What question did this study set out to answer?

This research evaluates transformer-based text summarization models for analyzing resumes in recruitment.

April 4, 2026

Evaluation of Transformer-based Text Summarization Models in Both Extractive and Abstractive Approaches on Resumes

MMMadhu Bala MyneniVignana Jyothi Institute of Management KAKavuri Sai AkarshVignana Jyothi Institute of Management MSMalloju SathvikaVignana Jyothi Institute of Management

Key Points

This research evaluates transformer-based text summarization models for analyzing resumes in recruitment.
Implemented both abstractive and extractive summarization techniques using transformer-based approaches.
Conducted experiments using BART for abstractive summarization and BERT, TextRank, and Luhn for extractive summarization.
Utilized Cosine similarity, Semantic BERT, and ROUGE-L to evaluate summarization effectiveness.
Analyzed real-time resumes to assess model performance.
Extractive summarization models, particularly BERT, outperformed classical methods like TextRank and Luhn.
BERT achieved a higher average similarity score of 84.34% compared to BART's 76.40% for job descriptions.
BERT produced more concise and fact-based summaries, while BART provided details with clearer context.

Abstract

Introduction:: Text summarization is applied to resume analysis in an AI-enabled recruitment process that needs a deeper contextual understanding. It can simplify the hiring process by quickly distilling essential candidate details. This research focuses on the evaluation of transformer-based models' performance with a specific emphasis on the development of an AI-enabled autonomous tool for resume analysis in the recruitment process. The study highlights both abstractive and extractive summarization techniques, utilizing large language models for the analysis. Methods:: It was implemented using both abstractive and extractive summarization with a transformer-based approach. Experiments were conducted using BART for abstractive summarization and BERT, TextRank, and Luhn for extractive summarization on resumes. Summaries were evaluated for job fit score using Cosine similarity, Semantic BERT, and ROUGE-L measures. Models were evaluated using real-time resumes collected from the study area. Results:: The evaluation results show that the transformer-based extractive summarization outperformed classical extraction methods (TextRank, Luhn) and it was observed that the extractive summarization (BERT model) achieved a higher average similarity score (84.34%) with job descriptions compared to the abstractive approach (BART model) (76.40%). BERT delivered more concise and fact-based summaries, while BART generated outputs that were clear and contextually detailed. Discussion:: The findings indicate that BERT offers superior alignment for candidate–job matching, making it highly effective for automated resume screening. However, BART provides richer and more readable summaries, suggesting that the choice of model should balance factual accuracy and interpretive clarity based on recruitment needs. The integration of transformer models can streamline hiring workflows, though further improvements are needed for domain adaptation and hybrid approaches. Conclusion:: Transformer-based language models can significantly enhance the efficiency of resume screening in recruitment. The comparative analysis shows that abstractive and extractive methods each have unique benefits, and choosing between them should depend on the specific needs of the summarization task.

AI에게 질문

Bookmark

View Full Paper