In this paper, we propose a filtering tool for WWW search engines. Because ordinary search engines select candidate pages based on appearance frequency of keywords, the candidate pages may contain every keyword. However, there are some cases that each keyword appears in individual context and they have no semantic relation. Such pages tend to be useless. We try to improve precision of a search engine by checking whether input keywords have a semantic relation in the candidate pages or not. Semantic relations between keywords are represented by using a structure, such as sentence, table, itemized list and so on. In this paper we deal with sentence and table as a structure to represent semantic relations. We propose a method to detect semantic relations between input keywords represented in the structure of sentence and table. We also show an experimental evaluation of our method.
展开▼