Personalizing XML Full Text Search in PIMENTO

XML search is increasing in popularity as more and larger XML repositories are becoming available. The accuracy of XML search varies across different systems and a lot of effort is put into designing scoring functions tailored to specific datasets. For example, the INEX effort [8] aims at improving the search relevance of IEEE XML data collections. To the best of our knowledge none of the existing XML search solutions incorporates user information to determine relevant query answers. In PIMENTO we argue that there is no scoring function that can meet all user-related information and advocate the idea of incorporating user profiles into XML search in order to customize query answers and improve search quality. Personalization is used in a variety of applications: in telecommunications it is used to direct user calls based on the caller context, in Web search the ranking of query answers may be modified using the user’s navigational and search patterns. In the relational database context, query personalization has been studied extensively [6,12] and shown to be effective in practice. In PIMENTO a user profile is composed of two kinds of preference rules: scoping rules and ordering rules. Scoping Rules are used to expand or restrict the original query result. Ordering Rules are combined with query scoring to customize the ranking of query answers, hence overriding the ranking strategy of the query engine. Query personalization in PIMENTO is defined as the process of rewriting a user query using scoping rules and ranking query answers using ordering rules. Enforcing scoping rules is not straightforward: there can be a large number of rewritings of the user query when scoping rules are considered. To enforce efficiently scoping rules we take into account existing query relaxation work [2,15]. A key contribution of our approach is that scoping rules can be incorporated into a single query plan without requiring actual query rewriting. Ultimately, the user is only interested in the top several answers. Consequently, understanding how to combine user profiles with topk processing is a key aspect of efficient query personalization. A core contribution of PIMENTO