시맨틱 웹 환경에서 온톨로지 기반의 지능형 상품 검색 시스템 설계 및 구현

Abstract

With development of information technologies, advances in Web technologies produced a large amount of information, made network such as Internet a critical element inevitable for daily life and brought improvement in quality of information like free information exchange over Internet. Conventional Web has put emphasis on convenience to facilitate use of HTML so that everyone can easily access Hypertext information, resulting in a drastic increase in the volume of information. As a result, users are now confronted with challenges in searching desired information. Typical approaches that have been used till now to search information are usually based on keywords using frequency of words or word information. Therefore, since many documents irrelevant to user's query are included in the searched results, user has to search the represented results further to obtain a desired result. To resolve this problem, interests have been focused on a semantic Web in which a search engine itself searches the information required for user through conceptualization, and ontology is at the center of it. Ontology can be regarded as a kind of dictionary that defines relations between words. It explicitly expresses the conceptual architecture in a specific domain to enable modeling of concepts closer to a real world around domain as a platform for building a search system. To build an ontology, we collected product related documents of each manufacturer, analyzed the documents and built a product domain ontology in OWL based on classification and features frequently found in those documents. In addition, synonym list, deduction and natural language processing functions were added to the ontology to allow expansion and deduction in searching. A database was built with the product data of about 40,000 items of Mart A operating on/off-line discount stores nationwide, and the user interface was developed using JSP and PowerBuilder9.0. We applied two techniques for ontology-based product search. One is searching by class in which users directly select a class of desired product from the ontology classes. The other is the natural language search method in which the keyword entered by user is searched through natural language processing. From the experiments on the two ontology-based searching methods, we could see that, unlike the conventional keyword search methods, it is possible to implement precise searching by giving meaning to the keyword itself using the concepts of classes and attributes, synonym list and deduction function defined in the ontology. Moreover, it is also feasible to search products with similar characters or those unexpected by user by semantic expansion of the searched products. By building an ontology-based intelligent product search system with this search architecture, we could obtain searching results that is impossible with the existing keyword-based searching methods and implement more precise searching.