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
Rate of Must−link Set Rate of Must−link Set Rate of Must−link Set

(d) heart (d=10) (e) letter (abcd) (d=10) (f) GCM (d=150)

Fig. 4. The performance with the change of rate for must-link set (d: reduced dimensionality)

performance when all constraints are must-link ones. The most probable reason would
be that the transformation from must-link constraints into cannot-link constraints can
not be performed when the necessary cannot-link constraints lack. This behavior is
consistent with the conclusion demonstrated in [9] that cannot-link constraints are more
important than must-link constraints in guiding the dimension reduction.
As implicated in the previous sections, the parameter λ that controls the balance
between P − Q and M, and the factor γ that is related to computing the similarity
between two points would influence the performance of SODRPaC. An analysis on
the two parameters is necessary to provide the guideline about how to choose their
values. PCA is employed as the baseline because existing methods can not hold such
two parameters simultaneously. Because of the different scale between λ and γ, λ-axis
and γ-axis are thus plotted as λ/(1 + λ) and γ/(1 + γ), respectively. The axis values
are then in the interval (0, 1). We empirically uncover two interesting patterns for most
of data sets and reduced dimensions as well. There are two regions where SODRPaC
are more likely to obtain its best performance. The first region is where λ/(1 + λ) is
small, as shown Fig. 5a, Fig. 5b, Fig. 5c, Fig. 5d, Fig. 5e and Fig. 5g. In this situation,
the variation of γ/(1 + γ) would not cause the dramatic change for the performance
of SODRPaC. The second region is where both λ/(1 + λ) and γ/(1 + γ) are large, as
shown in Fig. 5b, Fig. 5e, Fig. 5f, and Fig. 5h.
12 B. Tong and E. Suzuki

SODRPaC
0.8 PCA 0.9
0.85
0.9
0.8
0.7 0.8 0.85

Accuracy

Accuracy
Accuracy
Accuracy

0.75
0.8
0.6 0.7
0.7
0.75
0.65
0.6 0.7
0.5

0.8 0.8 0.8

0.8 0.5 0.6 0.5 0.6 0.5 0.6
0.5 0.6 0.4 0.4 0.4
0.4 0.2 0.2 0.2
1 0.2 1
λ/(1+λ)
1
λ/(1+λ)
1
γ/(1+γ) λ/(1+λ) γ/(1+γ) γ/(1+γ) γ/(1+γ) λ/(1+λ)

(a) monks-1 (d=3) (b) monks-1 (d=4) (c) monks-2 (d=3) (d) monks-2 (d=4)

0.8 0.9
0.99
0.75 0.85 0.8
Accuracy

Accuracy

Accuracy

Accuracy
0.7 0.8
0.7 0.98
0.65 0.75
0.6 0.97
0.6 0.7

0.8 0.8
0.5 0.6 0.5 0.6 0.5 0.6 0.8 0.5 0.6 0.8
0.2
0.4
0.2
0.4
1 0.2 0.4 0.2 0.4
1 1 1
γ/(1+γ) λ/(1+λ) λ/(1+λ) λ/(1+λ) γ/(1+γ) λ/(1+λ)
γ/(1+γ) γ/(1+γ)

(e) monks-3 (d=3) (f) monks-3 (d=4) (g) letter(abcd) (d=3) (h) letter(abcd) (d=15)

Fig. 5. The analysis for λ and γ (d: reduced dimensionality)

5 Conclusions and Future Works

In this paper, we have proposed a new linear dimension reduction method with must-
link constraints and cannot-link constraints, called SODRPaC, that can deal with the
multiple subclasses data. Inspired by the observation that handling the inter-subclass
must-link constraint is challenging for the existing methods, a new discriminant crite-
rion is invented by primarily transforming must-link constraints into cannot-link con-
straints. We also combine the manifold regularization into our dimension reduction
framework. The results of extensive experiments show the effectiveness of our method.
There are some other aspects of this work that merit further research. Although the
empirical choice of λ and γ is suggested, we do not as yet have a good understanding
of how to choose these two parameters which are also correlated with choice of the
number of the reduced dimensionality. Therefore, we are interested in automatically
identifying these three parameters and uncovering relationships among them. Another
possible would be to integrate the semi-supervised dimension reduction and clustering
in a joint framework with automatic subspace selection.

Acknowledgments. A part of this research is supported by the Strategic International

Cooperative Program funded by Japan Science and Technology Agency (JST) and by
the grant-in-aid for scientific research on fundamental research (B) 21300053 from the
Japanese Ministry of Education, Culture, Sports, Science and Technology. Bin Tong is
sponsored by the China Scholarship Council (CSC).
 Subclass-Oriented Dimension Reduction 13

