Document Level Closed Information Extraction

Document-level closed information extraction (DocIE) is the task of extracting fact triples consistent with predefined knowledge base entities and relations from unstructured text. DocIE involves subtasks such as mention detection, entity type recognition, named entity recognition, entity disambiguation, entity linking, and coreference resolution, aiming to capture long-distance dependencies and extract relationships between entities that are far apart in the document. This task is of great value for applications such as building knowledge graphs, question answering systems, knowledge discovery, and text summarization.