BCL: Bayesian In-Context Learning Framework for Information Extraction

2026-06-16

AI Summary

Current large language models exhibit unstable in-context learning performance and limited generalization capabilities on information extraction tasks. This work proposes the BCL framework, which, for the first time, integrates Bayesian inference with particle filtering into in-context learning. Through a systematic four-step process—initialization, observation, weight update, and resampling—the framework optimizes label representations in a principled manner. BCL provides a unified approach applicable to both sequence labeling and relation classification, the two dominant paradigms in information extraction. Extensive experiments across multiple benchmark tasks demonstrate that BCL consistently and significantly outperforms existing methods, exhibiting superior robustness and generalization ability.

Abstract

Existing information extraction (IE) tasks increasingly adopt in-context learning (ICL) with large language models. However, current approaches either show inconsistent performance across model scales or lack systematic optimization and generalizability. Building on this, we propose BCL (Bayesian In-Context Learning Framework for Information Extraction), the first optimization framework that uses particle filtering with Bayesian updates to systematically refine label representations across IE tasks. Through four steps initialization, observation, weight update, and resampling, BCL generalizes to both sequence labeling and relation classification paradigms. Extensive experiments demonstrate substantial and consistent improvements over existing approaches.

Problem

• information extraction
• in-context learning
• generalizability
• optimization
• large language models

Innovation

• Bayesian In-Context Learning
• Particle Filtering
• Information Extraction
• Label Representation Optimization
• Generalizable Framework