References
1. Parsons, L., Haque, E., Liu, H.: Subspace Clustering for High Dimensional Data: A Review.
In: SIGKDD Explorations, pp. 90–105 (2004)
2. Fukunaga, K.: Introduction to Statistical Pattern Recognition. Academic Press, San Diego
(1990)
3. Joliffe, I.: Principal Component Analysis. Springer, New York (1986)
4. Zhu, X.: Semi-supervised Learning Literature Survey. Technical Report Computer Sciences
1530, University of Wisconsin-Madison (2007)
5. Zhang, D., Zhou, Z.H., Chen, S.: Semi-supervised Dimensionality Reduction. In: Proceed-
ings of the 7th SIAM International Conference on Data Mining (2007)
6. Wang, F., Chen, S., Li, T., Zhang, C.: Semi-supervised Metric Learning by
Maximizing Constraint Margin. In: ACM 17th Conference on Information and Knowledge
Management, pp. 1457–1458 (2008)
7. Cevikalp, H., Verbeek, J., Jurie, F., Klaser, A.: Semi-supervised Dimensionality Reduction
Using Pairwise Equivalence Constraints. In: International Conference on Computer Vision
Theory and Applications (VISAPP), pp. 489–496 (2008)
8. Wei, J., Peng, H.: Neighborhood Preserving Based Semi-supervised Dimensionality Reduc-
tion. Electronics Letters 44, 1190–1191 (2008)
9. Tang, W., Xiong, H., Zhong, S., Wu, J.: Enhancing Semi-supervised Clustering: A Feature
Projection Perspective. In: Proc. of the 13th ACM SIGKDD International Conference on
Knowledge Discovery and Data Mining, pp. 707–716 (2007)
10. Zhou, D., Bousquet, O., Lal, T.N., Weston, J., Schölkopf, B.: Learning with Local and Global
Consistency. In: Proc. Advances in Neural Information Processing Systems, pp. 321–328
(2004)
11. Belkin, M., Niyogi, P.: Manifold Regularization: A Geometric Framework for Learning From
Labeled and Unlabeled Examples. Journal of Machine Learning Research 7, 2399–2434
(2006)
12. Sugiyama, M.: Dimensionality Reduction of Multimodal Labeled Data by Local Fisher Dis-
criminant Analysis. Journal of Machine Learning Research 8, 1027–1061 (2007)
13. Ertöz, L., Steinbach, M., Kumar, V.: A New Shared Nearest Neighbor Clustering Algorithm
and its Applications. In: Proc. of the Workshop on Clustering High Dimensional Data and its
Applications, Second SIAM International Conference on Data Mining (2002)
14. Yang, J., Zhang, D., Yang, J.Y., Niu, B.: Globally Maximizing, Locally Minimizing: Un-
supervised Discriminant Projection with Applications to Face and Palm Biometrics. IEEE
Trans. Pattern Analysis and Machine Intelligence 29(4), 650–664 (2007)
15. Blake, C., Keogh, E., Merz, C.J.: UCI Repository of Machine Learning Databases,
http://www.ics.uci.edu/˜mlearn/MLRRepository.html
16. Ramaswamy, S., Tamayo, P., Rifkin, R., et al.: Multiclass Cancer Diagnosis Using Tu-
mor Gene Expression Signatures. Proceedings of the National Academy of Sciences,
15149–15154 (1998)
17. He, X., Yan, S., Hu, Y., Niyogi, P.: Face Recognition Using Laplacianfaces. IEEE Transaction
on Pattern Analysis and Machine Intelligence 27, 316–327 (2005)
Other documents randomly have
different content
 The Meeting Of Mark And Johanna.

