Survey Manuscript

Knowledge-Centric Automated Issue Resolution: A Systematic Survey and Taxonomy¶

The latest manuscript presents a knowledge-centric survey of automated issue resolution, covering 133 studies, a three-layer taxonomy, and six research questions spanning knowledge definition, acquisition, representation, usage, workflow stages, and recent trends.

Open PDF Download PDF

Tables & Resources Citation Page

Surveyed studies 133

Manuscript length 91 pages

Research questions 6

Knowledge layers 3

Preview & Download¶

Manuscript Info¶

Title
Knowledge-Centric Automated Issue Resolution: A Systematic Survey and Taxonomy

Authors
Zhenxi Chen, Mingwei Liu, Zihao Wang, Zhanhui Ren, Leyan Hu, Xueying Du, Ying Wang, Chenxi Zhang, Chong Wang, Zhenchang Xing, Haofen Wang, Xin Peng, and Yanlin Wang

Status
Manuscript submitted to ACM

Public draft
July 2026

Keywords
Issue Resolution, Knowledge, Systematic Literature Review

Quick Reading Guide¶

The paper is best read in three passes: first the abstract and taxonomy overview, then the six RQ-driven analysis sections, and finally the discussion on underexplored knowledge types and future software knowledge infrastructures.

A recurring conclusion across the survey is that current systems are strong on code and tool grounding, but still weak on explicit background knowledge and design-level reasoning.

Abstract¶

Automated issue resolution is a knowledge-intensive software maintenance task in which LLM-based agents must acquire, represent, use, and update knowledge from repositories, execution environments, historical artifacts, documents, tools, and prior repair processes. This survey reviews 133 studies from a knowledge-centric perspective and organizes the field with a three-layer taxonomy: Background Knowledge, Repository Knowledge, and Procedural Knowledge. Across the six analytical dimensions, the paper shows that existing code knowledge and development tool knowledge remain dominant, while repository evolution knowledge and experiential knowledge are rising quickly. In contrast, design architecture knowledge and explicit background knowledge remain comparatively underexplored, suggesting that future systems need more reusable, traceable, and evolvable software knowledge infrastructures.

Citation¶

Recommended Reference Text¶

Zhenxi Chen, Mingwei Liu, Zihao Wang, Zhanhui Ren, Leyan Hu, Xueying Du, Ying Wang, Chenxi Zhang, Chong Wang, Zhenchang Xing, Haofen Wang, Xin Peng, and Yanlin Wang. 2026. Knowledge-Centric Automated Issue Resolution: A Systematic Survey and Taxonomy. Manuscript submitted to ACM.

BibTeX¶

@article{chen2026knowledgecentric,
  title   = {Knowledge-Centric Automated Issue Resolution: A Systematic Survey and Taxonomy},
  author  = {Chen, Zhenxi and Liu, Mingwei and Wang, Zihao and Ren, Zhanhui and Hu, Leyan and Du, Xueying and Wang, Ying and Zhang, Chenxi and Wang, Chong and Xing, Zhenchang and Wang, Haofen and Peng, Xin and Wang, Yanlin},
  year    = {2026},
  note    = {Manuscript submitted to ACM},
  url     = {https://github.com/SYSUSELab/Awesome-Knowledge-for-AIR}
}