ETF: An Entity Tracing Framework for Hallucination Detection in Code Summaries

TL;DR

This paper introduces ETF, a novel framework that uses static analysis and LLMs to detect hallucinations in code summaries, supported by a new dataset and achieving high accuracy.

Contribution

The paper presents the first dataset for hallucination detection in code summarisation and proposes ETF, a new entity tracing framework combining static analysis and LLM verification.

Findings

Achieved 73% F1 score in hallucination detection

Created the first dataset with ~10K samples for this task

Demonstrated ETF's effectiveness in localising errors

Abstract

Recent advancements in large language models (LLMs) have significantly enhanced their ability to understand both natural language and code, driving their use in tasks like natural language-to-code (NL2Code) and code summarisation. However, LLMs are prone to hallucination, outputs that stray from intended meanings. Detecting hallucinations in code summarisation is especially difficult due to the complex interplay between programming and natural languages. We introduce a first-of-its-kind dataset, CodeSumEval, with ~10K samples, curated specifically for hallucination detection in code summarisation. We further propose a novel Entity Tracing Framework (ETF) that a) utilises static program analysis to identify code entities from the program and b) uses LLMs to map and verify these entities and their intents within generated code summaries. Our experimental analysis demonstrates the framework's effectiveness, leading to a 73% F1 score. The proposed approach provides a method for detecting hallucinations by tracing entities from the summary to the code, allowing us to evaluate summary accuracy and localise the error within the summary.