It would be quite impossible, if we had the will to attempt it, for us

to go through the scene that took place between Johanna Oakley
and Mark Ingestrie in the magistrate's parlour. For about half an
hour they quite forgot where they were, or that there was any one
in the world but themselves. At the end of that period of time,
though, Sir Richard Blunt gently walked into the room.
"Well," he said, "have you come to any understanding about that
military man in the Temple Gardens?"
Johanna sprang towards the magistrate, and placing her arms upon
his breast, she kissed him on the cheek.
"Sir," she said, "you are our very dear friend, and I love you as I love
my father."
"God bless you!" said Sir Richard, "You have, by those few words,
more then repaid me for all that I have done. Are you happy?"
"Very, very happy."
"So very happy, sir," said Ingestrie, as his eyes glistened through
tears of joy, "that I can hardly believe in its reality."
"And yet you are both so poor."
"Ah, sir, what is poverty when we shall be together?"
"We will face that foe, Mark, I think," said Johanna, with a smile,
"and he shall not extort a tear from us."
"Well," said Sir Richard, as he opened his desk, "since you are not to
be knocked down by poverty, what say you to riches? Do you know
these, Mr. Ingestrie?"
"Why, that is my String of Pearls."
"Yes. I took this from Todd's escritoire myself, and they are yours
and Johanna's. Will you permit me always to call you Johanna?"
"Oh, yes—yes. Do so. All who love me call me Johanna."
"Very well. This String of Pearls, I have ascertained, is worth a
sufficient sum to place you both very far above all the primary
exigences of life. It will be necessary to produce them at the trial of
Sweeney Todd, but after that event they will be handed to you to do
what you please with them, when you can realise them at at once,
and be happy enough with the proceeds."
"If my poor friend, Thornhill," sighed Mark Ingestrie, "could but have
lived to see this day!"
"That, indeed, would have been a joy," said Johanna.
"Yes," said the magistrate; "but the grave has closed on his poor
remains—at least, I may say so figuratively. He was one of Todd's
victims, one of his numerous victims; for I do believe that, for a long
time, scarcely a week passed that did not witness some three or four
murders in that man's shop."
"Horrible!"
"You may well use that expression, in speaking of the career of
Sweeney Todd. It has been most horrible; but there cannot be a
doubt of his expiating his crimes upon the scaffold, together with his
partner in guilt, Mrs. Lovett."
Mark Ingestrie gave a shudder as that woman's name was
mentioned, for it put him in mind of the cellar where he had lived so
long, and where it was only by the most good fortune that he had
not terminated his career.
Before they could say any more, one of the officers in attendance
upon Sir Richard, announced Colonel Jeffery.
"Ah, that is your dreadful military rival," said Sir Richard to Ingestrie.
"That is the gentleman whom you saw in the garden of the Temple
with Johanna."
"I have much to thank him for. His conduct to Johanna has been
most noble."
The colonel smiled when he saw Mark Ingestrie and Johanna, for he
well knew, from private information he had got from the magistrate,
that Mark Ingestrie and Mrs. Lovett's cook were identical; and
holding out his hand to the young man, he said—
"Accept of my best and sincerest wishes, Mr. Ingestrie."
"And you, sir," said Mark, "accept of my best thanks. Our gratitude is
largely due to you, sir."
"I am quite repaid by this very happy result; and I have the pleasure
of informing you, Sir Richard, that poor Tobias is very much better
indeed."
"Which I am rejoiced to hear," said Sir Richard. "And now, my dear
Johanna, it is time for you to go home. You will hear from me in the
morning, for I intend to do myself the pleasure of calling upon your
father, and explaining all to him; for there are some circumstances
that he is yet in ignorance of, and particularly concerning Mr.
Ingestrie."
"I will walk with you to your door, Johanna," said Mark rising and
tottering.
"No," said Sir Richard Blunt; "that must not be to-night. Do not let
him, Johanna. He is by far too weak and unwell to do anything of
the kind. A calm and long night's rest here will do him a world of
good. Business prevents me from leaving the office; but I daresay
the colonel will see Johanna in safety."
"With pleasure," said Colonel Jeffery, "if Mr. Ingestrie has no
objection to my doing so."
"Sir," said Mark, "there is no one in all the world that I would more
cheerfully see protecting my Johanna. I feel that I am in too great a
state of exhaustion to go out. I leave her to your care, sir."
"That is right," said Sir Richard Blunt. "Now, good-night, Johanna,
and God bless you. You will see me in the morning, recollect."
Mark Ingestrie took a parting embrace of Johanna, and then she
went off with the colonel, who, on their road home, told her how he
and Arabella had got so far as to fix their wedding day, and how he
should not feel at all happy unless both she and Mark Ingestrie were
at the ceremony.
"Indeed, he hoped," he said, "that they might give the parson only
one trouble, by being married upon the same occasion."
Johanna warded this last part of the colonel's speech; but she was
fervent in her hopes that he and Arabella would be so very happy,
and in her praises of her young friend; so in very pleasant discourse
indeed, they reached the old spectacle-maker's shop, and then the
colonel shook hands with Johanna, and bade her a kind and friendly
adieu, and she was let in by—to her immense surprise—her mother!
Mrs. Oakley fell upon Johanna's neck in a passion of tears, crying—
"Come, my child—come to your mother's heart, and tell her that you
forgive her for much past neglect and unkindness."
"Oh, mother," said Johanna, "do not speak so. There is nothing to
forgive; and if you are happy and we are all good friends, we will
never think of the past."
"That's right, my dear," said Mr. Oakley, from the passage; "that's
right, my love. Come in, both of you." But it is necessary that we
should briefly state how it was that this wonderful change in the
behaviour of Mrs. Oakley came about, and for that purpose we must
retrace our steps a little.
The reader will be so good as to recollect that the last time Mrs.
Oakley was introduced to his notice she was encumbered by Mr.
Lupin, and had the pleasure of introducing that gentleman to the
notice of Big Ben the beef-eater, who had quickly put all idea of
escape out of the question, as regarded that highly religious
personage.
At that point the presence of other events compelled us to leave the
lady, and repair to Todd's shop, and to Mrs. Lovett's little concern in
Bell Yard.
The appearance of Lupin's face when he found that he was in the
grasp of Big Ben, would have been quite a study for a painter. It
transcended all description, and for the moment seemed as if he
were bidding farewell to this world and to all his iniquities in it,
without the intervention of the law. But in a few moments he
recovered from this condition, and sliding on to his knees, and in a
whining tone, he cried—
"Mercy, Mercy! Oh, let me go!"
"At the end of a rope," said Big Ben. "Easy does it. What has he
been and done, Mrs. O.?"
"Murder, murder!"
A crowd of people soon began to collect around them, and then
Lupin made an effort to thrust himself out of the grasp of Big Ben,
but the only result of the effort was very nearly to strangle himself.
"You are killing the man, you great brute!" cried a woman. "You are
throttling the poor man."
"He will be murdered," shouted another female. "Oh, you great
wretch, do you want to take his life?"
"Listen to me," said Mrs. Oakley. "He has murdered his poor wife,
and that is the reason I have asked that he should be held tight."
"Murdered his wife!" exclaimed about twelve females in chorus.
"Murdered his wife? Then hanging is a great deal too good for him.
Hold him tight, sir, do. Oh, the wretch!"
The tide of popular feeling fairly turned against Mr. Lupin, and Big
Ben had as much difficulty now in preserving the half dead wretch
from popular fury as if he had been accused of any other crime, he
might have had to prevent popular sympathy from aiding his escape.
"Oh!" cried one lady, of rather extensive proportions, who was the
wife of a baker, "I should like to have him in a brisk oven for an hour
and a half."
"And I," said the lady of a butcher, "would see him slaughtered
without so much as winking at him."
"And serve him right, the wagabone!" cried Big Ben. "Come along,
will you, you ill-looking scarecrow! Easy does it. Will you walk? Oh,
very well, don't. Who are you?"
A little man with a constable's staff in his hand, rushed before Ben,
crying out—
"What is it? what is it? I'm a constable. What is it?"
"Murder!" said Mrs. Oakley. "I give that man in charge for murdering
his wife. I saw him do it."
"That will do," said the constable. "Give him to me. I'll take him. He
dare not resist me. I'll have him."
Big Ben looked at the constable and then he shook his head, as he
said very gravely—
"I tell you what it is, my little man, you ain't fit to tussle with such a
fellow as this—I'll take him along for you. Where is he to go?"
"To the round-house, in course; but I'm a constable. I must take him
—I will take him! Give him to me, sir, directly—I will have him—I
must go with him!"
"Wait a minute," said Ben. "Easy does it! You must go with him, you
say? Very good—easy does everything!"
With this, Ben grasped Mr. Lupin round the middle, and placed him
under his left arm, and suddenly pouncing, then, upon the
constable, he caught him up and placed him under the right arm;
and then away he walked, to the admiration of the populace, and
paying about as much attention to the kicking of the constable and
the kicking of Mr. Lupin, as though they were two dogs that he was
carrying home.
And so the murderer was taken to the round-house, where Mrs.
Oakley duly preferred the charge against him, and promised to
substantiate it before a magistrate when called upon so to do.
 CHAPTER CXVII.
SHOWS HOW MRS. OAKLEY RECONCILED HERSELF
TO EVERYBODY AT HOME.

