A Hybrid Approach to Mining Conditions
Download as PPTX, PDF0 likes80 views
This document presents a hybrid approach to condition mining that uses both computational linguistics and deep learning. It proposes using a deep regressor trained on condition candidates generated from sentences to score the candidates. The main methods involve generating candidates from new sentences, scoring them with the regressor, and selecting the best candidates. Experimental results on English and Spanish reviews show the approach outperforms baselines, with variants using CNNs and CNN-BiGRUs performing best on average. Analysis of the results finds these variants statistically significantly outperform other methods. The approach overcomes issues in prior work and achieves good accuracy.
![Inputs
Sentence Conditions
I think that the lens is beyond excellent for
amateurs.
[“for amateurs”]
The resolution of this camera is 13Mp. []
Flash is tacky when using outdoors. [“when using outdoors”]
… …
16](https://image.slidesharecdn.com/fogallego-hais18-180705131304/85/A-Hybrid-Approach-to-Mining-Conditions-16-320.jpg)
A Hybrid Approach to Mining Conditions
- 1. A Hybrid Approach to Mining Conditions Fernando O. Gallego, and Rafael Corchuelo
- 2. Opinion mining Attribute Polarity “lens” Positive 2 Attribute Polarity “resolution” Neutral “Flash” Negative I think that the lens is beyond excellent for amateurs. The resolution of this camera is 13Mp. Flash is tacky when using outdoors.
- 3. But wait! 3 The opinion is only true in a certain situation
- 4. Opinion mining (with conditions) Attribute Polarity “lens” Positive (for amateurs) 4 Attribute Polarity “resolution” Neutral “Flash” Negative (when using outdoors) I think that the lens is beyond excellent for amateurs. The resolution of this camera is 13Mp. Flash is tacky when using outdoors.
- 7. Condition mining 7 I think that the lens is beyond excellent for amateurs. The resolution of this camera is 13Mp. Flash is tacky when using outdoors. for amateurs when using outdoors
- 9. Handcrafted patterns • Mausam et al. (2012) – OpenIE extraction – Dependency tree – Adverbial clauses • Chikersal et al. (2015) – Opinion mining – Basic connectives – “then”/comma 9
- 10. And the problem is… 10
- 11. Variability of conditions 11 0/1st/2nd/3rd conditionals If you do sth Even if sby fell down If sth had passed Should you help me When sth happens May it be accepted For sby To sby During my event While doing sth After/before sth If it occurs
- 12. Machine learning • Nakayama et al (2015): – SVM/CRF Model – 3k Japanese sentences – Several lexicons used 12
- 13. And the problem is… 13
- 15. Our solution Computational linguistics + Deep learning
- 16. Inputs Sentence Conditions I think that the lens is beyond excellent for amateurs. [“for amateurs”] The resolution of this camera is 13Mp. [] Flash is tacky when using outdoors. [“when using outdoors”] … … 16
- 17. Main methods 17
- 18. Main methods 18
- 19. Train (1/4) • Create a subset of training examples for each sentence 19 S1 S1 S1 S1 ts ts ts ts
- 20. Train (2/4) • Generate condition candidates for a given sentence 20 outdoors tacky Flash is usingcop advcl nsubj advmod when c1: c2: advmod Flash is tacky when using outdoors when using outdoors
