Smart Resume Analyzer & Job Match Recommender: A Machine Learning and NLP-Based Approach for Automated Resume Screening

Abstract

Recruiters today are buried under resumes the moment a job opening goes live, and sorting through them by hand is slow, tiring, and rarely consistent from one reviewer to the next. Qualified applicants sometimes get missed simply because no one had time to read carefully, while candidates on the other end are left guessing why they were turned down. We built a Smart Resume Analyzer & Job Match Recommender to take some of that guesswork out of the process. Resumes uploaded as PDF or DOCX are read using PyMuPDF and python-docx, cleaned up, and compared against a job description supplied by the user. The comparison itself is done through TF-IDF vectorization over unigrams and bigrams, feeding into a Logistic Regression model that was trained to separate \"good\" resume-to-job fits from \"poor\" ones and to report how confident it is in that call. On top of the match prediction, the tool checks the resume\'s text against a list of common technical skills, flags whatever the job asks for that the resume doesn\'t mention, and turns that gap into a short list of things worth learning. Everything runs through a Streamlit interface, so a recruiter or a job seeker can use it without touching any code. In testing against a labelled set of resume-job pairs, the pipeline held up well: it stayed fast, its reasoning was easy to trace back to actual keywords, and it gave applicants something concrete to work on instead of a plain rejection.

Introduction

The text presents a Smart Resume Analyzer & Job Match Recommender designed to make resume screening faster, more transparent, and more useful for job seekers. The system addresses the limitations of manually reviewing large numbers of applications by using Natural Language Processing (NLP), TF-IDF, and Logistic Regression to compare resumes with job descriptions.

Problem

Online job portals have greatly increased the number of applications recruiters receive, while resume screening is still often performed manually. This can lead to:

• Slow and inconsistent screening.
• Reviewer fatigue and changing evaluation standards.
• Qualified candidates being overlooked because they use different terminology.
• Little or no feedback explaining why a candidate was rejected.

The proposed system aims to automate the initial matching process while also providing skill-gap feedback to candidates.

Proposed Approach

The system uses a lightweight pipeline:

Resume + Job Description → Text Extraction → Preprocessing → TF-IDF → Logistic Regression → Skill Matching → Gap Analysis → Recommendations

The main technologies are:

• PyMuPDF: Extracts text from PDF resumes.
• python-docx: Extracts text from DOCX resumes.
• TF-IDF: Converts resume and job-description text into numerical features and gives greater importance to distinctive terms.
• Logistic Regression: Predicts whether the resume and job description are a good match and provides a confidence score.
• Skill-matching module: Identifies skills present in the job description but missing from the resume.
• Streamlit: Provides a simple web interface for users.

Literature Review

Previous research has explored traditional text classification, TF-IDF, recommender systems, Word2Vec, BERT, resume parsing, and Applicant Tracking Systems (ATS).

The text identifies a trade-off:

• TF-IDF and traditional ML methods are inexpensive, fast, and interpretable but have limited semantic understanding.
• Word2Vec and BERT-based approaches provide deeper language understanding but generally require more computational resources.
• Many existing resume-matching systems provide only a match/no-match result without explaining missing skills or suggesting improvements.

The proposed system attempts to fill this gap by combining lightweight matching with actionable skill recommendations.

System Architecture

The system is divided into seven modules:

1. Input handling: Accepts PDF/DOCX resumes and job descriptions.
2. Text extraction: Extracts readable text from uploaded files.
3. Preprocessing: Converts text to lowercase and removes unnecessary characters.
4. Feature extraction: Uses TF-IDF with unigrams and bigrams, limited to 5,000 features.
5. Classification: Logistic Regression predicts the match category and probability.
6. Skill extraction: Compares required job skills with skills found in the resume.
7. Output: Displays the match result, confidence score, matched skills, missing skills, recommendations, and extracted resume text.

