CSS: Contrastive Semantic Similarity for Uncertainty Quantification of LLMs

TL;DR

This paper introduces a novel CLIP-based contrastive semantic similarity method to improve uncertainty quantification in large language models, leading to more reliable response filtering in question-answering tasks.

Contribution

It proposes a new CLIP-based feature extraction approach for better uncertainty estimation in LLMs, surpassing traditional NLI-based methods.

Findings

Outperforms baseline methods in estimating LLM response reliability

Effective in filtering unreliable LLM generations in QA tasks

Demonstrates robustness across multiple LLMs and datasets

Abstract

Despite the impressive capability of large language models (LLMs), knowing when to trust their generations remains an open challenge. The recent literature on uncertainty quantification of natural language generation (NLG) utilises a conventional natural language inference (NLI) classifier to measure the semantic dispersion of LLMs responses. These studies employ logits of NLI classifier for semantic clustering to estimate uncertainty. However, logits represent the probability of the predicted class and barely contain feature information for potential clustering. Alternatively, CLIP (Contrastive Language-Image Pre-training) performs impressively in extracting image-text pair features and measuring their similarity. To extend its usability, we propose Contrastive Semantic Similarity, the CLIP-based feature extraction module to obtain similarity features for measuring uncertainty for text pairs. We apply this method to selective NLG, which detects and rejects unreliable generations for better trustworthiness of LLMs. We conduct extensive experiments with three LLMs on several benchmark question-answering datasets with comprehensive evaluation metrics. Results show that our proposed method performs better in estimating reliable responses of LLMs than comparable baselines. Results show that our proposed method performs better in estimating reliable responses of LLMs than comparable baselines. The code are available at \url{https://github.com/AoShuang92/css_uq_llms}.