A Personalized Federated Framework for Document-Level Biomedical Relation Extraction

Document-Ievel relation extraction (RE), a critical task in natural language processing, involves classifying relationships between entities mentioned in documents. Traditional document-level RE methods often rely on centralized datasets, which necessitates the collection of documents from various sources and their storage on a single machine. However, this brings up issues regarding privacy protection, and also limit the diversity of data. In this paper, we solve the document-level RE problem in federated learning scenario and propose a personalized federated learning framework named PF-BRE, specifically designed for biomedical RE. This framework enables multiple clients to collaboratively train a robust model without sharing the raw data. Additionally, to simulate the varying data characteristics of different clients in real-world application scenarios, we process the existing biomedical RE datasets to create distinct clients for personalized federated training. Exper-iments show that, compared to centralized training, the PF - BRE framework not only achieves impressive performance but also protects the data privacy of each client.