arXiv subjects

Yanzhen Zou

Publications and source records attributed to Yanzhen Zou.

12 recordsLinked to original sources

arXiv2026

FeatX: Editing Software by Editing Features for Repository-Level Code Evolution

Large language models (LLMs) are increasingly used for software evolution, yet most interaction paradigms remain code-centric and require manual context management and prompt iteration. We present FeatX, a feature-oriented tool for editing software by editing features. Given an existing repository, FeatX extracts a hierarchical epic-feature structure with explicit feature-to-code mappings, then invokes a three-stage Evolution Agent to translate feature edits into code patches. The workflow is exposed through four coordinated panels. Across a controlled user study and replay experiments on 38 real-world feature-editing commits, FeatX significantly reduces cognitive load and improves usability compared with vanilla ChatGPT. It also achieves a 42.6\% relative improvement in function-level modification localization F1 over strong LLM baselines, at substantially lower cost (\$0.07 in total). The tool and collected dataset are available at https://github.com/a496263365/FeatX/tree/demo, with a demonstration video at https://youtu.be/OZqKZ4Ii-yM.

cs.SE↗

arXiv2025

RepoSummary: Feature-Oriented Summarization and Documentation Generation for Code Repositories

Repository summarization is a crucial research question in development and maintenance for software engineering. Existing repository summarization techniques primarily focus on summarizing code according to the directory tree, which is insufficient for tracing high-level features to the methods that collaboratively implement them. To address these limitations, we propose RepoSummary, a feature-oriented code repository summarization approach that simultaneously generates repository documentation automatically. Furthermore, it establishes more accurate traceability links from functional features to the corresponding code elements, enabling developers to rapidly locate relevant methods and files during code comprehension and maintenance. Comprehensive experiments against the state-of-the-art baseline (HGEN) demonstrate that RepoSummary achieves higher feature coverage and more accurate traceability. On average, it increases the rate of completely covered features in manual documentation from 61.2% to 71.1%, improves file-level traceability recall from 29.9% to 53.0%, and generates documentation that is more conceptually consistent, easier to understand, and better formatted than that produced by existing approaches.

cs.SE↗

arXiv2025

Tool-integrated Reinforcement Learning for Repo Deep Search

Issue localization, the process of identifying code locations that need modification to resolve software issues, is a critical yet challenging task in software development. The semantic gap between natural language issue descriptions and faulty code requires complex multi-hop reasoning through code dependencies. Existing LLM-based agents attempt to address this by integrating repository retrieval tools. However, this transforms issue localization into a demanding task we call Repo Deep Search, which requires the LLM to effectively utilize various repository retrieval tools throughout a multi-step reasoning and navigation process. To tackle this challenge, we present ToolTrain, a two-stage tool-integrated training framework combining rejection-sampled supervised fine-tuning and tool-integrated reinforcement learning to enhance LLMs' ability to use retrieval tools for issue localization. Experimental results show that ToolTrain-trained models achieve state-of-the-art performance, with our 32B model even surpassing Claude-3.7 on function-level localization. The results also show that improved localization performance translates to better end-to-end issue resolution performance. This further demonstrates that training for issue localization is a viable and effective strategy for improving automated software development.

cs.SE↗

arXiv2025

Issue Retrieval and Verification Enhanced Supplementary Code Comment Generation

Issue reports have been recognized to contain rich information for retrieval-augmented code comment generation. However, how to minimize hallucinations in the generated comments remains significant challenges. In this paper, we propose IsComment, an issue-based LLM retrieval and verification approach for generating method's design rationale, usage directives, and so on as supplementary code comments. We first identify five main types of code supplementary information that issue reports can provide through code-comment-issue analysis. Next, we retrieve issue sentences containing these types of supplementary information and generate candidate code comments. To reduce hallucinations, we filter out those candidate comments that are irrelevant to the code or unverifiable by the issue report, making the code comment generation results more reliable. Our experiments indicate that compared with LLMs, IsComment increases the coverage of manual supplementary comments from 33.6% to 72.2% for ChatGPT, from 35.8% to 88.4% for GPT-4o, and from 35.0% to 86.2% for DeepSeek-V3. Compared with existing work, IsComment can generate richer and more useful supplementary code comments for programming understanding, which is quantitatively evaluated through the MESIA metric on both methods with and without manual code comments.

