MeTMaP: Metamorphic Testing for Detecting False Vector Matching Problems in LLM Augmented Generation

• DOI: 10.48550/arxiv.2402.14480
• Publication Date: 2024-02-22
• Authors: Guanyu Wang; Yuekang Li; Yi Liu; Gelei Deng; Tianlin Li; Guosheng Xu; Yang Liu; Haoyu Wang; Kailong Wang

Augmented generation techniques such as Retrieval-Augmented Generation (RAG) and Cache-Augmented (CAG) have revolutionized the field by enhancing large language model (LLM) outputs with external knowledge cached information. However, integration of vector databases, which serve a backbone for these augmentations, introduces critical challenges, particularly in ensuring accurate matching. False matching databases can significantly compromise integrity reliability LLM outputs, leading to misinformation or erroneous responses. Despite crucial impact issues, there is notable research gap methods effectively detect address false matches LLM-augmented generation. This paper presents MeTMaP, metamorphic testing framework developed identify systems. We derive eight relations (MRs) from six NLP datasets, form our method's core, based on idea that semantically similar texts should match dissimilar ones not. MeTMaP uses MRs create sentence triplets testing, simulating real-world scenarios. Our evaluation over 203 configurations, involving 29 embedding models 7 distance metrics, uncovers significant inaccuracies. The results, showing maximum accuracy only 41.51\% tests compared original emphasize widespread issue need effective detection mitigation applications.