Ranking Web Pages Using Collective Knowledge

Indexing is a crucial technique for dealing with the massive amount of data present on the web. Indexing can be performed based on words or on phrases. Our approach aims to efficiently index web documents by employing a hybrid technique in which web documents are indexed in such a way that knowledge available in the Wikipedia and in meta-content is efficiently used. Our preliminary experiments on the TREC dataset have shown that our indexing scheme is a robust and efficient method for both indexing and for retrieving relevant web pages. We ranked term queries in different ways, depending if they were found in Wikipedia pages or not. This paper presents our preliminary algorithm and experiments for the ad-hoc and diversity tasks of the TREC 2011 Web track. We ran our system on the subset B (50 million web documents) from the ClueWeb09 dataset. Categories and Subject Description Web Information Retrieval: Content Analysis, Indexing, and Ranking