Cross-Language Information Retrieval Based on Category Matching of Web Directories

With the popularity of the Internet, more and more languages are becoming to be used for Web documents. Accordingly, Cross-Language Information Retrieval (CLIR), a method to retrieve documents written in one or more languages using a query written in another language, has been actively studied. A variety of methods, including employing corpus statistics for translation of terms and disambiguation of translated terms, are studied and a certain results has been obtained. However, since corpus-based methods depend much on the domain of the content, such methods have a potential problem that the retrieval effectiveness might be poor for domains which do not match the content of the corpus. In this paper, we propose a method to employ a Web directory which has multiple language versions such as Yahoo! for CLIR of Web documents. Feature terms are extracted from Web documents in a category, and one or more correspondent categories are determined by comparing similarities of categories across languages. We intend to resolve ambiguities of dictionary translation and to improve the retrieval effectiveness by limiting the categories to be retrieved.