High-speed Detection of Ontological Knowledge and Bi-directional Lexico-Syntactic Patterns from the Web

Abstract

We propose a high-speed method of detecting ontological knowledge from the Web. Ontological knowledge in this paper means a term related to a given term. For example, hypernyms and hyponyms are basic related terms that are treated in dictionaries. Synonyms and coordinate terms are also well-defined related terms. Topic terms and description terms represent topics of the given term and they are vaguely defined. There are other related terms such as abbreviations and nicknames. The proposed method can be used for detecting many kinds of related terms. It extracts related terms from text resources only from Web search results, which consist of the titles, snippets, and URLs of Web pages. We use two different kinds of lexico-syntactic patterns to extract related terms from the search results, and these are called bi-directional lexico-syntactic patterns. The proposed method can be applied to both languages where words are separated by a space such as English and Korean and ones where words are not separated by a space such as Japanese and Chinese. The proposed method does not need any advanced natural language processing such as morphological analysis or syntactic parsing. It works relatively fast and has excellent precision. We also propose a method of automatically discovering superior bi-directional lexico-syntactic patterns using Web search engines because it is sometimes difficult to find appropriate patterns to detect related terms in a certain relationship.