A Proposal on

Job Recommendation and Resume

Screening System Using Natural Language
Processing

Submitted at:
Birat Multiple College, Biratnagar, Morang
Submitted by:
Bijesh Regmi
Nisip Sigdel
Bibek Karki
[Link] VII Semester
Birat Multiple College, Biratnagar, Morang
May, 2025
 ABSTRACT

This study proposes an intelligent system designed to optimize job search and
recruitment processes through advanced resume screening and personalized job
recommendations. Leveraging Natural Language Processing (NLP) techniques—
including BERT (Bidirectional Encoder Representations from Transformers),
lemmatization, vectorization, and Cosine Similarity—the system processes both
resumes and job descriptions to ensure precise semantic matching. For job seekers,
resumes are preprocessed through lemmatization to normalize text, enriched with
BERT-based embeddings to capture contextual meaning, and vectorization to
emphasize key terms. For recruiters, job descriptions follow the same processing
pipeline. The system then applies Cosine Similarity to compare candidate resumes
with job descriptions, generating ranked lists of job opportunities for applicants and
top candidates for recruiters. Additionally, a feedback mechanism enables recruiters to
provide evaluations, which are used to iteratively refine future resume on job seeker’s
side. By unifying contextual understanding, text normalization, and semantic
similarity into a single framework, the system enhances the accuracy, efficiency, and
fairness of the job-candidate matching process.

Keyword: Natural Language Processing (NLP), Resume Screening, BERT,

vectorization, Cosine Similarity, Candidate Evaluation

i
 Table of Content
SN

0 Abstraction i

1 Introduction 1

2 Statement of the Problem 1-2

3 Objectives 2

4 Methodology 3-11
4.1. Requirement Identification 3-6
4.2. Feasibility Study 6-8
4.3. High level Design 9-11
5 Expected Outcomes 12
6 References 12-13
1. Introduction
In today’s competitive job market, both job seekers and recruiters face numerous
challenges in finding the right match. Candidates often struggle to identify
opportunities that align with their skills and experience, while recruiters face
difficulties in efficiently screening large volumes of resumes to find qualified
applicants. This mismatch not only slows down the hiring process but also increases
the risk of overlooking potential talent due to inefficient or biased screening methods.

We believe that the integration of intelligent technologies into the recruitment process
can significantly improve the efficiency, fairness, and accuracy of job-candidate
matching. With the growing availability of job-related data and advancements in
Natural Language Processing (NLP), we see an opportunity to address the existing
gaps through automation and semantic analysis.

Our proposed system leverages NLP techniques—such as lemmatization, BERT

(Bidirectional Encoder Representations from Transformers), vectorization, and Cosine
Similarity—to create a robust framework for resume screening and personalized job
recommendations. By analyzing both resumes and job descriptions through the same
semantic pipeline, the system ensures meaningful matching based on skills,
qualifications, and experience rather than simple keyword overlaps.

Additionally, the system features a feedback mechanism that allows recruiters to

provide constructive input on candidate resumes. This feedback can help job seekers
identify and address shortcomings in their profiles, ultimately improving their chances
of securing relevant employment. In a world where the job market is rapidly evolving,
such a system can empower both job seekers and employers by making the hiring
process more data-driven and less prone to human bias.

2. Statement of the Problem

The current recruitment process is inefficient, time-consuming, and often biased due
to manual screening and basic keyword filters.

1
Job seekers struggle to find roles matching their skills, while recruiters are
overwhelmed by irrelevant applications.

Traditional systems fail to understand the context and semantics of resumes and
job descriptions, leading to poor matches, missed opportunities, and hiring
inefficiencies.

These issues are worse in regions with limited career support, such as Nepal.

There is a clear need for an intelligent, NLP-based system that can semantically
analyze resumes and job descriptions, reduce bias, and offer feedback to
job seekers.

This study explores:

i. Limitations of current resume screening and job recommendation systems.

ii. Use of NLP to semantically improve resume-job matching.

iii. Effectiveness of such systems in enhancing match accuracy and reducing bias.

3. Objectives
The primary objective of this project is to develop a job recommendation and resume
screening system utilizing Natural Language Processing to enhance the efficiency
accuracy, and fairness of the recruitment process in a single platform.

The specific objectives of the study are listed below:

i. To develop an automated, NLP-powered system for accurate and unbiased matching

between candidate resumes and job descriptions, enhancing efficiency in the
recruitment process.

ii. To provide personalized job recommendations for candidates and enable recruiters
to give feedback for iterative resume improvement.

