Ontology-based Concept Weighting for Effective Information Retrieval

Abstract

Current keyword-based search, which is used actively in Web search, is easy and efficient to use but has a number of shortcomings in terms of performance. In order to overcome such shortcomings, research is being conducted on semantic-based search by various organizations. Recently emerged ontology-based semantic search is another search method for overcoming the limitations of keyword-based search. This study is about semantic search using ontology and particularly about efficient subject search. The ordinary search method that We use commonly also searches for documents containing contents intended by the user, but subject search finds documents that represent best the subject that the user is interested in. In order to prove the efficiency of ontology-based semantic search, this study first built a prototype system and experimented on the system, and found that well-defined ontology is required for efficient semantic search. Another finding of this study is that semantic search needs a method for the effective use of ontology, which is an organized system for knowledge delivery. Ontology, the core of semantic Web, is created by various institutions on diverse subjects. However, it is a highly time-consuming and labor-intensive work to establish an ontology manually. Accordingly, We need to integrate established ontologies and expand them. This study proposed an ontology matching method using DTW for the expansion of ontology. In addition, I conducted an experiment to compare performance using the Benchmark dataset of OAEI in order to prove the efficiency of the proposed DTW-based ontology matching method. The result of the experiment showed that the ontology matching method proposed in this study improved the accuracy of ontologies by over 5% compared to existing systems. The most important part of this study is effective information search and, in particular, subject search of documents using ontology. Subject-based document search is applicable to Web search but it can be applied most effectively in searching learning contents as in e-learning. This study proposed a method for applying weight to the vector elements of VMS using ontology for effective search while utilizing the conventional VSM-based search technique. Using the proposed method, I can create document vectors representing well the subjects of documents and use the vectors in existing search methods. In order to prove the efficiency of search method proposed in this study, I conducted an experiment using 852 mathematical contents of the 8th grade in middle school and TREC WT10G data. The experiment using mathematical contents showed that the method proposed in this study was superior to the existing TF-IDF method for various types of queries. In the experiment using TREC data, the proposed method showed performance higher than or equal to that of existing IR and semantic methods. These results suggest that the subject search method invented in this study can be extended for ordinary web search.