- 21. Train (3/4) • Score each candidate 21 c1: c2: Flash is tacky when using outdoors when using outdoors 0.8560 1.0000
- 22. Train (4/4) • Train a deep regressor from training set 22
- 24. Main methods 24
- 25. Apply (1/5) • Generate condition candidates 25 outdoors tacky Flash is usingcop advcl nsubj advmod when c1: c2: advmod Flash is tacky when using outdoors when using outdoors
- 26. Apply (2/5) • For each condition candidate it checks whether it must be considered or not 26
- 27. Apply (3/5) • The regressor scores the candidate 27 c1: c2: Flash is tacky when using outdoors when using outdoors 0.8560 1.0000
- 28. Apply (4/5) • If score is equal to or greater than a given threshold, it is considered 28
- 29. Apply (5/5) • It keeps the best non- overlapped candidates 29
- 31. Hardware & software configuration • Intel Xeon E5-2690 • 4 threads at 2.60 Ghz • 2 GiB of RAM • Nvidia Tesla K10 GPU • CentOS Linux 7.3 • Snowball 1.2.1 • Stanford Core NLP 3.8.0 • Python 3.5.4 • Gensim 2.3.0 • Keras 2.0.8 & Theano 1.0 31
- 34. Results 34
- 36. Well done! • It overcomes the problems found in the literature • Comprehensive experimental analysis • It achieves good results 36
- 38. Condition mining’s main applications 38 Inf. Extraction Opinion mining Recommenders
- 39. Detailed example (1/3) 39 who cake if you be lik- ’s try nsubj dobj someone then john advmod xcompadvcl mark acl:relcl copnsubj case If you are someone who likes cakes then try John’s
- 40. Detailed example (2/3) 40 ’s case john xcomp try - then advmod cake dobj lik- acl:relcl who nsubj someone advcl be cop you nsubj if mark c1 : c2: c3: c4: cake dobj lik- acl:relcl who nsubj someone advcl be cop you nsubj if mark cake dobj lik- acl:relcl who nsubj ’s case john xcomp who cake if you be lik- ’s try nsubj dobj someone then john advmod xcompadvcl mark acl:relcl copnsubj case
- 42. Our Neural Networks CNN: Convolution Input ld Output .9l1.2d Activation relu Kernel 3 Drop-out 0.2000 Convolution Input .9l1.2d Output .6l.3d Activation relu Kernel 17 Drop-out 0.2000 Pooling Input .6l.3d Output .6l1 Functor max Pool global Dense Input .6l1 Output .3l1 Activation linear Drop-out 0.2000 Dense Input .3l1 Output 11 Activation tanh Drop-out 0.0000 MLP: GRU: Dense Input ld Output l.5d Activation tanh Drop-out 0.2000 Dense Input l.5d Output 11 Activation linear Drop-out 0.0000 GRU Input ld Output l1 Activation tanh Drop-out 0.1500 Dense Input l1 Output .3l1 Activation linear Drop-out 0.2000 Dense Input .3l1 Output 11 Activation tanh Drop-out 0.0000 BiGRU: BiGRU Input ld Output 2l1 Activation tanh Drop-out 0.1500 Dense Input 2l1 Output .3l1 Activation linear Drop-out 0.2000 CNN-BiGRU: Convolution Input ld Output .9l.3d Activation relu Kernel 3 Drop-out 0.0000 Pooling Input .9l.3d Output .9l.3d Functor max Pool 2 BiGRU Input .9l.3d Output 4l1 Activation tanh Drop-out 0.1500 Dense Input .4l1 Output .3l1 Activation linear Drop-out 0.2000 Dense Input 3l1 Output 11 Activation tanh Drop-out 0.0000 Dense Input .3l1 Output 11 Activation tanh Drop-out 0.0000
