Fact-Level Confidence Calibration and Self-Correction
This addresses the issue of inadequate calibration for multi-fact responses in LLMs, enabling better self-evaluation and correction, though it is an incremental improvement over existing methods.
The paper tackles the problem of confidence calibration in LLMs for long-form generation by proposing a Fact-Level Calibration framework that aligns confidence with relevance-weighted correctness at the fact level, and experiments show that their method, ConFix, effectively mitigates hallucinations across four datasets and six models without external knowledge.
Confidence calibration in LLMs, i.e., aligning their self-assessed confidence with the actual accuracy of their responses, enabling them to self-evaluate the correctness of their outputs. However, current calibration methods for LLMs typically estimate two scalars to represent overall response confidence and correctness, which is inadequate for long-form generation where the response includes multiple atomic facts and may be partially confident and correct. These methods also overlook the relevance of each fact to the query. To address these challenges, we propose a Fact-Level Calibration framework that operates at a finer granularity, calibrating confidence to relevance-weighted correctness at the fact level. Furthermore, comprehensive analysis under the framework inspired the development of Confidence-Guided Fact-level Self-Correction ($\textbf{ConFix}$), which uses high-confidence facts within a response as additional knowledge to improve low-confidence ones. Extensive experiments across four datasets and six models demonstrate that ConFix effectively mitigates hallucinations without requiring external knowledge sources such as retrieval systems.