Date of Publication

12-16-2019

Document Type

Master's Thesis

Degree Name

Master of Science in Computer Science

Subject Categories

Computer Sciences

College

College of Computer Studies

Department/Unit

Computer Science

Thesis Adviser

Ethel Chua Joy Ong

Defense Panel Chair

Charibeth Cheng

Defense Panel Member

Nathalie Rose Lim-Cheng
Ethel Chua Joy Ong

Abstract/Summary

Abstract The area of story content generation has been widely explored in the field of natural language processing. Previously, analogy-based methodologies have been used to provide an approach to this task. However, with the improvement of tech-nology, more and more research have been tapping into recurrent neuralnetworks - specifically, long short-term memory networks (LSTM) to accomplish this task. More specifically, they train the LStM models to learn the story, either through sequences of scenes in a story, or more commonly through sequences of action events, and allow them to predict a subsequent event based on this input. While the approach proves to be more complex, general findings from these researches an inability to provide consistently decent respons. In these researches, the rec-curring problem is attributed mainly to the poor quality of the training dataset. This research takes this opportunity and provides an alternative method of ex-traction and encoding event sequences in stories. The performance analysis of the event extraction system over 8 stories yielded the system an F1 score of 82%. The effectiveness of the encoding was evaluated by utilizing the encoded events extracted from a set of children’s stories in training an LSTM network. Results show the system’s ability to generate a decent response for more than half the time, with its ability limited by the current size of the dataset.

Abstract Format

html

Language

English

Format

Electronic

Accession Number

CDTG007957

Keywords

Natural language generation (Computer science); Parsing (Computer grammar); Computational linguistics

Upload Full Text

wf_yes

Embargo Period

12-5-2022

Download

Share

COinS