- 43. Detailed results Lang Proposal q = 0.2500 q = 0.5000 q = 0.7500 P R F1 P R F1 P R F1 en MB 0.6270 0.6144 0.6206 0.6270 0.6144 0.6206 0.6270 0.6144 0.6206 CB 0.7979 0.4642 0.5870 0.7979 0.4642 0.5870 0.7979 0.4642 0.5870 Averages 0.7125 0.5393 0.6038 0.7125 0.5393 0.6038 0.7125 0.5393 0.6038 MLP 0.4741 0.7799 0.5897 0.5612 0.5271 0.5436 0.5739 0.4582 0.5096 GRU 0.9999 0.4421 0.6131 0.9999 0.4421 0.6131 0.9999 0.4421 0.6131 BiGRU 0.5448 0.5262 0.5353 0.8999 0.4421 0.5929 0.9999 0.4421 0.6131 CNN 0.5908 0.7546 0.6628 0.6211 0.6278 0.6244 0.6571 0.5432 0.5948 CNN-BiGRU 0.5586 0.8052 0.6596 0.6318 0.6529 0.6422 0.7327 0.4914 0.5883 Averages 0.6336 0.6616 0.6121 0.7428 0.5384 0.6033 0.7927 0.4754 0.5838 es MB 0.6699 0.5285 0.5909 0.6699 0.5285 0.5909 0.6699 0.5285 0.5909 CB 0.7953 0.4399 0.5665 0.7953 0.4399 0.5665 0.7953 0.4399 0.5665 Averages 0.7326 0.4842 0.5787 0.7326 0.4842 0.5787 0.7326 0.4842 0.5787 MLP 0.4232 0.8295 0.5604 0.5382 0.5678 0.5526 0.5771 0.4465 0.5034 GRU 0.5246 0.7483 0.6168 0.7089 0.4304 0.5356 0.9999 0.4153 0.5869 BiGRU 0.5321 0.7451 0.6209 0.6335 0.4692 0.5391 0.9999 0.4153 0.5869 CNN 0.5997 0.7519 0.6672 0.6606 0.6521 0.6563 0.7065 0.5467 0.6164 CNN-BiGRU 0.5227 0.8221 0.6390 0.6195 0.6968 0.6559 0.6843 0.5369 0.6017 Averages 0.5205 0.7794 0.6209 0.6321 0.5633 0.5879 0.7935 0.4721 0.5790 43
- 44. Statistical analysis q = 0.2500 q = 0.5000 Proposal Ranking Comparison z p-value Proposal Ranking Comparison z p-value CNN 1.0000 CNN x CNN - - CNN-BiGRU 1.4000 CNN-BiGRU x CNN-BiGRU - - CNN-BiGRU 2.0000 CNN x CNN-BiGRU 1.4142 0.1573 CNN 1.6000 CNN-BiGRU x CNN 0.2828 0.7773 BiGRU 3.5000 CNN x BiGRU 3.5355 0.0008 MLP 3.1000 CNN-BiGRU x MLP 2.4042 0.0324 MLP 4.1000 CNN x MLP 4.3841 0.0000 BiGRU 4.2000 CNN-BiGRU x BiGRU 3.9598 0.0002 GRU 4.4000 CNN x GRU 4.8083 0.0000 GRU 4.7000 CNN-BiGRU x GRU 4.6669 0.0000 (a) (b) q = 0.7500 Proposal Ranking Comparison z p-value Proposal Ranking Comparison z p-value CNN 1.3000 CNN x CNN - - CNN0.25 1.4000 CNN0.25 x CNN0.25 - - CNN-BiGRU 1.7000 CNN x CNN-BiGRU 0.5657 0.5716 CNN-BiGRU0.50 1.8000 CNN0.25 x CNN-BiGRU0.50 0.5657 0.5716 MLP 3.0000 CNN x MLP 2.4042 0.0324 MB 3.4000 CNN0.25 xMB 2.8284 0.0094 GRU 4.5000 CNN x GRU 4.5255 0.0000 CNN0.75 3.7000 CNN0.25 xCNN0.75 3.2527 0.0034 BiGRU 4.5000 CNN x BiGRU 4.5255 0.0000 CB 4.7000 CNN0.25 x CB 4.6669 0.0000 (c) (d) 44
Editor's Notes
- #2: Thanks for attending my presentation. My name is Fernando O. Gallego and I co-authored this paper with Rafael Corchuelo, both from the University of Seville. -- Copyright (C) 2018 The Distributed Group The use of these slides is hereby constrained to the conditions of the TDG Licence, a copy of which you may download from http://www.tdg-seville.info/License.html
- #3: First of all, let’s introduce an example to understand the problem. Opinion mining is a set of natural language processing tasks whose main goal is to determine whether an opinion of a document or an aspect is positive, negative, or neutral.
- #4: But wait! There is a problem that you likely didn’t notice.
- #5: There are some clauses in the sentence, which are known as conditions, that changes the sense of the opinion. For instance, the positive opinion about “lens” is only true if you consider amateur photographers. Alike, the negative opinion regarding “Flash” is only true if the user uses the camera outdoors.
- #6: This is the roadmap of my presentation: I’ll start with a broad introduction to the problem, then I’ll report on our proposal, then on some experimental results, and, finally, I’ll present some conclusions.
