Web search using query type classification

Abstract

The massive and heterogeneous Web exacerbates IR problems and short user queries make them worse. Due to these difficulties, nowadays classic IR that focuses on content information is not enough to retrieve various types of answer documents. To compensate for the insufficiencies of content information, previous studies combined multiple types of evidence such as PageRank with content information. However, in some cases, the combination of multiple types of evidence degrades the retrieval performance of a search engine. Each type of evidence has designated queries and answer documents. The use of evidence for an inappropriate query degrades the retrieval performance. We have to use each type of evidence according to its properties. To do this, the analysis of a query is needed. In this work, we investigate the property of each type of information for a search engine according to a query type. In addition, we propose Web search and MctaSearch that exploit a query type. User queries can be classified into three types according to a user``s intention; a topic relevance task, a homepage finding task, and a service finding task. The intention of a topic relevance task, a homepage finding task, and a service finding task are informational need, navigational need, and transactional need, respectively. We investigate the properties of content, link, and URL information according to a query type. In addition, we propose and investigate service link information that uses the existence of a service hyperlink for a service finding task. Each type of information shows different effect in Web search according to a query type. In a homepage finding task, combining link and URL information with content information improves the retrieval performance of a search engine. However, in a topic relevance task and a service finding task, it degrades the retrieval performance. In addition, retrieval algorithms such as TFIDF and OKAPI, also show different effect in Web search. Each type ...