Yuchen Fu, Zifeng Cheng, Zhiwei Jiang, Zhonghui Wang, Yafeng Yin, Zhengliang Li, Qing Gu. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

Bug localization, which aims to localize software faults specific to a bug report, is valuable for improving software developers' efficiency. This task is often formulated as an information retrieval problem, where potentially buggy files are retrieved and ranked according to their textual similarity to a target bug report. To address this task, many methods have been proposed, primarily focusing on resolving the semantic gap between natural language in bug reports and programming language in source files. Recently, Large Language Models (LLMs) have demonstrated strong capabilities in seamlessly understanding both natural language and programming language, potentially addressing the semantic gap issue through natural conversation. However, the limited context length of prompt and capability of long-context comprehension in existing LLMs make it impossible to directly load the entire codebase (e.g., thousands of code files) into LLMs' prompts to retrieve buggy code files. In this paper, we explore how to leverage existing LLMs for projectlevel bug localization without requiring additional fine-tuning and propose an LLM-based Bug Localization framework, LLM-BL. Our core idea is to enable LLMs to perform bug localization tasks using listwise ranking instructions and to avoid exceeding the context length limit by compressing the codebase through file filtering and content reduction. Specifically, LLM-BL consists of four modules: report expansion, candidate retrieval, content reduction, and LLM-based bug localization. Among them, the first three modules are used to retrieve potential buggy code files and extract bug-related information from the code files, respectively, while the last module enables the LLM to perform bug localization through listwise file ranking. We select three widely used LLMs (i.e., ChatGPT, Llama 3, CodeLlama) as the base LLMs in our framework and conduct extensive experiments on six public projects with two report types (i.e., Java and Python). Experimental results demonstrate that the two LLMs, ChatGPT and Llama 3, can more effectively understand task intentions and more accurately localize buggy files than CodeLlama. Compared to existing methods, LLM-BL achieves better localization performance without requiring any fine-tuning, in a plug-and-play manner. These results demonstrate that both ChatGPT and Llama 3 are effective zero-shot rankers for bug localization.

Summary With the development of smartphones, mobile applications play an irreplaceable role in our daily life, which characteristics often commit code changes to meet new requirements. This characteristic can introduce defects into the software. To provide immediate feedback to developers, previous researchers began to focus on just‐in‐time (JIT) software defect prediction techniques. JIT defect prediction aims to determine whether code commits will introduce defects into the software. It contains two scenarios, within‐project JIT defect prediction and cross‐project JIT defect prediction. Regardless of whether within‐project JIT defect prediction or cross‐project JIT defect prediction all need to have enough labeled data (within‐project JIT defect prediction assumes that have plenty of labeled data from the same project, while cross‐project JIT defect prediction assumes that have sufficient labeled data from source projects). However, in practice, both the source and target projects may only have limited labeled data. We propose the MTL‐DNN method based on multi‐task learning to solve this question. This method contains the data preprocessing layer, input layer, shared layers, task‐specific layers, and output layer. Where the common features of multiple related tasks are learned by sharing layers, and the unique features of each task are learned by the task‐specific layers. For verifying the effectiveness of the MTL‐DNN approach, we evaluate our method on 15 Android mobile apps. The experimental results show that our method significantly outperforms the state‐of‐the‐art single‐task deep learning and classical machine learning methods. This result shows that the MTL‐DNN method can effectively solve the problem of insufficient labeled training data for source and target projects.