BCL: Bayesian In-Context Learning Framework for Information Extraction

The article introduces BCL (Bayesian In-Context Learning Framework), a novel optimization framework for information extraction tasks that employs particle filtering with Bayesian updates to enhance label representations. BCL's systematic approach includes four key steps: initialization, observation, weight update, and resampling, allowing it to generalize across both sequence labeling and relation classification tasks. The framework shows significant and consistent performance improvements over current methods, highlighting its potential for practitioners working with large language models in information extraction.