Knowledge Graph Creation from Structure Knowledge

Open Access
- Author:
- Wang, Shuting
- Graduate Program:
- Computer Science and Engineering
- Degree:
- Doctor of Philosophy
- Document Type:
- Dissertation
- Date of Defense:
- June 05, 2017
- Committee Members:
- Clyde Lee Giles, Dissertation Advisor/Co-Advisor
Clyde Lee Giles, Committee Chair/Co-Chair
Jesse Louis Barlow, Committee Member
George Kesidis, Committee Member
Bruce A Desmarais Jr., Outside Member - Keywords:
- Knowledge Graph
Web Knowledge Base
Education Data Mining - Abstract:
- Knowledge graphs, which organize and structure knowledge by linked concepts, have been widely used as teaching and learning tools for science education tasks. Typically, they are manually created by domain experts to serve as ``ontology'' or ``knowledge base'' for different purposes. However, with the tremendous growth of massive online educational data, automating the creation of concept maps becomes necessary. The task is challenging as it requires not only extracting domain concepts from educational content but also identifying the semantic relations among them. This thesis proposes to perform concept map extraction from high-quality academic resources, such as textbooks and research papers, inside of which more meaningful structures and semantics can be leveraged. Moreover, we leverage other types of structure knowledge such as web knowledge bases to help create the concept map. With rich semantic information and a large-scale hyperlink network, it has been widely used in learning and education. We first present a work on concept hierarchy extraction using rich structure in textbooks. However, this work did not explicitly identify concept relations embedded in the textbooks. To resolve this, we perform joint optimization for concept extraction and prerequisite relations identification using rich book structures. In the last work, instead of only being interested in the prerequisites, we aim to extract a concept map with multiple types of relationships, such as ``is-a' and ``has-part''. Experimental results show that our proposed model achieves the state-of-the-art concept map extraction result on concept maps, manually created from six textbooks. Lastly, this thesis performs some pilot studies on using the automatically extracted concept map for educational purposes. Two systems are introduced in this thesis. One is BBookx - an automatic textbook creation system~\footnote{https://bbookexp.psu.edu/}; Another is an automatic assessment system using prerequisite concept maps.