cs.SE↗

arXiv2025

LLMigrate: Transforming "Lazy" Large Language Models into Efficient Source Code Migrators

Rewriting C code in Rust provides stronger memory safety, yet migrating large codebases such as the 32-million-line Linux kernel remains challenging. While rule-based translators (e.g., C2Rust) provide accurate yet largely unsafe Rust programs, recent Large Language Model (LLM) approaches produce more idiomatic, safe Rust programs but frequently exhibit "laziness", omitting significant portions of the target code. To address the issue, in this paper, we present LLMigrate, an LLM-based C-to-Rust translation tool that splits modules into discrete functions, translating them individually, and then reintegrating them. LLMigrate uses static analysis to retain necessary context, pairs GPT-4o (a state-of-the-art LLM) with compiler-driven translation and program-repair techniques for complex core functions, and leverages call-graph-guided translation to ensure consistent interfaces. Evaluations on three representative Linux kernel modules (math, sort, and ramfs) show that LLMigrate requires modifying less than 15\% of the target code, significantly outperforming a pure GPT-4o-based migration.

cs.PL↗

arXiv2025

SoRFT: Issue Resolving with Subtask-oriented Reinforced Fine-Tuning

Mainstream issue-resolving frameworks predominantly rely on commercial models, leading to high costs and privacy concerns. Existing training approaches for issue resolving struggle with poor generalization and fail to fully leverage open-source development resources. We propose Subtask-oriented Reinforced Fine-Tuning (SoRFT), a novel training approach to enhance the issue resolving capability of LLMs. We decomposes issue resolving into structured subtasks: file localization, function localization, line localization, and code edit generation. SoRFT consists of two training stages: (1) rejection-sampled supervised fine-tuning, Chain of Thought (CoT) data is filtered using ground-truth before fine-tuning the LLM, and (2) rule-based reinforcement learning, which leverages PPO with ground-truth based rewards. We evaluate the SoRFT-trained model on SWE-Bench Verified and SWE-Bench Lite, achieving state-of-the-art (SOTA) performance among open-source models (e.g., resolve 21.4% issues on SWE-Bench Verified with SoRFT-Qwen-7B). The experimental results demonstrate that SoRFT significantly enhances issue-resolving performance, improves model generalization, and provides a cost-efficient alternative to commercial models.

cs.SE↗

arXiv2024

Repository Structure-Aware Training Makes SLMs Better Issue Resolver

Language models have been applied to various software development tasks, but the performance varies according to the scale of the models. Large Language Models (LLMs) outperform Small Language Models (SLMs) in complex tasks like repository-level issue resolving, but raise concerns about privacy and cost. In contrast, SLMs are more accessible but under-perform in complex tasks. In this paper, we introduce ReSAT (Repository Structure-Aware Training), construct training data based on a large number of issues and corresponding pull requests from open-source communities to enhance the model's understanding of repository structure and issue resolving ability. We construct two types of training data: (1) localization training data, a multi-level progressive localization data to improve code understanding and localization capability; (2) code edit training data, which improves context-based code editing capability. The evaluation results on SWE-Bench-verified and RepoQA demonstrate that ReSAT effectively enhances SLMs' issue-resolving and repository-level long-context understanding capabilities.

cs.SE↗

arXiv2024

Dehallucinating Parallel Context Extension for Retrieval-Augmented Generation

Large language models (LLMs) are susceptible to generating hallucinated information, despite the integration of retrieval-augmented generation (RAG). Parallel context extension (PCE) is a line of research attempting to effectively integrating parallel (unordered) contexts, while it still suffers from hallucinations when adapted to RAG scenarios. In this paper, we propose DePaC (Dehallucinating Parallel Context Extension), which alleviates the hallucination problem with context-aware negative training and information-calibrated aggregation. DePaC is designed to alleviate two types of in-context hallucination: fact fabrication (i.e., LLMs present claims that are not supported by the contexts) and fact omission (i.e., LLMs fail to present claims that can be supported by the contexts). Specifically, (1) for fact fabrication, we apply the context-aware negative training that fine-tunes the LLMs with negative supervisions, thus explicitly guiding the LLMs to refuse to answer when contexts are not related to questions; (2) for fact omission, we propose the information-calibrated aggregation which prioritizes context windows with higher information increment from their contexts. The experimental results on nine RAG tasks demonstrate that DePaC significantly alleviates the two types of hallucination and consistently achieves better performances on these tasks.