When Ben and Mrs. Oakley had thus disposed of Mr. Lupin, and left
him to his solitary and not very pleasant reflections in a cell of the
round-house, they found themselves together in the open street,
and Ben, as he cast a woeful glance at her, said—
"Well, how does yer feel now? Easy does it! Oh, you aint a-been and
behaved yourself properly lately—you is like the old bear as we calls
Nosey. He's always a-doing what he shouldn't, and always a-never
doing what he should."
"Ben?"
"Well, blaze away. What is yer going to say now?"
"I feel, Ben, that I am a very different woman from what I was—
very different."
"Then you must have gained by the exchange, for you was, I will say
it, anything but a pleasant bit o' goods. There's poor old Oakley a-
making of spectacles all days, and a-wearing of his old eyes out—
and there's Miss Johanna, bless her heart! as wise a little bit o'
human nature as you'd wish to see, whether she's in petticoats or
the other things; and yet you neglects 'em both, all for to run arter a
canting snivelling wagabone like this Lupin, that we wouldn't have
among the beasteses at the Tower, if so be he'd come and offer
himself."
"I know it, Ben—I know it."
"You know it! Why didn't you know it before?"
"I don't know, Ben; but my eyes are open now. I have had a lesson
that to my dying day I shall never forget. I have found that piety
may only be a cloak with which to cover up the most monstrous
iniquity."
"Oh, you have made that discovery, have you?"
"I have, indeed, Ben."
"Well, I knowed as much as that when I was a small baby. It only
shows how back'ard some folks is in coming for'ard with their
edication."
"Yes, Ben."
"Well, and what is you going to be arter now?"
"I wish to go home, and I want you to come with me, and to say a
kind word for me; I want you to tell them how I now see the error of
my ways, and how I am an altered woman, and mean to be a very—
very different person than I was."
Here Mrs. Oakley's genuine feelings got the better of her, and she
began to weep bitterly; and Ben, after looking at her for a few
moments, cried out—
"Why, it's real, and not like our hyena that only does it to gammon
us! Come, mother Oakley, just pop your front paw under my arm,
and I'll go home with you; and if you don't get a welcome there, I'm
not a beef-eater. Why, the old man will fly right bang out of his wits
for joy. You should only see what a house is when the mother and
the wife don't do as she ought. Mother O., you should see what a bit
of fire there is in the grate, and what a hearth."
"I know it—I ought to know it."
"You ought to know it!" added Ben, putting himself into an oratorial
attitude. "You should only see the old man when dinner time comes
round. He goes into the parlour and he finds no fire; then he says
—'Dear me!'"
"Yes—yes."
"Then he gives a boy a ha'penny to go and get him something that
don't do him no sort of good from the cook's shop, and sometimes
the boy nabs the ha'penny and the shilling both, and ain't never
heard of again by any means no more."
"No doubt, Ben."
"Then, when tea comes round, it don't come round at all, and the
old man has none; but he takes in a ha'porth of milk in a jug without
a spout, and he drinks that up, cold and miserable, with a penny-
loaf, you see."
"Yes—yes."
"And then at night, when there ought to be a little sort of comfort
round the fireside, there ain't none."
"But Johanna, Ben—there is Johanna?"
"Johanna?"
"Yes. Is she not there to see to some of her father's comforts? She
loves him—I know she does, Ben!"
Ben placed his finger by the side of his nose, and in an aside to
himself, he said—
"Now I'll touch her up a bit—now I'll punish her for all she has done,
and it will serve her right." Then, elevating his voice, he added—"Did
you mention Johanna?"
"Yes, Ben, I did."
"Then I'm sorry you did. Perhaps you think she's been seeing to the
old man's comforts a little—airing his night-cap, and so on—Eh? Is
that the idea?"
"Yes, I know that she would do anything gladly for her father. She
was always most tenderly attached to him."
"Humph!"
"Why do you say, Humph, Ben?"
"Just answer me one question, Mrs. O. Did you ever hear of a young
girl as was neglected by her mother—her mother who of all ought to
be the person to attend to her—turning out well?"
"Do not terrify me, Ben."
"Well, all I have got to say is, that Johanna can't be in two places at
once, and as she isn't at home, how, I would ask any reasonable
Christian, can she attend to the old man?"
"Not at home, Ben?"
"Not—at—home!"
"Oh, Heaven! why did I not stay in that dreadful man's house, and
let him murder me! Why did I not tell him at once that I knew of his
crime, and implore him to make me his next victim! Oh, Ben, if you
have any compassion in your disposition you will tell me all, and
then I shall know what to hope, and what to dread."
"Well," said Ben, "here goes then."
"What goes?"
"I mean I'm a-going to tell you all, as you seem as if you'd like to
know it."
"Do! Oh, do!"
"Then of course Johanna being but a very young piece of goods, and
not knowing much o' the ways o' this here world, and the habits and
manners o' the wild beasteses as is in it, when she found as the old
house wasn't good enough for her mother, she naturally enough
thought it wasn't good enough for her, you know."
"Oh, this is the most dreadful stroke of all!"
"I should say it were," said Ben, quite solemnly. "Take it easy
though, and you'll get through it in the course of time. Well then,
when Johanna found as everything at home was sixes and sevens,
she borrowed a pair of what do call 'ems of some boy, and a jacket,
and off she went."
"She what?"
"She put on a pair of thingumys—well, breeches then, if you must
have it—and away she went, and the last I saw of her was in Fleet
Street with 'em on."
"Gracious Heaven!"
"Very likely, but that don't alter the facts of the case, you know, Mrs.
O. On she had 'em, and all I can say is that you might have knocked
me down flat to see her, that you might. I didn't think I should ever
have got home to the beasteses in the Tower again, it gave me such
a turn."
"Lost! Lost!"
"Eh? What do you say? What have you lost now?"
"My child! My Johanna!"
"Oh! Ah, to be sure. But then you know, Mrs. O, you ought to have
staid at home, and gived her ever so much good advice, you know;
and when you saw she was bent upon putting on the boy's things,
you as a mother ought to have said, 'My dear, take your legs out of
that if yer pleases, and if yer don't, I'll pretty soon make you,' and
then staid and gived the affair up as a bad job that wouldn't pay,
and took to morals."
"Yes—yes. 'Tis I, and I only, who am to blame. I have been the
destruction of my child. Farewell, Ben. You will perhaps in the course
of time not think quite so badly of me as you now do. Farewell!"
"Hold!" cried Ben as he clutched the arm of Mrs. Oakley only the
more tightly in his own: "What are you at now?"
"Death is now my only resource. My child is lost to me, and I have
driven her by my neglect to such a dreadful course. I cannot live
now. Let me go, Ben. You will never hear of me again."
"If I let you go may I be—Well, no matter—no matter. Come on. It's
all one, you know, a hundred years hence."
"But at present it is madness and despair. Let me go, I say. The river
is not far off, and beneath its waters I shall at least find peace for
my breaking heart. Let my death be considered as some sort of
expiation of my sins."
"Stop a bit."
"No—no—no."
"But I say, yes. Things ain't quite so bad as you think 'em, only it
was right o' me, you know, just to let you know what they might
have been."
"What do you tell me?"
"Why that there ain't a better girl than Johanna in all the world, and
that if all the mothers that ever was or ever will be, had neglected
her and set her all their bad examples in the universal world, she
would still be the little angel that she is now, and no mistake."
"Then she is not from home? It is all a fable?"
"Not quite, Mrs. O. just you trot on now comfortably by the side of
me, and I will tell you the whole particulars, and then you will find
that there ain't no occasion to go plumping into the river on
Johanna's account."
Poor Mrs. Oakley, with delight beaming upon every feature of her
face, now listened to Ben while he explained the whole matter to
her, as far as he himself was cognisant of it; and if he did not offer
to be very explicit in minor details, she at all events heard from him
quite enough to convince her that Johanna was all that the tenderest
mother could wish.
"Oh, Ben," she said, as the tears coursed each other down her
cheeks, "how could you torture me as you have done?"
"All for your own good," said Ben. "It only lets you see what might
have happened if Johanna had not been the good little thing that
she is, that's all."
"Well, perhaps it is for the best that I should have suffered such a
pang, and I only hope that Heaven will accept of it as some sort of
expiation of my wickedness. If you had not held me, Ben, I should
certainly have taken my life."
"Not a doubt about it," said Ben; "and a pretty kittle of fish you
would then have made of the whole affair. However, that's all right
enough now, and as for old Oakley, all you have got to do is to go
into the shop and say to him. 'Here I am, and I am sorry for the
past, which I hope you will forgive, and for the future I will strive to
be a good wife.'"
"Must I say that, Ben?"
"Yes, to be sure. If you are ashamed to say what's right, you may
depend upon it you haven't much inclination to do it."
"You have convinced me, Ben. I will humble myself. It is fit and
proper that I should. So I will say as nearly as I can recollect just
what you have told me to say."
"You can't do better; and here we are at the corner of the street.
Now if you would rather go in by yourself without me, only say the
word, and I'm off."
Mrs. Oakley hesitated for a moment and then she said—
"Yes, Ben, I would rather go alone."
"Very good. I think it's better too, so good-by; and I'll call to-morrow
and see how you are all getting on."
"Do so, Ben. No one can possibly be more welcome than you will be.
You will be sure to come to-morrow?"
"Rather."
With this Ben walked away, and Mrs. Oakley entered the house.
What then passed we do not feel that we ought to relate. The
humiliations of human nature, although for the best of purposes,
and for the ultimate happiness of the parties themselves, are not
subjects for the pen of the chronicler. Suffice it, that Mr. and Mrs.
Oakley were perfectly reconciled, and were happy upon that day.
 CHAPTER CXVIII.
TAKES A PEEP AT TOBIAS AT THE COLONEL'S
HOUSE.

