Copyright © 2026 Authors retain the copyright of this article. This article 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.

@article{209524,
        author = {Sheetal Munna Rajbhar and Ms. Cynthia Shinde},
        title = {AI-Based Duplicate Bug Detection and Intelligent Bug Relationship Analysis Using Natural Language Processing},
        journal = {International Journal of Innovative Research in Technology},
        year = {2026},
        volume = {13},
        number = {5},
        pages = {2104-2108},
        issn = {2349-6002},
        url = {https://ijirt.org/article?manuscript=209524},
        abstract = {Most bug tracking systems used in the software development lifecycle treat individual bug reports as isolated entries with no inherent means of determining if one report describes the same defect as another. Given two reports, one stating “the application crashes when I try to log in” and another stating “the application closes after submission of login credentials,” it is entirely reasonable to assume that they describe the same bug, but a keyword-based system would be incapable of recognizing this. This project implements a bug management system with an intelligent layer recognizing potential duplicates based on semantic similarities between reports using sentence embeddings and TF-IDF comparison, providing an alternative to the usual create-view-update-assign cycle in bug tracking software. For each new report, semantic similarity is calculated using two different approaches – cosine similarity of TF-IDF vectors and cosine similarity of pretrained Sentence-Transformer embeddings – and the report is classified as a potential duplicate, related to an existing report, or distinct based on the combined similarity score. Additionally, a rule-based system is used to assign a priority and severity classification to each report with natural language reasoning provided for each recommendation to allow for manual review. The system was implemented as a functioning web application using FastAPI, MySQL, and React instead of a standalone script, and benchmarked using a manually-created labeled dataset of eight reports with three pairs of duplicates.},
        keywords = {Duplicate Bug Detection, Natural Language Processing, Semantic Similarity, Sentence Embeddings, Transformer Models, TF-IDF, Cosine Similarity, Bug Triage, Software Testing},
        month = {October},
        }