Injecting Knowledge Base Information into End-to-End Joint Entity and Relation Extraction and Coreference Resolution

We consider a joint information extraction (IE) model, solving named entity\nrecognition, coreference resolution and relation extraction jointly over the\nwhole document. In particular, we study how to inject information from a\nknowledge base (KB) in such IE model, based on unsupervised entity linking. The\nused KB entity representations are learned from either (i) hyperlinked text\ndocuments (Wikipedia), or (ii) a knowledge graph (Wikidata), and appear\ncomplementary in raising IE performance. Representations of corresponding\nentity linking (EL) candidates are added to text span representations of the\ninput document, and we experiment with (i) taking a weighted average of the EL\ncandidate representations based on their prior (in Wikipedia), and (ii) using\nan attention scheme over the EL candidate list. Results demonstrate an increase\nof up to 5% F1-score for the evaluated IE tasks on two datasets. Despite a\nstrong performance of the prior-based model, our quantitative and qualitative\nanalysis reveals the advantage of using the attention-based approach.\n