The more stirring events of our story, have compelled us in some

measure to neglect poor Tobias. He had suffered very much from
that visit of Todd's to the colonel's house, and it had a very
prejudicial effect upon his mind too, inasmuch as it deprived him of
that feeling of security, which had before possessed him beneath
that roof.
The colonel felt this very acutely, and he could not help perceiving
by Tobias's manner, that the faith he put in his assurance that Todd
could not possibly again come near him, was not full and complete.
Under these circumstances, then, it was a very great satisfaction to
the colonel to be able to make the gratifying communication he had
it in his power to make to Tobias, on the morning following the
arrest of Todd and Mrs. Lovett.
The illness contingent upon the fright that Todd had given the poor
boy, or the relapse as we might call it, had in a great measure worn
off, and if Tobias's mind could have been quite at ease, his recovery
would have been as rapid as any one could possibly have wished or
expected.
As soon as he was up and about upon the following morning, then,
after the arrests, the colonel sought Tobias's room, and with a
cheerful smile upon his face he said—
"Well, Tobias, I come to bring you good news."
"Indeed, sir?" said Tobias his colour coming and going in flushes. "I
am very weak, and—and if—"
"Come, come, Tobias. What I am going to tell you will strengthen
you, I know. Todd is in Newgate!"
Tobias drew a long breath.
"Todd is in Newgate?" he replied. "Todd is in Newgate? The walls are
very thick. I am safe now."
"Yes, you are, indeed, Tobias. The walls of Newgate are thick, and
the doors are massive and well-guarded. Be assured that Todd will
never issue out at them but to his execution. Your old cunning
enemy is at length more powerless by a great deal than you are,
and from this moment you may completely banish all fear from your
mind upon his account."
"And the woman, sir, Mrs. Lovett?"
"She is in Newgate likewise."
"Both, both, and their crimes then are all known at last, and there
will be no more murders, and no more poor boys driven mad as I
was! Oh, God be thanked, it is indeed all over now, all over."
With this Tobias burst into tears, and relieved his surcharged heart
of a load of misery. In the course of about five minutes he looked up
with such a great smile of happiness upon his face, that it was quite
a joy to see it.
"And you, sir, you," he said, "my dear friend have done all this!"
"Not all, Tobias. I have helped in every way that lay in my power to
bring the affair about, but it is Sir Richard Blunt the magistrate, who
has toiled day and night almost in the matter, and who has at last
brought it to so successful an issue, that the guilt of both Todd and
Mrs. Lovett can be distinctly and clearly proved, without the shadow
of a doubt."
"Unhappy wretches!"
"They are, indeed, Tobias, unhappy wretches, and may Heaven have
mercy upon them. Some other old friends of yours, too, will, before
nightfall I think, find a home in Newgate."
"Indeed, sir, whom mean you?"
"The folks at the madhouse at Peckham. Sir Richard would have had
them apprehended some time ago, but he was afraid that it might
give the alarm to Todd, before the affair was ripe enough to enable
him to be arrested, with a certainty of his crimes being clearly
understood and brought home to him. Now, however, that is all over,
and they will be punished."
"They are very, very wicked. I think, sir, they are almost worse than
Sweeney Todd."
"They are, if anything; but they will meet with their deserts, never
fear; and as Minna Gray is expected every moment, so your mother
tells me, I will not deprive you of the gratification of giving her the
piece of news yourself. Of course, all the town will know it soon
through the medium of the press; and Sir Richard Blunt, too, will be
here in the course of the morning, to arrange with you concerning
your evidence."
"My evidence? Shall I be wanted?"
"Yes, Tobias. Surely you would not like so notorious a criminal to find
a loop-hole of escape, from the want of your evidence?"
"Oh, no, no—I will go. I have only to tell the truth, and that should
never be denied for or against. I will go, sir."
"You are right, Tobias. It is a duty you owe to society. If some one
long ago, and before you even had the evil fortune to go into his
shop, had found out and exposed the iniquities of Sweeney Todd,
how much misery would have been spared in this world both to you
and to others!"
"Ah, yes, sir; and yet—"
"Yet what, Tobias?"
"I was only thinking, sir, that what at times seems like our very
worst misfortunes, at times turn out to be the very things that are
the making of us."
"Indeed, Tobias?"
"Yes, sir. If I had not been Sweeney Todd's boy, and if he had not
persecuted me in the way he did, I should never have known what it
was to have the friend I now have in you, sir; and perhaps she
whom I love so dearly, would not have thought so much of me, if
she had not deeply pitied me for all that I suffered."
"There is profound philosophy in what you say, my poor boy," replied
the colonel; "and if we could only bring ourselves to think, when
things apparently go wrong with us, that after all it is for the best,
we should be much happier than we are now; but with our short-
sighted wisdom, we hastily take upon ourselves to decide upon
matters concerning the issues of which we know nothing, and so by
anticipation we make ourselves pleased or sorrowful, when the
precise contrary may be the real result."
"Yes, sir," said Tobias, "I have had time to think of that, and of many
other strange things, as I lay here."
