arXiv subjects

QiHong Chen

Publications and source records attributed to QiHong Chen.

3 recordsLinked to original sources

arXiv2026

ProjAgent: Procedural Similarity Retrieval for Repository-Level Code Generation

Repository-level code generation requires implementing target functions while accounting for complex cross-file dependencies and project-specific conventions. Existing retrieval methods predominantly rely on lexical, structural, or semantic similarity, often overlooking repository functions that implement similar procedural logic despite differing in identifiers or application domains. We propose ProjAgent, a repository-level code generation system that introduces procedural similarity as an explicit retrieval signal. ProjAgent decomposes the target function into intermediate reasoning steps and employs an agentic workflow to retrieve repository functions that exhibit similar procedural behavior at each step. The retrieved procedural context is integrated with conventional semantic retrieval to construct a richer repository context for code generation. ProjAgent further incorporates a conservative static-analysis feedback loop that iteratively repairs generated code using compiler and static-analysis feedback. Evaluated on REPOCOD, ProjAgent achieves 41.14% Pass@1, outperforming existing retrieval-based baselines. These results demonstrate that procedural similarity is an effective and previously unexplored retrieval dimension for repository-level code generation.

cs.SE↗

arXiv2025

From Bias To Improved Prompts: A Case Study of Bias Mitigation of Clone Detection Models

The issue of clone code has persisted in software engineering, primarily because developers often copy and paste code segments. This common practice has elevated the importance of clone code detection, garnering attention from both software engineering researchers and industry professionals. Their collective concern arises from the potential negative impacts that clone code can have on software quality. The emergence of powerful Generative Large Language Models (LLMs) like ChatGPT has exacerbated the clone code problem. These advanced models possess code generation capabilities that can inadvertently create code clones. As a result, the need to detect clone code has become more critical than ever before. In this study, we assess the suitability of LLMs for clone code detection. Our results demonstrate that the Palm model achieved a high F1 score of 89.30 for the avatar dataset and 86.41 for the poolC dataset. A known issue with LLMs is their susceptibility to prompt bias, where the performance of these models fluctuates based on the input prompt provided. In our research, we delve deeper into the reasons behind these fluctuations and propose a framework to mitigate prompt bias for clone detection. Our analysis identifies eight distinct categories of prompt bias, and our devised approach leveraging these biases yields a significant improvement of up to 10.81% in the F1 score. These findings underscore the substantial impact of prompt bias on the performance of LLMs and highlight the potential for leveraging model errors to alleviate this bias.

cs.SE↗

arXiv2024

A Deep Dive Into Large Language Model Code Generation Mistakes: What and Why?

Recent advancements in Large Language Models (LLMs) have led to their widespread application in automated code generation. However, these models can still generate defective code that deviates from the specification. Previous research has mainly focused on the mistakes in LLM-generated standalone functions, overlooking real-world software development situations where the successful generation of the code requires software contexts such as external dependencies. In this paper, we considered both of these code generation situations and identified a range of \textit{non-syntactic mistakes} arising from LLMs' misunderstandings of coding question specifications. Seven categories of non-syntactic mistakes were identified through extensive manual analyses, four of which were missed by previous works. To better understand these mistakes, we proposed six reasons behind these mistakes from various perspectives. Moreover, we explored the effectiveness of LLMs in detecting mistakes and their reasons. Our evaluation demonstrated that GPT-4 with the ReAct prompting technique can achieve an F1 score of up to 0.65 when identifying reasons for LLM's mistakes, such as misleading function signatures. We believe that these findings offer valuable insights into enhancing the quality of LLM-generated code.

cs.SE↗