NEURAL NAMED ENTITY RECOGNITION AND TEMPORAL RELATION EXTRACTION

Abstract

Automatically identifying information of interest from texts is one of the most difficult challenges. One crucial step towards information extraction is named entity recognition, where many entities are embedded in other entities (i.e., nested entities). Nested entities contain rich fine-grained information, which is essential in understanding texts. However, most work ignored nested entity recognition though they are common in many domains. In addition to the semantic information expressed in named entities, temporal information conveyed by named entities is another important dimension in understanding texts. Temporally classifying the relations (e.g., before) between entities is known as temporal relation extraction, which is required in many tasks such as text summarisation. The thesis is the first comprehensive research focusing on nested entity recognition for information extraction using neural network methods. In this research, we describe our work on (1) neural nested entity recognition; (2) evaluation on different domains of corpora; (3) task-specific evaluation including (a) neuroscience entity extraction; (b) screening reference documents; (c) extraction of medication and adverse drug information; (d) and extraction of chronic obstructive pulmonary disease phenotypes. In addition to nested entity recognition, we further investigate neural temporal relation extraction, which focuses on the extraction of both intra-sentence and inter-sentence temporal relations.

Date of Award	1 Aug 2020
Original language	English
Awarding Institution	The University of Manchester
Supervisor	Sophia Ananiadou (Supervisor) & Riza Theresa Batista-Navarro (Supervisor)