Using anchor text refined by page importance to improve web retrieval

As important part of web page contents, anchor texts have been widely used and proved to be useful in web information retrieval systems, especially for the navigational queries. But previous work only focused on how to determine the importance of anchor texts for a given destination page. From global perspective, web link-structure is very useful to determine page importance, and this paper proposes a method that combines page importance and relevance between anchor texts and their destination pages together to build new anchor-based retrieval models. Experimental results show that the combined models are better than the models which only consider the relevance between anchor texts and their destination pages.