Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Progressive Code Integration for Abstractive Bug Report Summarization

Shaira Sadia Karim, Abrar Mahmud Rahim, Lamia Alam, Ishmam Tashdeed, Lutfun Nahar Lota, Md. Abu Raihan M. Kamal, Md. Azam Hossain

TL;DR

Bug reports are unstructured and verbose, hindering quick comprehension. The authors introduce a progressive code-integration framework that combines textual reports with full code snippets via structured prompts and hierarchical code summarization, with parameter-efficient fine-tuning. The approach outperforms extractive baselines and achieves competitive semantic fidelity relative to state-of-the-art abstractive methods across four datasets and eight LLMs. The work demonstrates the value of jointly leveraging textual and code information for bug comprehension and highlights directions for human evaluation and larger multimodal datasets and metrics.

Abstract

Bug reports are often unstructured and verbose, making it challenging for developers to efficiently comprehend software issues. Existing summarization approaches typically rely on surface-level textual cues, resulting in incomplete or redundant summaries, and they frequently ignore associated code snippets, which are essential for accurate defect diagnosis. To address these limitations, we propose a progressive code-integration framework for LLM-based abstractive bug report summarization. Our approach incrementally incorporates long code snippets alongside textual content, overcoming standard LLM context window constraints and producing semantically rich summaries. Evaluated on four benchmark datasets using eight LLMs, our pipeline outperforms extractive baselines by 7.5%-58.2% and achieves performance comparable to state-of-the-art abstractive methods, highlighting the benefits of jointly leveraging textual and code information for enhanced bug comprehension.

Progressive Code Integration for Abstractive Bug Report Summarization

TL;DR

Bug reports are unstructured and verbose, hindering quick comprehension. The authors introduce a progressive code-integration framework that combines textual reports with full code snippets via structured prompts and hierarchical code summarization, with parameter-efficient fine-tuning. The approach outperforms extractive baselines and achieves competitive semantic fidelity relative to state-of-the-art abstractive methods across four datasets and eight LLMs. The work demonstrates the value of jointly leveraging textual and code information for bug comprehension and highlights directions for human evaluation and larger multimodal datasets and metrics.

Abstract

Bug reports are often unstructured and verbose, making it challenging for developers to efficiently comprehend software issues. Existing summarization approaches typically rely on surface-level textual cues, resulting in incomplete or redundant summaries, and they frequently ignore associated code snippets, which are essential for accurate defect diagnosis. To address these limitations, we propose a progressive code-integration framework for LLM-based abstractive bug report summarization. Our approach incrementally incorporates long code snippets alongside textual content, overcoming standard LLM context window constraints and producing semantically rich summaries. Evaluated on four benchmark datasets using eight LLMs, our pipeline outperforms extractive baselines by 7.5%-58.2% and achieves performance comparable to state-of-the-art abstractive methods, highlighting the benefits of jointly leveraging textual and code information for enhanced bug comprehension.

Paper Structure

This paper contains 21 sections, 21 equations, 2 figures, 6 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Abstractive Bug Report Summarization
  4. Prompt Engineering
  5. Approach
  6. Problem Statement
  7. Data Preprocessing
  8. Prompt Engineering
  9. Fine-Tuning
  10. Research Questions
  11. Experimental Setup
  12. Dataset
  13. Models
  14. Evaluation Metrics
  15. Experimental Configurations
  16. ...and 6 more sections

Figures (2)

  • Figure 1: End-to-end pipeline for abstractive bug report summarization. Here long buggy code is chunked and summarized, combined with bug-report text through prompting and refined using LoRA-based fine-tuning to produce summaries.
  • Figure 2: Overview of the iterative prompt refinement process. Prompts were progressively adjusted based on the quality of model-generated summaries to improve accuracy, completeness, and readability.