R2Code: A Self-Reflective LLM Framework for Requirements-to-Code Traceability

Yifei Wang; Jacky Keung; Xiaoxue Ma; Zhenyu Mao; Kehui Chen; Yishu Li

R2Code: A Self-Reflective LLM Framework for Requirements-to-Code Traceability

Yifei Wang
, Jacky Keung
, Xiaoxue Ma
, Zhenyu Mao
, Kehui Chen
, Yishu Li^*

^*Corresponding author for this work

Department of Computer Science

Research output: Chapters, Conference Papers, Creative and Literary Works › RGC 32 - Refereed conference paper (with host publication) › peer-review

Abstract

Accurate requirement-to-code traceability is crucial for software maintenance. However, existing IR- and embedding-based methods are heavily dependent on lexical similarity, often yielding incomplete or inconsistent links across projects and languages and incurring high cost from long-context retrieval and prompting.

This paper presents R2Code, an LLM-based semantic traceability framework designed to improve trace link accuracy while reducing inference cost. R2Code integrates three components:
1) a decomposition-enhanced Bidirectional Alignment Network (BAN) that aligns four-layer requirement semantics with corresponding code structures to support cross-level semantic matching;
2) a Self-Reflective Consistency Verification (SRCV) module that conducts explanation-guided consistency checking to calibrate link reliability; and
3) a Dynamic Context-Adaptive Retrieval (DCAR) mechanism that adjusts retrieval granularity and filters contexts using semantic-overlap weighting for efficient context utilization.

Experiments on five public datasets spanning multiple domains and two programming languages demonstrate that R2Code consistently outperforms the strongest baselines, achieving an average F1 gain of 7.4%, while reducing token consumption by up to 41.7% through adaptive context control.

Original language	English
Title of host publication	2026 IEEE 50th Annual Computers, Software, and Applications Conference (COMPSAC)
Publication status	Published - 7 Jul 2026
Event	2026 IEEE 50th Annual Computers, Software, and Applications Conference (COMPSAC): Agentic AI: Innovative Software, Computing Paradigms, and Applications for Migrating from Reactive Models to Proactive Systems - Escuela Técnica Superior de Ingenieros de Caminos, Canales y Puertos, Madrid, Spain Duration: 7 Jul 2026 → 10 Jul 2026 https://ieeecompsac.computer.org/2026/

Conference

Conference	2026 IEEE 50th Annual Computers, Software, and Applications Conference (COMPSAC)
Abbreviated title	COMPSAC 2026
Place	Spain
City	Madrid
Period	7/07/26 → 10/07/26
Internet address	https://ieeecompsac.computer.org/2026/

Bibliographical note

Information for this record is supplemented by the author(s) concerned.

Since this conference is yet to commence, the information for this record is subject to revision.

Other Access Options

Check CityUHK Library

Permanent Link

Right click to copy link

Cite this

@inproceedings{0ef10a8017f642269f0feb5ca76dfb5f,

title = "R2Code: A Self-Reflective LLM Framework for Requirements-to-Code Traceability",

abstract = "Accurate requirement-to-code traceability is crucial for software maintenance. However, existing IR- and embedding-based methods are heavily dependent on lexical similarity, often yielding incomplete or inconsistent links across projects and languages and incurring high cost from long-context retrieval and prompting. This paper presents R2Code, an LLM-based semantic traceability framework designed to improve trace link accuracy while reducing inference cost. R2Code integrates three components: 1) a decomposition-enhanced Bidirectional Alignment Network (BAN) that aligns four-layer requirement semantics with corresponding code structures to support cross-level semantic matching; 2) a Self-Reflective Consistency Verification (SRCV) module that conducts explanation-guided consistency checking to calibrate link reliability; and 3) a Dynamic Context-Adaptive Retrieval (DCAR) mechanism that adjusts retrieval granularity and filters contexts using semantic-overlap weighting for efficient context utilization. Experiments on five public datasets spanning multiple domains and two programming languages demonstrate that R2Code consistently outperforms the strongest baselines, achieving an average F1 gain of 7.4\%, while reducing token consumption by up to 41.7\% through adaptive context control.",

