Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

CLUE: Concept-Level Uncertainty Estimation for Large Language Models

Yu-Hsiang Wang, Andrew Bai, Che-Ping Tsai, Cho-Jui Hsieh

TL;DR

A novel framework for Concept-Level Uncertainty Estimation (CLUE) for LLMs is proposed, which can provide more interpretable uncertainty estimation results compared with sentence-level uncertainty, and could be a useful tool for various tasks such as hallucination detection and story generation.

Abstract

Large Language Models (LLMs) have demonstrated remarkable proficiency in various natural language generation (NLG) tasks. Previous studies suggest that LLMs' generation process involves uncertainty. However, existing approaches to uncertainty estimation mainly focus on sequence-level uncertainty, overlooking individual pieces of information within sequences. These methods fall short in separately assessing the uncertainty of each component in a sequence. In response, we propose a novel framework for Concept-Level Uncertainty Estimation (CLUE) for LLMs. We leverage LLMs to convert output sequences into concept-level representations, breaking down sequences into individual concepts and measuring the uncertainty of each concept separately. We conduct experiments to demonstrate that CLUE can provide more interpretable uncertainty estimation results compared with sentence-level uncertainty, and could be a useful tool for various tasks such as hallucination detection and story generation.

CLUE: Concept-Level Uncertainty Estimation for Large Language Models

TL;DR

A novel framework for Concept-Level Uncertainty Estimation (CLUE) for LLMs is proposed, which can provide more interpretable uncertainty estimation results compared with sentence-level uncertainty, and could be a useful tool for various tasks such as hallucination detection and story generation.

Abstract

Large Language Models (LLMs) have demonstrated remarkable proficiency in various natural language generation (NLG) tasks. Previous studies suggest that LLMs' generation process involves uncertainty. However, existing approaches to uncertainty estimation mainly focus on sequence-level uncertainty, overlooking individual pieces of information within sequences. These methods fall short in separately assessing the uncertainty of each component in a sequence. In response, we propose a novel framework for Concept-Level Uncertainty Estimation (CLUE) for LLMs. We leverage LLMs to convert output sequences into concept-level representations, breaking down sequences into individual concepts and measuring the uncertainty of each concept separately. We conduct experiments to demonstrate that CLUE can provide more interpretable uncertainty estimation results compared with sentence-level uncertainty, and could be a useful tool for various tasks such as hallucination detection and story generation.
Paper Structure (32 sections, 10 equations, 4 figures, 11 tables)

This paper contains 32 sections, 10 equations, 4 figures, 11 tables.

Table of Contents

  1. Introduction
  2. Motivation
  3. Information Entanglement Issue
  4. Breaking Down Sequences
  5. Related Work
  6. Uncertainty Estimation for LLMs
  7. Hallucination in LLM Generation
  8. Methodology
  9. Concept Extraction
  10. Concept-level Uncertainty Calculation
  11. Concept Scorer
  12. Uncertainty Calculation
  13. Experiments
  14. Experimental Settings
  15. Models
  16. ...and 17 more sections

Figures (4)

  • Figure 1: Our proposed framework of concept-level uncertainty. $o_i$ denotes the $i$-th output sequence, $C_i$ denotes the extracted concepts from $o_i$, and $c_i$ denotes the $i$-th concept in the concept pool.
  • Figure 2: Screenshot of sequence-level human annotation interface presented to MTurkers.
  • Figure 3: Screenshot of concept-level human annotation interface presented to MTurkers.
  • Figure 4: ROC Curves and PR Curves on different thresholds of concept score. The results indicate that our method demonstrates better performance when utilizing tighter thresholds.