2
[Link]

4.1. Requirement identification

a. Study of existing system

Current recruitment processes, both manual and automated, face significant

challenges. Manually, recruiters contend with an overwhelming volume of resumes,
leading to resume fatigue, inconsistent screening, and unconscious biases based on
factors like names, education institutions, or even resume formatting. This often
results in a limited perspective, overlooking candidates with hidden qualities or non-
traditional backgrounds, and relying on potentially misleading or fabricated
information

Existing Systems and basic screening tools aim to alleviate this by parsing resumes
and matching keywords. While they significantly speed up data entry and initial
filtering, they often suffer from lack of contextual understanding and nuance. These
systems primarily rely on exact keyword matches, leading to limitations such as:
 Inability to recognize synonyms, abbreviations, or variations of keywords
 Potential for inherited bias towards some group of people, gender, names etc.
 Commercial job recommendation platforms often provide basic filters
(location, salary, keywords) and can offer "job alerts," but their
recommendation engines typically lack deep semantic understanding.

b. Literature Review
• Studied Natural Language Processing (NLP) in Recruitment:
[Link]
This study highlights how NLP enhances recruitment by automating candidate
screening and improving the interpretation of job descriptions and resumes through
semantic analysis.

3
• Studied Resume Parsing Techniques:
[Link]
Resume_Parser_Analysis_Using_Machine_Learning_and_Natural_Language_Proces
sing
This paper analyzes different machine learning and NLP-based approaches to extract
structured data from unstructured resumes for efficient processing.

• Studied Job-Resume Matching Algorithms:

[Link]
The research explores advanced matching algorithms that use deep learning and
embedding techniques to improve the accuracy of resume-job fit.

• Studied Bias and Fairness in Automated Hiring:

[Link]
This paper examines ethical concerns in automated hiring systems, focusing on how
bias can be introduced and how fairness can be improved through responsible AI
design.

ii. Requirement analysis

The goal of the proposed system is to bridge the gap between job seekers and
recruiters by providing intelligent, accurate, and personalized job-candidate matching.
To achieve this, both functional and non-functional requirements must be carefully
analyzed and defined.
1. Functional Requirements
These are the core features and capabilities the system must perform:
a. Resume and Job Description Input
 The system must allow users to upload or enter resume content and job
descriptions in common formats (e.g., .txt, .pdf, or form input).
 Both candidates and recruiters should have separate portals/interfaces for
input.

4
b. Text Preprocessing:
 The system must clean and normalize the text using lemmatization to reduce
words to their base forms, removal of stop words, special characters, and
irrelevant symbols.
 Tokenization to split text into meaningful units.

c. Feature Extraction
 The system should extract meaningful information from text using
vectorization to give weight to important terms.
d. Similarity Matching
 The system must implement Cosine Similarity to compare resumes against job
descriptions (for job recommendations), job descriptions against resumes (for
candidate ranking). Results should be sorted by relevance score and displayed
in ranked order.
e. Feedback Mechanism
 Recruiters should be able to provide structured feedback on resumes.
 Feedback may include ratings, strengths, weaknesses, and suggestions for
improvement.
 The system should store this feedback and optionally use it to recommend
changes to resumes.
f. Result Presentation
 The system must display a list of suitable jobs for candidates, ranked by
relevance,
 A list of top candidates for recruiters, based on semantic match, each match
should include a similarity score and a brief explanation.
g. Account Management
 Users (both recruiters and candidates) should be able to register/login.
 Each user should have a profile storing their submitted resumes or job listings.

5
2. Non-Functional Requirements
 These ensure the system performs reliably, efficiently, and securely.
a. Usability
 The system must have a simple, intuitive UI for both job seekers and
recruiters.
 Responses and results should be easy to understand, even for non-technical
users.
b. Performance
 The system should respond within a few seconds for standard-sized
documents. It must support concurrent users without noticeable performance
degradation.
c. Scalability
 The system should be designed to handle a growing database of resumes and
job listings.
d. Accuracy
 The use of BERT embeddings and vectorization must ensure a high level of
semantic matching accuracy.
 Matching algorithms should be validated using relevant metrics
e. Security and Privacy
 Resume and job data must be securely stored and accessible only to authorized
users.
 Feedback should be confidential and used responsibly to improve the system.
f. Maintainability
 The system should be well-documented, allowing future updates or model
enhancements.

4.2. Feasibility Study

i. Technical Feasibility
The proposed system is technically feasible, given the availability and maturity of the
technologies required. The key components—Natural Language Processing (NLP),
semantic similarity analysis, and machine learning—are widely supported by modern
libraries and frameworks. Some of the technologies and tools include:
6
  BERT (Bidirectional Encoder Representations from Transformers) for
contextual word embeddings.
 Cosine Similarity for semantic comparison.
 Python as the primary programming language due to its robust support for data
science and NLP
 Node/ExpressJs for backend server and React for frontend development.
ii. Operational Feasibility
The proposed system is operationally feasible and can be easily integrated into
existing recruitment workflows. The system is designed to serve two user groups:
 Job Seekers, who upload resumes and receive personalized job
recommendations.
 Recruiters, who input job descriptions and receive ranked lists of qualified
candidates.
The interface will be user-friendly and require minimal training. Additionally, the
feedback system allows recruiters to guide applicants in improving their resumes,
making the system interactive and valuable in the long run.
iii. Economic Feasibility
From an economic perspective, the system is feasible with a relatively low cost of
development and maintenance:
 Development Tools: Most of the required tools and frameworks (e.g., Python,
Scikit-learn, Transformers library) are open-source and freely available.
 Infrastructure: Initial development can be done using free or low-cost cloud
resources. For deployment, a basic server would suffice.
 Manpower: A small team of developers with experience in NLP and web
development can build the product.


