Relation identification for reasoning

This thesis aims to implement relation identification between entities in one sentence, which is a basic project for further applications in nature language processing. Two pre-labelled corpuses including Sem-Eval 2010 task 8 and Tacred Relation Extraction Dataset are utilised here, and sentence...

Full description

Saved in:
Bibliographic Details
Main Author: Wang, Xuehan
Other Authors: Mao Kezhi
Format: Theses and Dissertations
Language:English
Published: 2019
Subjects:
Online Access:http://hdl.handle.net/10356/78682
Tags: Add Tag
No Tags, Be the first to tag this record!
Institution: Nanyang Technological University
Language: English
Description
Summary:This thesis aims to implement relation identification between entities in one sentence, which is a basic project for further applications in nature language processing. Two pre-labelled corpuses including Sem-Eval 2010 task 8 and Tacred Relation Extraction Dataset are utilised here, and sentence modelling is done by extracting sentence-level features incorporating word embedding and position embedding. The classification task is supported by deep learning algorithms including Convolutional Neural Network and Recurrent Neural Network with Long ShortTerm Memory cell. The Effectiveness of CNN, LSTM and two combinations of these two models were investigated aiming to achieve better performance in relation identification. The whole project is implemented on python and the results were shown that the combination of LSTM-CNN model delivered highest accuracy rate and F1 score among all four models.