"Then you have done yourself some good, Tobias. But I hear a light
footstep upon the stairs, and I will now leave you, for I can guess by
that heightened colour that you hear it likewise, and I know that two
may be good company but three none."
Tobias would have said something deprecatory of the colonel leaving
him, and he did begin, but with a smile his kind and hospitable
friend took his leave, and Tobias soon had the satisfaction of relating
to the young girl, whom he was so tenderly attached to, that nothing
further was now to be feared from Sweeney Todd or from Mrs.
Lovett.
We may now leave Tobias in good company; and it was really
surprising to those who have not made a habit of noting the intimate
connection there is between the mind and the body, to see how
from the very moment that he felt assured there was nothing further
to apprehend from Sweeney Todd, Tobias's health picked up and
improved. The absolute dread with which that bold impious bad man
had inspired the boy, had been the sole cause of keeping him in so
delicate a state. His dreams had been all of Todd; but now that word
Newgate, in conjunction with Todd's name, was a spell that brought
with it peace and security.
Tobias, as he sat with the hand of the young and fair girl who had
pleased his boyish fancy in his own, was now truly happy.
When Johanna got home, after being escorted from Sir Richard
Blunt's house in Craven Street by Colonel Jeffery, she found her
mother at home, and not a little surprised was she to find herself
suddenly clasped in that mother's arms, a most unwonted process
for Mrs. Oakley to go through.
"Oh, my child, my dear child!" sobbed the now repentant woman.
"Can you forgive me as your father has done?"
"Forgive you, mother? Oh, do not speak to me in such a way as that.
It is quite a joy to find you—you are really my mother?"
"You might well doubt it, my dear child; but the future is before us
all, and then you will find that it was only when I could not have
been in my right mind, that I preferred any place to my own home."
Old Oakley wiped his eyes as he said to Johanna—
"Yes, my darling, your mother has come back to us now in every
sense of the word, and all the past is to be forgotten, except such of
it as will be pleasant to remember. Your good friend, and I may say
the good friend of us all, Sir Richard Blunt, sent us a letter to say
that you would be here to-night, and God bless him my child, for
watching over you as he did."
"Oh, how perilous an enterprise you went upon, my darling," said
Mrs. Oakley.
The door of the adjoining room was partially open, and from it now
stepped forward Arabella, saying—
"It is I who ought to ask pardon of you all for advising that step; and
you will grant me that pardon I am sure, if upon no other ground,
upon that that I have suffered greatly for my folly and precipitation."
"My dear Arabella," said Johanna, "you must not blame yourself in
such a way. How pleased I am to find you here, my dear friend. Ah!
at one time how little did we ever expect to meet all thus, in this
little room!"
Johanna and Arabella embraced each other, and while they were so
occupied, big Ben came out of the room from whence Arabella had
proceeded, and flinging his arms round them both, he made a great
roaring noise, in imitation of the largest of the bears in the Tower
collection.
At the moment, Johanna was alarmed, and could not conceive what
it was; but Arabella, who knew that Ben had been in the room,
waiting for some opportunity of coming out in a highly practical
manner, only laughed, and then Johanna knew in a moment who it
was, and she cried—
"Ben, it is you!"
"Yes, it's me," said Ben, "and I'm only astonished at you two girls
fancying I was going to be quiet, and see all that kissing and
hugging going on, and not come in for any of it. Don't kick now, for I
must kiss you both, and there's an end of it. It's no use a-kicking."
To the credit of both Arabella and Johanna we may state, that they
neither of them kicked, but very quietly let Ben kiss them both.
"Well," said Ben as he plumped himself down upon a chair after the
salute. "Well!—Murder! Where am I going to now?"
"Dear me," said Mrs. Oakley. "All four legs of the chair are broken
off, and Ben is on the floor."
"Really, Ben," said Mr. Oakley, "you ought to be perfectly careful
when you sit down."
"Easy does it," said Ben. "I really thought I was going to kingdom
come. Pull me, Johanna, my dear. Pull me up."
Johanna shook her head, and declined the Herculean attempt, so
that Ben had to scramble to his feet the best way he could, and then
as he sat down upon the sofa which was sufficiently strong to
withstand any shocks, Mrs. Oakley asked him what it was he had
been upon the point of saying, when the chair had so very
unceremoniously given way with him; but Ben had quite forgotten it,
only he said he recollected something else that was quite as good,
and that was that he ordered to come about that hour a foaming
tankard of mulled wine, and then he winked at Mrs. Oakley and
hoped she had no medicine in the house to put in it.
"Oh, no, Ben," she said, "and if there isn't a knock at the door; and
if you ordered it at the Unicorn's Tail, you may depend that's it."
"Very good," said Ben, and then he proceeded to the door and found
that it was the boy from the Unicorn's dorsal appendage with the
spiced wine; and after whispering to bring a similar quantity in half
an hour, and to keep on at it every half hour until further orders, Ben
took it into the parlour, and a happier party than was there could not
have been found in all London.
 CHAPTER CXIX.
THE CRIMINALS IN NEWGATE.—TODD'S ATTEMPT
AT SUICIDE.