4. Schedule Feasibility
 The project has been carefully broken down into manageable phases with
clearly defined deliverables and timelines.

 Each phase targets specific milestones and have deliverables.

 Additionally, the final few weeks are reserved for preparing comprehensive
documentation, internal reviews, final presentation, and submission.

7
  With proper planning, task distribution, and time management, the system can
be successfully designed, implemented, tested, and delivered within the given
academic schedule.
S. Week 1st 2n 3r 4t 5t 6t 7t 8t 9t 10t 11t 12t 13t 14t
d d h h h h h h h h h h h
N Work
1. Research, Topic
selection and
proposal
2. Requirement
Gathering and
planning
3. System Design
And
architecture
4. Development

5. Model
Integration
6. Testing and bug
Fixing
7. Security and
Optimization
8. Final
Integration,
Documentation

Fig: Gantt Chart of project timeline

8
4.3 High-Level Design of the System

System Architecture:

1. User Interface:

Frontend for resume input and job description and result output

2. Backend (Node/Express):

 Handles routing, form submission, and DB operations

 Implements authentication and authorization

 Passes data to model and returns results

3. Database:

 Stores users’ data, login information, resume data, job description, re-
sult of NLP model

4. Algorithms Used:
 Tokenization Algorithm: splits raw text from resume and job descrip-
tion into tokens for further processing
 Stemming Algorithm: Reduces words to their root form to standard-
ize variations of the same term.
 Lemmatization Algorithm: Reduces words to their dictionary form
(lemma) while preserving meaning, considering context and part-of-
speech
 Part-Of-Speech Tagging Algorithm: Assigns grammatical categories
(e.g., noun, verb, adjective) to tokens to support lemmatization and en-
tity recognition.
 Named Entity Recognition Algorithm: Identifies and classifies
named entities (e.g., person names, organizations) in text for
anonymization and contextual analysis.

9
 BERT Algorithm: Generates contextualized embeddings for semantic
matching between resumes and job descriptions.
 Cosine Similarity Algorithm: Measures the similarity between re-
sume and job description embeddings to rank candidates.

10
 Flowchart of the system

Fig: Flowchart of the proposed system

11
5. Expected Outcomes
 A full stack web app for job recommendation and resume screening

 Candidate and job ranking based on custom filters

 Improved User experience for both job seekers and recruiters

 Enhanced efficiency in recruitment and job application process

 Bias Reduction and fairer decision-making by focusing on skills and

experience

 Significant reduction in manual effort, saving time and cost

[Link]
[1] M. Alamelu, D. Kumar, R. Sanjana, J. Sree, A. Devi, and D. Kavitha, “Resume
Validation and Filtration using Natural Language Processing,” 2021 10th Int. Conf. on
Internet of Everything, Microwave Engineering, Communication and Networks, 2021.
[Online]. Available: [Link]

[2] S. Chakraborty et al., “A survey on application of Natural Language Processing in

Recruitment Domain,” Procedia Computer Science, vol. 167, pp. 2060–2068, 2020.
[Online]. Available: [Link]

[3] J. Devlin, M.-W. Chang, K. Lee, and K. Toutanova, “BERT: Pre-training of Deep
Bidirectional Transformers for Language Understanding,” in Proc. NAACL-HLT,
2019. [Online]. Available: [Link]

[4] N. J. S. Rani, A. B. S. Reddy, and A. V. N. Kumar, “NLP-Based Bi-Directional

Recommendation System: Towards Recommending Jobs to Job Seekers and Resumes
to Recruiters,” Big Data Cogn. Comput., vol. 6, no. 4, p. 147, 2022. [Online].
Available: [Link]

[5] “Cosine similarity,” Wikipedia, [Online]. Available:

[Link]

[6] “Wikipedia,” Wikipedia, [Online]. Available: [Link]

12
[7] M. Meena, D. Sharma, and A. Singh, “Application of Natural Language
Processing in Recruitment,” Journal of Ambient Intelligence and Humanized
Computing, 2024. [Online]. Available:
[Link]

[8] D. Kumar and P. Sinha, “Resume Parser Analysis Using Machine Learning and
Natural Language Processing,” [Link], 2023. [Online]. Available:
[Link]
arning_and_Natural_Language_Processing

[9] H. Wang, Y. Zhang, and L. Lin, “A Joint Embedding Model for Job-Resume
Matching,” arXiv preprint arXiv:2109.06501, 2021. [Online]. Available:
[Link]

[10] B. Raghavan, S. Barocas, J. Kleinberg, and K. Levy, “Mitigating Bias in

Algorithmic Hiring: Evaluating Claims and Practices,” arXiv preprint
arXiv:1906.09208, 2019. [Online]. Available: [Link]

13