Zejun Zhang
Zejun Zhang 张则君
Assistant Professor · Nanjing University
南京大学 · 准聘助理教授
中文

About

简介

I am an Assistant Professor in Software Institute at Nanjing University. Before joining Nanjing University, I was a Research Scientist at Nanyang Technological University (NTU), advised by Prof. Yi Li. I received my Ph.D. from the School of Computing at the Australian National University (ANU), advised by Prof. Zhenchang Xing.

现任南京大学准聘助理教授。此前任职于南洋理工大学（NTU），担任 Research Scientist。于澳大利亚国立大学（ANU）计算机学院获博士学位，导师为 邢振昌教授。

研究方向聚焦可信智能软件工程，覆盖软件全生命周期，重点关注 AI 辅助开发的可理解性、可靠性与一致性。首先致力于实现人与 AI 之间对程序语义、开发意图及业务需求的双向理解；在此基础上，借助程序分析与重构优化技术，保证 AI 生成代码的质量；最终在软件持续迭代过程中，维持代码、文档、UI、配置、接口等各类工件之间的语义同步。

在 ICSE、FSE、TSE 等软件工程领域顶级会议和期刊上发表多篇论文，并获 ICSE 2024 ACM SIGSOFT 杰出论文奖。重视学术成果与实际开发的衔接，积极将研究落地为可用工具，已开发 LintCFG AI Skill（DSL 驱动的代码规范配置生成工具链）、RIdiom IDE 插件、RefactoringIdioms Python 包，以及 DeIdiom 在线平台。

My research focuses on trustworthy intelligent software engineering, addressing the entire software lifecycle with a progressive focus on three core objectives of AI-assisted development: understandability, reliability, and consistency. The work aims to first achieve mutual understanding between developers and AI systems regarding program semantics, developer intent, and business requirements. It then employs program analysis, refactoring, and optimization to ensure the quality and reliability of AI-generated code. Finally, as software evolves, it maintains semantic synchronization across all artifacts including code, documentation, UI, configurations, and APIs to keep them consistent throughout iterative development.

I have published papers in top software engineering venues, including ICSE, FSE, and TSE. My work received the ACM SIGSOFT Distinguished Paper Award at ICSE 2024.

A defining theme of my research is bridging research and practice through usable tools. I developed the LintCFG AI skill, a DSL-driven linter configuration generation toolchain, followed by the RIdiom IDE plugin, the RefactoringIdioms Python package, and the DeIdiom web platform, bringing research ideas into practical software development tools.

Beyond tools, I also release open-source benchmarks and datasets for the community, including MULocBench, a benchmark for localizing code and non-code issues in software projects.

除工具之外，我也向社区开源基准数据集，包括面向软件项目中代码与非代码问题定位的基准 MULocBench。

I am looking for motivated undergraduate, Master’s, and Ph.D. students. If you are interested, please email me your CV at zejun.zhang@nju.edu.cn.

欢迎本科生、硕士生与博士生加入。请将简历发送至 zejun.zhang@nju.edu.cn。

Publications

论文

* Corresponding author * 通讯作者

Awards

荣誉

Invited Talks

受邀报告

Service

学术服务

Journal Reviewer

期刊审稿

Conference Reviewer

会议审稿

Benchmarks

基准数据集

MULocBench

Benchmark for Issue Localization · 1,100 Issues · 46 Python Projects
Issue 定位基准 · 1,100 个 Issue · 46 个 Python 项目

A benchmark for localizing code and non-code issues in software projects. Beyond pull-request issues and code locations, MULocBench covers diverse issue types, root causes, location scopes, and file types — including commits, comments, configurations, and documentation — offering a realistic testbed for issue resolution. State-of-the-art localization methods and five LLM-based prompting strategies remain below 40% Acc@5 and F1 even at the file level.

面向软件项目中代码与非代码问题定位的基准数据集。区别于已有基准仅关注 Pull Request 类 Issue 与代码位置，MULocBench 覆盖更多样的 Issue 类型、 根因、定位范围与文件类型（包括提交、评论、配置文件与文档），为 Issue 解决提供了更贴近真实的评测平台。实验显示，即使是文件级定位，当前最优 方法与五种 LLM 提示策略的 Acc@5 与 F1 仍低于 40%。

Paper Dataset

Tools

研究工具

LintCFG

AI-Powered Linter Configuration · AI Skill · Web Platform · Python Package
AI 驱动的代码规范配置工具 · AI Skill · 在线平台 · Python 包

Automatically transforms coding standards into formal rules and validated linter configurations, enabling end-to-end code quality enforcement.

将自然语言代码规范自动转换为形式化规则与经过验证的 Linter 配置，支持端到端的软件质量保障。

AI Skill → Web Platform → Python Package →

RIdiom

Python Code Refactoring Tool · Python Package · JetBrains IDE Plugin
Python 代码重构工具 · Python 第三方库 · JetBrains IDE 插件

An automated tool for refactoring and improving Python code.

面向 Python 的自动化代码重构工具。

PyPI → JetBrains Plugin →

DeIdiom

Code Comprehension Tool · Web Platform
代码理解工具 · 在线平台

Transforms hard-to-understand Python code into more readable equivalents to facilitate code understanding — the system behind the ICSE 2024 Distinguished Paper.

将难理解的 Python 代码转换为更易读的等价代码，帮助开发者理解代码——相关工作获 ICSE 2024 杰出论文奖。

Web Platform →

Teaching

教学

Gray PDF / Code / Data buttons mean the URL is not filled in yet.

灰色的 PDF / Code / Data 表示链接尚未填写。