It is grievous to turn from the contemplation of so pleasant and

grateful a scene as that that was taking place at the old spectacle-
maker's house, to dive into the interior of Newgate. But thither it is
that now we would conduct the reader.
The state of mind that Todd was in after his arrest, was one that
such a man with such strong passions as he had was exceedingly
unlikely to come to. It is difficult to describe it, but if we say that he
was mentally stunned, we shall be as near the mark as language will
permit us to be.
He walked, and looked, and spoke very much like a man in a dream;
and it is really doubtful whether, for some hours, he comprehended
the full measure of the calamity that had befallen him on his
apprehension.
At Newgate they are quite accustomed to find this unnatural
calmness in great criminals immediately after their arrest, so they
take their measures accordingly.
Sir Richard Blunt had given some very special instructions to the
Governor of Newgate concerning his prisoner, when he should arrive
and be placed in his custody, so everything was ready for Todd. How
little he suspected that for two days and two nights the very cell he
was to occupy in Newgate had been actually pointed out, and that
the irons in which his limbs were to be encompassed were waiting
for him in the lobby!
He was placed in a small stone room that had no light but what
came from a little orifice in the roof, and that was only a borrowed
light after all, so that the cell was in a state of semi-darkness
always.
Into this place he was hurried, and the blacksmith who was in the
habit of officiating upon such occasions, riveted upon him, as was
then the custom, a complete set of irons.
All this Todd looked at with seeming indifference. His face had upon
it an unnatural flush, and probably Todd had never looked so
strangely well in health as upon the occasion of the first few hours
he spent in Newgate.
"Now, old fellow," said one of the turnkeys, "I'm not to be very far
off, in case you should happen to want to say anything; and if you
give a rap at the door, I'll come to you."
"In case I want to say anything?" said Todd.
"Yes, to be sure. What, are you asleep?"
"Am I asleep?"
"Why, he's gone a little bit out of his mind," said the blacksmith, as
he gathered up his tools to be gone.
The turnkey shook his head.
"Are you quite sure you have made a tight job of that?"
"Sure? Ay, that I am. If he gets out of them, put me in 'em, that's
all. Oh, no! It would take—let me see—it would take about half a
dozen of him to twist out o' that suit of armour. They are just about
the best we have in the old stone jug."
"Good."
"Yes, they are good."
"I mean very well. And now Mr. Sweeney Todd, we will leave you to
your own reflections, old boy, and much good may they do you.
Good-night, old fellow. I always says good-night to the prisoners,
cos it has a tender sort o' sound, and disposes of 'em to sleep. It's
kind o' me, but I always was tender-hearted, as any little chick, I
was."
Bang went the cell door, and its triple locks were shot into their
hoops. Todd was alone.
He had sat down upon a stool that was in the cell; and that stool,
with a sort of bench fastened to the wall, was the only furniture it
contained; and there he sat for about half an hour, during which
time one of the most extraordinary changes that ever took place in
the face of any human being, took place in his.
It seemed as if the wear and tear of years had been concentrated
into minutes; and in that short space of time he passed from a
middle aged, to be an old man.
Then reflection came!
"Newgate!" he cried as he sprang to his feet.
The chains rattled and clanked together.
"Chains—Newgate—a cell—death! Found out at last! At the moment
of my triumph—defeated—detected! Newgate—chains—death!"
He fell back upon the stool again, and sat for the space of about two
minutes in perfect silence. Then he sprang up again with such a wild
yell of rage and mental agony, that not only the cell, but the whole
of that portion of the prison, echoed again with it.
The turnkey opened a small wicket in the door, which when it was
opened from without, still was defended by iron bars across it, and
peering into the cell, he said—
"Hilloa! What now?"
"Hilloa!" shouted Todd. "Air—air!"
"Air? Why what do you mean by gammoning a fellow in that sort o'
way for, eh? Haven't you got lots o' air? Well, of all the unreasonable
coves as ever I comed across, you is the worstest. Be quiet, will
you?"
"No—no! Death—death! Give me the means of instant death. I am
going mad—mad—mad!"
"Oh, no yer ain't. It's only yer first few hours in the stone-jug that
has comed over you a little, that's all, old fellow. You'll soon pick up,
and behave yourself like any other christian. All you have got to do is
never to mind, and then it's nothink at all, old chap."
Clap went shut the little wicket door again.
"Help! Help!" shouted Todd. "Take these irons off me. It is only a
dream after all. Back, back you grinning fiends—why do you look at
me when you know that it is not real? No—no, it cannot be, you
know that it cannot be real."
"Be quiet will you?" shouted the turnkey.
"Keep off, I say. All is well. Mrs. Lovett dead—quite dead. The boy to
die too. The house in a blaze—all is well arranged. Why do you mock
and joke at me?"
"Well, I never!" said the turnkey. "I do begin to think now that he's
getting queer in the upper story. I have heard of its driving some of
'em mad to be bowled out when they didn't expect it, more 'special
when it's a hanging affair. I wonder what he will say next? He's a
regular rum un, he is."
"What have I done?" shouted Todd. "What have I done? Nothing—
nothing. The dead tell no tales. All is safe—quite safe. The grave is a
good secret keeper. I think Tobias is dead too—why not? Mrs. Lovett
is dead. This is not Newgate. These are not chains. It is only the
nightmare. Ha! ha! ha! It is only the nightmare—I can laugh now!"
"Oh, can you?" said the turnkey. "It's rather an odd sort o' laugh
though, to my thinking. Howsomdever, there's no rule agin grinning,
so you can go on at it as long as you like."
"Mercy!" suddenly shrieked Todd, and then down he fell upon the
floor of the cell, and lay quite still. The turnkey looked curiously in at
him, through the little grating.
"Humph!" he said, "I must go and report him to the Governor, and
he will do whatsomdever he likes about him; but I suppose as they
will send the doctor to him, and all that ere sort o' thing, for it won't
do to let him slip out o' the world and quite cheat the gallows; oh
dear no."
Muttering these and similar remarks to himself, the turnkey went, as
he was bound in duty to do upon any very extraordinary conduct
upon the part of any prisoner in his department, to report what Todd
was about to the Governor.
"Ah!" said that functionary, the surgeon, "and I will soon come to
him. I fully expected we should have some trouble with that man. It
really is too bad, that when people come into the prison, they will
not be quiet. It would be just as well for them, and much more
comfortable for me."
"Werry much, sir," said the turnkey.
"Well—well, he shall be attended to."
"Werry good, sir."
The turnkey went back and took up his post again outside Todd's
door, and in the course of ten minutes or so, without making the
least hurry of the subject, the Governor and the jail surgeon arrived
and entered the cell.
Todd was picked up, and then it was found that he had struck his
head against the stone floor, and so produced a state of insensibility,
but whether he had done it on purpose or by accident, they could
come to no opinion.
"Lay him on the bench," said the surgeon, "I can do nothing with
him. He will come to himself again in a little while, I daresay, and be
all right again in the morning."
"He seems really, indeed, to be a very troublesome man," said the
Governor to the surgeon.
"Very likely. Have you a mind for a game of cribbage to-night,
Governor? I suppose this fellow will hang?"
"Yes, I don't mind a game. Yes, they will tuck him up."
With this they left Todd's cell, and the turnkey closed the door, and
made the highly philosophical remark to himself of—
"Werry good."
Todd remained until the morning in a state of insensibility, and when
he awakened from it he was very much depressed in strength
indeed. He lay for about two hours gazing on the ceiling of his cell,
and then the door was opened, and the turnkey appeared with a
bason of milk-and-water and a lump of coarse bread.
"Breakfast!" he cried.
Todd glared at him.
"Breakfast; don't you understand that, old cock? However, it's all one
to me. There it is—take it or leave it."
Todd did not speak, and the not over luxurious meal was placed on
the table, or rather upon the end of the bench upon which he lay,
and which served the purpose of a table.
The moment Todd heard the door of the cell closed behind the
turnkey, he rose from his recumbent posture, and, although he
staggered when he got to his feet, he seized the bason, and at once,
without tasting any of its contents, broke it against the corner of the
bench to fragments.
"I shall elude them yet!" he said. "They think they have me in their
toils—but I shall elude them yet!"
He selected a long jagged piece of the broken bason, and dragging
down his cravat with one hand, he was upon the very point of
plunging it into his throat with the other, when the turnkey sprang
into the cell.
 Todd In Newgate, Tries To Commit Suicide.