author = "Yifei Wang and Jacky Keung and Xiaoxue Ma and Zhenyu Mao and Kehui Chen and Yishu Li",

note = "Information for this record is supplemented by the author(s) concerned. Since this conference is yet to commence, the information for this record is subject to revision.; 2026 IEEE 50th Annual Computers, Software, and Applications Conference (COMPSAC) : Agentic AI: Innovative Software, Computing Paradigms, and Applications for Migrating from Reactive Models to Proactive Systems, COMPSAC 2026 ; Conference date: 07-07-2026 Through 10-07-2026",

year = "2026",

month = jul,

day = "7",

language = "English",

booktitle = "2026 IEEE 50th Annual Computers, Software, and Applications Conference (COMPSAC)",

url = "https://ieeecompsac.computer.org/2026/",

}

TY - GEN

T1 - R2Code

T2 - 2026 IEEE 50th Annual Computers, Software, and Applications Conference (COMPSAC)

AU - Wang, Yifei

AU - Keung, Jacky

AU - Ma, Xiaoxue

AU - Mao, Zhenyu

AU - Chen, Kehui

AU - Li, Yishu

N1 - Information for this record is supplemented by the author(s) concerned. Since this conference is yet to commence, the information for this record is subject to revision.

PY - 2026/7/7

Y1 - 2026/7/7

N2 - Accurate requirement-to-code traceability is crucial for software maintenance. However, existing IR- and embedding-based methods are heavily dependent on lexical similarity, often yielding incomplete or inconsistent links across projects and languages and incurring high cost from long-context retrieval and prompting. This paper presents R2Code, an LLM-based semantic traceability framework designed to improve trace link accuracy while reducing inference cost. R2Code integrates three components: 1) a decomposition-enhanced Bidirectional Alignment Network (BAN) that aligns four-layer requirement semantics with corresponding code structures to support cross-level semantic matching; 2) a Self-Reflective Consistency Verification (SRCV) module that conducts explanation-guided consistency checking to calibrate link reliability; and 3) a Dynamic Context-Adaptive Retrieval (DCAR) mechanism that adjusts retrieval granularity and filters contexts using semantic-overlap weighting for efficient context utilization. Experiments on five public datasets spanning multiple domains and two programming languages demonstrate that R2Code consistently outperforms the strongest baselines, achieving an average F1 gain of 7.4%, while reducing token consumption by up to 41.7% through adaptive context control.

AB - Accurate requirement-to-code traceability is crucial for software maintenance. However, existing IR- and embedding-based methods are heavily dependent on lexical similarity, often yielding incomplete or inconsistent links across projects and languages and incurring high cost from long-context retrieval and prompting. This paper presents R2Code, an LLM-based semantic traceability framework designed to improve trace link accuracy while reducing inference cost. R2Code integrates three components: 1) a decomposition-enhanced Bidirectional Alignment Network (BAN) that aligns four-layer requirement semantics with corresponding code structures to support cross-level semantic matching; 2) a Self-Reflective Consistency Verification (SRCV) module that conducts explanation-guided consistency checking to calibrate link reliability; and 3) a Dynamic Context-Adaptive Retrieval (DCAR) mechanism that adjusts retrieval granularity and filters contexts using semantic-overlap weighting for efficient context utilization. Experiments on five public datasets spanning multiple domains and two programming languages demonstrate that R2Code consistently outperforms the strongest baselines, achieving an average F1 gain of 7.4%, while reducing token consumption by up to 41.7% through adaptive context control.

M3 - RGC 32 - Refereed conference paper (with host publication)

BT - 2026 IEEE 50th Annual Computers, Software, and Applications Conference (COMPSAC)

Y2 - 7 July 2026 through 10 July 2026

ER -

R2Code: A Self-Reflective LLM Framework for Requirements-to-Code Traceability

Abstract

Conference

Bibliographical note

Other Access Options

Permanent Link

Fingerprint

Cite this