I will present an intelligent web retrieval system that is able to rank web pages by using collective knowledge to enhance a standard vector space model. The index contains separate information about the frequency of the terms in Wikpedia articles, in home pages, and in other types of web pages, instead of using a generic term frequency for the whole text collection. We also filter out spam. We present results on the ClueWeb collection, for two sets of queries, for an adhoc retrieval task and for a diversity task (which aims at retrieving not only relevant information, but also information for different aspects of the queries). We obtained competitive results when comparing to other systems from the Web track at TREC 2011.