Modeling Task Interactions in Document-Level Joint Entity and Relation Extraction

Abstract

We target on the document-level relation extraction in an end-to-end setting, where the model needs to jointly perform mention extraction, coreference resolution (COREF) and relation extraction (RE) at once, and gets evaluated in an entity-centric way. Especially, we address the two-way interaction between COREF and RE that has not been the focus by previous work, and propose to introduce explicit interaction namely Graph Compatibility (GC) that is specifically designed to leverage task characteristics, bridging decisions of two tasks for direct task interference. Our experiments are conducted on DocRED and DWIE; in addition to GC, we implement and compare different multi-task settings commonly adopted in previous work, including pipeline, shared encoders, graph propagation, to examine the effectiveness of different interactions. The result shows that GC achieves the best performance by up to 2.3/5.1 F1 improvement over the baseline.

Towards Consistent Document-level Entity Linking: Joint Models for Entity Linking and Coreference Resolution

“…Thus, [93] introduce Span-BERT, a BERT model pre-trained directly on spans instead of tokens (as is the case of BERT) in text. This model has been successfully used as the backbone to achieve state-of-the-art results in numerous information extraction tasks [94][95][96]. We use span-based models in our work described in Chapters 2, 3 and 4 of this thesis.…”

Coreference Resolution Based on High-Dimensional Multi-Scale Information

“…This fundamental NLP task can benefit various applications, such as Information Extraction [3,4], Question Answering [5,6], Machine Translation [7,8], and Summarization [9,10], which are of great research value. Coref requires document-level encoding.…”