Domain Level Personalization Technique

We propose a novel technique for web search personalization that exploits the clustering of the results of web searches. Our approach is based on an automatic characterization of the user search history through the collection of semantic domains and web sources chosen by the user. The semantic domains are the terms extracted directly from the clusters contents which describe the main topics covered by the involved documents. The web sources are the web roots of the urls of the documents. The idea is that a user submits a query to a general purpose search engine and then selects clusters or documents from the resulting list. In the first case we can assume that, for that particular query, the user is interested in the topics covered by the selected clusters. We use these information to construct a user profile by assigning the clusters semantic domains to the query submitted by the user. In the second case we keep trace of the relevance and reliability of web sources of the selected documents, by assigning them to the submitted query.