On the Performance of Large Language Models for Code Change Intent Classification
Bibliographic record
Abstract
Modern Code Review (MCR) is an essential practice in software engineering, supporting early defect detection, enhancing code quality, and fostering knowledge. To manage code review tasks effectively, developers need to understand the intent behind code changes, such as a bug fix, test, refactoring, or new feature. Traditional methods for categorizing code changes in MCR rely on rule-based heuristics with predefined keywords. However, these methods lack context regarding the code changes, leading to limited generalizability, particularly when dealing with sparsely documented changes. This paper addresses these limitations by investigating the potential of Large Language Models (LLMs) for changes' intent classification. We introduce LLM Change Classifier (LLMCC), an LLM-based approach that classifies code changes based on their underlying intent. We evaluate the effectiveness of LLMCC by conducting an empirical study on three open-source projects: Android, OpenS tack, and Qt. The performance of LLMCC was benchmarked against traditional heuristic methods, conventional machine learning algorithms (including Decision Trees and Random Forests), and state-of-the-art transformer models (including BERT and RoBERTa). Results show that LLMCC significantly enhances code change intent classification accuracy, achieving up to a 33 % improvement in F1 score over heuristic-based methods. Additionally, LLMCC outperformed both traditional machine learning and transformer models, achieving an average 77% improvement in terms of Matthew Correlation Coefficient (MCC). These findings underscore the potential of LLMCC to streamline code change intent classification.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.010 | 0.038 |
| Meta-epidemiology (narrow) | 0.002 | 0.001 |
| Meta-epidemiology (broad) | 0.001 | 0.002 |
| Bibliometrics | 0.005 | 0.002 |
| Science and technology studies | 0.001 | 0.001 |
| Scholarly communication | 0.002 | 0.003 |
| Open science | 0.002 | 0.002 |
| Research integrity | 0.002 | 0.003 |
| Insufficient payload (model declined to judge) | 0.001 | 0.001 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".