Date of Award

Spring 5-19-2018

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Computational Analysis and Modeling

First Advisor

Ben Choi

Abstract

This dissertation describes a knowledge-based system to create abstractive summaries of documents by generalizing new concepts, detecting main topics and creating new sentences. The proposed system is built on the Cyc development platform that consists of the world’s largest knowledge base and one of the most powerful inference engines. The system is unsupervised and domain independent. Its domain knowledge is provided by the comprehensive ontology of common sense knowledge contained in the Cyc knowledge base. The system described in this dissertation generates coherent and topically related new sentences as a summary for a given document. It uses syntactic structure and semantic features of the given documents to fuse information. It makes use of the knowledge base as a source of domain knowledge. Furthermore, it uses the reasoning engine to generalize novel information.

The proposed system consists of three main parts: knowledge acquisition, knowledge discovery, and knowledge representation. Knowledge acquisition derives syntactic structure of each sentence in the document and maps words and their syntactic relationships into Cyc knowledge base. Knowledge discovery abstracts novel concepts, not explicitly mentioned in the document by exploring the ontology of mapped concepts and derives main topics described in the document by clustering the concepts. Knowledge representation creates new English sentences to summarize main concepts and their relationships. The syntactic structure of the newly created sentences is extended beyond simple subject-predicate-object triplets by incorporating adjective and adverb modifiers. This structure allows the system to create sentences that are more complex. The proposed system was implemented and tested. Test results show that the system is capable of creating new sentences that include abstracted concepts not mentioned in the original document and is capable of combining information from different parts of the document text to compose a summary.

Download

Share

COinS