cs.CL↗

arXiv2024

Decomposing God Header File via Multi-View Graph Clustering

God Header Files, just like God Classes, pose significant challenges for code comprehension and maintenance. Additionally, they increase the time required for code recompilation. However, existing refactoring methods for God Classes are inappropriate to deal with God Header Files because the code elements in header files are mostly short declaration types, and build dependencies of the entire system should be considered with the aim of improving compilation efficiency. Meanwhile, ensuring acyclic dependencies among the decomposed sub-header files is also crucial in the God Header File decomposition. This paper proposes a multi-view graph clustering based approach for decomposing God Header Files. It first constructs and coarsens the code element graph, then a novel multi-view graph clustering algorithm is applied to identify the clusters and a heuristic algorithm is introduced to address the cyclic dependencies in the clustering results. To evaluate our approach, we built both a synthetic dataset and a real-world God Header Files dataset. The results show that 1) Our approach could achieve 11.5% higher accuracy than existing God Class refactoring methods; 2) Our decomposition results attain better architecture on real-world God Header Files, evidenced by higher modularity and acyclic dependencies; 3) We can reduce 15% to 60% recompilation time for historical commits that require recompiling.

cs.SE↗

arXiv2024

MESIA: Understanding and Leveraging Supplementary Nature of Method-level Comments for Automatic Comment Generation

Code comments are important for developers in program comprehension. In scenarios of comprehending and reusing a method, developers expect code comments to provide supplementary information beyond the method signature. However, the extent of such supplementary information varies a lot in different code comments. In this paper, we raise the awareness of the supplementary nature of method-level comments and propose a new metric named MESIA (Mean Supplementary Information Amount) to assess the extent of supplementary information that a code comment can provide. With the MESIA metric, we conduct experiments on a popular code-comment dataset and three common types of neural approaches to generate method-level comments. Our experimental results demonstrate the value of our proposed work with a number of findings. (1) Small-MESIA comments occupy around 20% of the dataset and mostly fall into only the WHAT comment category. (2) Being able to provide various kinds of essential information, large-MESIA comments in the dataset are difficult for existing neural approaches to generate. (3) We can improve the capability of existing neural approaches to generate large-MESIA comments by reducing the proportion of small-MESIA comments in the training set. (4) The retrained model can generate large-MESIA comments that convey essential meaningful supplementary information for methods in the small-MESIA test set, but will get a lower BLEU score in evaluation. These findings indicate that with good training data, auto-generated comments can sometimes even surpass human-written reference comments, and having no appropriate ground truth for evaluation is an issue that needs to be addressed by future work on automatic comment generation.

cs.SE↗

arXiv2021

Comprehensive Integration of API Usage Patterns

Nowadays, developers often reuse existing APIs to implement their programming tasks. A lot of API usage patterns are mined to help developers learn API usage rules. However, there are still many missing variables to be synthesized when developers integrate the patterns into their programming context. To deal with this issue, we propose a comprehensive approach to integrate API usage patterns in this paper. We first perform an empirical study by analyzing how API usage patterns are integrated in real-world projects. We find the expressions for variable synthesis is often non-trivial and can be divided into 5 syntax types. Based on the observation, we promote an approach to help developers interactively complete API usage patterns. Compared to the existing code completion techniques, our approach can recommend infrequent expressions accompanied with their real-world usage examples according to the user intent. The evaluation shows that our approach could assist users to integrate APIs more efficiently and complete the programming tasks faster than existing works.

cs.SE↗

arXiv2020

From API to NLI: A New Interface for Library Reuse

Developers frequently reuse APIs from existing libraries to implement certain functionality. However, learning APIs is difficult due to their large scale and complexity. In this paper, we design an abstract framework NLI2Code to ease the reuse process. Under the framework, users can reuse library functionalities with a high-level, automatically-generated NLI (Natural Language Interface) instead of the detailed API elements. The framework consists of three components: a functional feature extractor to summarize the frequently-used library functions in natural language form, a code pattern miner to give a code template for each functional feature, and a synthesizer to complete code patterns into well-typed snippets. From the perspective of a user, a reuse task under NLI2Code starts from choosing a functional feature and our framework will guide the user to synthesize the desired solution. We instantiated the framework as a tool to reuse Java libraries. The evaluation shows our tool can generate a high-quality natural language interface and save half of the coding time for newcomers to solve real-world programming tasks.

cs.SE↗