Finding Preferred Query Relaxations in Content-based Recommenders

In content-based recommender systems, product proposals are generated by exploiting deep knowledge about the items in the catalog. In many implementations of such systems, the users´ requirements are directly viewed as constraints that all items in the proposal must fulfil and determining the set of suitable products thus corresponds at least initially to constructing an adequate query to the catalog. In such approaches, however, the problem can easily arise that the catalog query fails because none of the items in the catalog fulfils all of the user´s constraints. One general way of dealing with such situations is to relax the catalog query by eliminating individual subqueries and to search for items that fulfil as many constraints as possible. Finding such `maximal succeeding subqueries´ (XSS), however, is not a trivial problem because not all of the potentially many XSSs for a failing query are equally suitable for the user, which means that determining one arbitrary XSS is not sufficient in realistic settings. In this paper we present a new technique for determining all maximal succeeding subqueries of a query in an efficient way which allows us to determine optimal or `preferred´ solutions within the limited time frames of Interactive recommendation sessions. By evaluating the individual subqueries independently in advance and combining these partial results, we can compute all XSSs in a way that no further costly catalog queries are required. The approach has been implemented in the knowledge-based Advisor Suite recommender system and has been successfully evaluated in several real-world recommender applications