"Hold a bit!" he cried. "We don't allow that sort of thing here with
any of our customers. You should have thought of those games
before you got into the stone jug!"
With one powerful blow, the turnkey struck the piece of the broken
bason from the hand of Todd, and with another he felled him to the
floor.
"None o' your nonsense," he said; and then he carefully collected
the pieces of the broken bason.
"Why should you grudge me the means of death," said Todd, "when
you know that you have brought me here among you to die?"
"Contrary to rules."
"In mercy, I ask you only to give me leave to take my own life, for I
have failed in the object of my living."
"Contrary to rules."
The turnkey left the cell, then, as coolly as if nothing had happened,
and carefully locked the door again, while he went to report the
attempted suicide of the prisoner to the proper quarter.
Foiled, then, in every way, Todd looked round the cell for some
means of ridding himself of his life and his troubles together; but he
found none. He then paced the cell to and fro like a maniac, as he
muttered to himself—
"All lost—lost—lost—all lost! Foiled, too, at the moment when I
thought myself most secure—when I had made every preparation to
leave England for ever! Oh, dolt that I was, not to have done so long
ago, when I had half—ay, when I had only a quarter of the sum that
I should this day have fled with! In my dreams I have seen myself as
I am now, and the sight has shaken me, but I never thought to be
so in reality. Is there any hope for me? What do they know?—what
can they know?"
Upon these questions, Todd paused in his uneasy walk in the cell,
and sat down upon the low stool to think. His head rested upon his
breast, and he was profoundly still.
 CHAPTER CXX.
A LUNCHEON AT SIR RICHARD BLUNT'S.—THE DOG
AND HIS OLD FRIEND.

We willingly leave Todd to his own reflections upon the disastrous

state of his affairs, while we solicit the attention of our readers to
the private house and office of Sir Richard Blunt again, in Craven
Street.
The worthy magistrate had quite a party to lunch on that day, and
he had fixed the hour as eleven when he wished to see his friends.
Those friends consisted of Johanna Oakley, Mark Ingestrie, Mr. and
Mrs. Oakley, Colonel Jeffery, Arabella Wilmot, and Big Ben, who was,
at the special request of Johanna, gladly included in the party.
A happier party than that could not very well have been found
throughout the whole length and breadth of London; and there was
but one slight shade of disquietude upon the face of Johanna, when
she at times thought that at one o'clock she would have to attend
the police-office at Bow Street to give her testimony against Todd
the murderer.
"Well," said Ben, "here we are alive—all alive, and as merry as so
many grigs; and all I can say is, my tulips, that I will show the wild
beasteses to anybody as likes to come to the Tower, free, gratis and
for nothing. Take it easy, Mr. Ingestrie, and don't be casting sheep's-
eyes at Johanna. The little love of a thing ain't at all used to it—
indeed, she ain't; and the only person as she lets love her above a
bit, and takes it easy with, is me; so don't come any nonsense."
Welcome to Our Bookstore - The Ultimate Destination for Book Lovers
Are you passionate about books and eager to explore new worlds of
knowledge? At our website, we offer a vast collection of books that
cater to every interest and age group. From classic literature to
specialized publications, self-help books, and children’s stories, we
have it all! Each book is a gateway to new adventures, helping you
expand your knowledge and nourish your soul
Experience Convenient and Enjoyable Book Shopping Our website is more
than just an online bookstore—it’s a bridge connecting readers to the
timeless values of culture and wisdom. With a sleek and user-friendly
interface and a smart search system, you can find your favorite books
quickly and easily. Enjoy special promotions, fast home delivery, and
a seamless shopping experience that saves you time and enhances your
love for reading.
Let us accompany you on the journey of exploring knowledge and
personal growth!

ebookball.com