Event Structure Detection and Representation Using Deep Learning

Abstract

As the amount of unstructured text generated every day grows, so does the need to process them automatically. The goal of this thesis is to develop models using deep learning that can automatically detect and generate event structure representations and to investigate their effectiveness as sentence semantic representations. We provide an overview of deep learning for NLP in Chapter 2 as background for subsequent chapters. We define in Chapter 3 the properties of an event structure and present the state-of-the-art approaches for both general and biomedical domains. We then introduce our proposed approach on biomedical event detection. In Chapter 4, we discuss the experiments settings in which we evaluate our neural models and present the results and analyses. We then present in Chapter 5 a review of the approaches in sentence representation in both general and biomedical domains. We discuss in Chapter 6 the results of our preliminary experiments on a state-of-the-art approach in the general domain and examine its limitations. Finally, we investigate the effectiveness of event representations in a biomedical semantic similarity task in Chapter 7.

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