CHIP 2023 Task Overview: Complex Information and Relation Extraction of Drug-Related Materials

Drug labels or package insert are legal documents that include significant information and are highly valuable. However, it contains both structured and unstructured information, which is challenging to extract. We construct a drug package insert information extraction dataset, which consists of 1,000 electronic files with a total of 17,000 structured fields and 24,580 entity relationships annotated. It is used in the CHIP2023 “Complex Information and Relation Extraction of Drug-related Materials” evaluation competition, in order to promote the development of printed material recognition and entity relationship extraction technology. Participants need to recognize structured fields from dataset and extract entity relationship from specified fields. Finally, we provide a concise overview and outline of their methods and discuss the potential value of the dataset in further study.