Implementation

The prototype was implemented using Python 3.11, Streamlit, scikit-learn, Pandas, PyMuPDF, and python-docx. It was designed to operate on relatively modest hardware, reportedly running during development on an i3-class computer with 4 GB RAM, without requiring a GPU or cloud infrastructure.

The Logistic Regression model uses an 80/20 training-test split, while the TF-IDF vectorizer uses up to 5,000 unigram and bigram features.

Results

The prototype produced outputs such as:

• Good Match: 71.28% match score, with skills including Python, C++, C, R, HTML, and Excel.
• Fair Match: 75% match score, identifying Python, Git, and Agile as matched skills while flagging Docker, Kubernetes, and AWS as missing skills.

The system provides recommendations based on these skill gaps, making the result more useful than a simple acceptance/rejection classification.

Limitations

The main limitation is keyword-based skill matching. For example, a candidate who writes “K8s” instead of “Kubernetes” may incorrectly appear to lack Kubernetes experience. Similarly, different descriptions of the same skill may not be recognized.

The paper also notes that the reported results are representative prototype outputs rather than a comprehensive benchmark. A larger held-out dataset with formal precision, recall, and F1-score evaluation is identified as future work.

Conclusion

What this work adds up to is a resume-matching tool that doesn\'t ask for much in the way of compute, is straightforward to explain to someone without a machine-learning background, and — importantly — doesn\'t stop at a bare match/no-match verdict. Pairing the TF-IDF and Logistic Regression pipeline with the skill-gap and recommendation layer means a candidate walks away from the tool with something to actually work on, not just a rejection. Keeping the whole thing lightweight also means it can realistically run on modest hardware and respond in real time, which was one of the original goals. There\'s a fairly clear list of next steps. Swapping TF-IDF for something like BERT [5] would let the system catch matches that depend on meaning rather than exact wording — the \"K8s\" versus \"Kubernetes\" problem mentioned earlier is a good example of where this would help. Named Entity Recognition could pick up skills and experience that are implied rather than explicitly named. Hooking into live job-portal APIs would turn this from a one-off matcher into something closer to a recommendation engine. On the infrastructure side, adding a proper database for resume history and some basic analytics would make it more useful for repeated use rather than single-session checks. And, as noted above, running the system against a larger and more varied dataset with formal precision, recall, and F1 numbers is the natural next step to back up what\'s currently a promising but still preliminary set of results.

References

[1] F. Sebastiani, \"Machine learning in automated text categorization,\" ACM Computing Surveys, vol. 34, no. 1, pp. 1–47, 2002. [2] C. D. Manning, P. Raghavan, and H. Schütze, Introduction to Information Retrieval. Cambridge, U.K.: Cambridge University Press, 2008. [3] F. Ricci, L. Rokach, and B. Shapira, Recommender Systems Handbook. New York, NY, USA: Springer, 2011. [4] T. Mikolov, K. Chen, G. Corrado, and J. Dean, \"Efficient estimation of word representations in vector space,\" in Proc. Int. Conf. Learning Representations (ICLR), 2013. [5] 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. [6] D. Jurafsky and J. H. Martin, Speech and Language Processing. Boston, MA, USA: Pearson, 2020. [7] P. Pokharel, \"Resume parser using natural language processing,\" 2022. [8] A. K. Sinha, M. A. K. Akhtar, and M. Kumar, \"Automated resume parsing and job domain prediction using machine learning,\" Indian J. Science and Technology, 2023. [9] S. Modak, P. Shinde, A. Tiwari, and S. Nalamwar, \"Resume-job matching using machine learning: A review,\" Int. J. Recent and Innovation Trends in Computing and Communication, 2024. [10] Y. Kanojia et al., \"Resume parser using machine learning,\" Int. J. Novel Research and Development, 2024. [11] S. Yadav, S. Ursal, S. Tate, and A. Thade, \"Resume analysis using NLP and ATS algorithm,\" Int. J. Latest Technology in Engineering, Management & Applied Science, 2025.

Copyright

Copyright © 2026 Mr. Adhikesavan C, Dr. V. Shanmuganeethi. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.