PhD 2022F: Liyan Xu

Document-Level Information Extraction w/ Global Inference and Challenging Domains

Liyan Xu

Date: 2022-10-21 / 4:00 ~ 5:00 PM
Location: MSC E306 (


In this thesis proposal, we address various document-level extraction tasks, including end-to-end coreference resolution and relation extraction, and specifically focus on the aspect of global inference. Our motivation is based on the fact that different from sentence-level tasks, document-level extraction naturally yields a global structure consisting of entities from the entire documents and their dependencies/interactions among them, which brings opportunities on how to leverage these global interactions and context to further improve the extraction quality and performance. In our proposal, we discuss the global inference from three perspectives: (1) higher-order inference on coreference resolution; (2) task interactions on joint entity and relation extraction; (3) generation-based joint extraction. In particular, the first two are presented as our finished work, and the last one is our proposed topic.