- #7: Let’s start with the introduction
- #8: Simply put, condition mining is a task whose goal is to identify conditions from a piece of text.
- #9: Currently, there are two approaches in the literature, namely: handcrafted patterns and machine learning.
- #10: Handcrafted or user-defined patterns clearly describe how to identify a condition in a text by means of connectives, pos-tags, dependency tags, or another clue words. There are two proposals in this way: Mausam, who studied the problem in the field of entity-relation extraction and uses adverbial clauses from the dependency tree; and Chikersal, who studied the problem in the field of opinion mining and uses basic connectives and tokens “then”/comma.
- #11: Unfortunately, the previous proposals are not appealing because of the human effort when handcrafting such patterns.
- #12: Furthermore, the results typically fall short regarding recall because of the variability of the conditions.
- #13: The only existing machine-learning proposal was introduced by Nakayama et al, who worked in the field of opinion mining in Japanese. They devised a model that is based on several features from opinion expressions, which requires to provide some specific-purpose dictionaries, taxonomies, and heuristics. They used Conditional Random Fields and Support Vector Machines to learn classifiers of syntactic units of the sentences.
- #14: Unfortunately, their proposal was only evaluated on a small dataset with 3,155 sentences regarding hotels and the best F1 score attained was 0.58. As a conclusion, this proposal is not generally applicable and its effectiveness is poor
- #15: Then, we’ll describe our proposal
- #16: Our solution is a hybrid approach that combines computational linguistics and deep learning. It does not have any of the problems found in the related work.
- #17: Our inputs are a set of sentences with its corresponding sets of labelling. Those sets identify the conditions for each sentence.
- #18: These are our proposal’s main methods.
- #19: Method “train” returns a regressor that computes a score that assesses how likely a candidate condition is an actual condition.
- #20: The procedure is repeated for every input sentence to compute a subset of training examples.
- #21: The procedure starts by generating a set of condition candidates from the sentence’s dependency tree. The heuristic used is quite simple, we consider every non-leaf node in the dependency tree and compute all of the sequences of tokens that originate from that node.
- #22: For each candidate, we computed a score that represents how likely it is a condition.
- #23: And finally, we train a deep regressor using well-known Deep learning networks.
- #24: We have experimented with a dozens neural network alternatives but the best ones are those that we present in our paper, namely: Multilayer Perceptron, Gated Recurrent Unit Network, Bidirectional Gated Recurrent Unit Network, Convolutional Neural Network, and a hybrid neural network composed of both Convolutional layers and Bidirectional Gated Recurrent Unit layers.
- #25: Method “apply” returns the conditions found in a sentence by means of that regressor.
- #26: We first need to compute the set of candidate conditions of the sentence. This method is the same as one used in main method train.
- #27: The procedure is repeated for every candidate to check whether it must be considered or not
- #28: Given a candidate, we need to score it. In this case, we use the regressor that we trained before.
- #29: If the score is equal to or greater than a given threshold, it is added to the result set.
- #30: Finally, we remove the conditions that overlap others with a higher score.
- #31: Now, let me show you our experimental results.
- #32: This is our hardware and software configuration. As you can see, it’s a pretty regular configuration with recent versions of software components.
- #33: We used a dataset with almost 4 million sentences in English and Spanish. In addiction, we just increased the amount of sentences by means of new languages like French or Italian, and we uploaded it to Kaggle platform.
- #34: We used the Handcrafted patterns proposals as baselines. But the Machine Learning proposal wasn’t considered because it is not clear if it can be customised to deal with languages other than Japanese and its best F1 was 0.58; neither could we find an implementation or the dataset.
- #35: In this slide we present our results in terms of F1 score. Our best alternatives are, namely: CNN and CNN-BiGRU, which beats the related work proposals. We performed statistical analysis to determine which alternative is the winner.
- #36: It’s time for conclusions
- #37: Our conclusions are that we’ve present a proposal that overcomes the problems found in the literature. Our experimental analysis covers a variety of alternatives and it achieves promising results.
- #38: Thanks for attending this presentation
- #44: Corregir el símbolo del threshold.
- #45: Corregir el símbolo del threshold.