A Personalized Retrieval Model based on Influence Diagrams

A key challenge in information retrieval is the use of contextual evidence within the ad-hoc retrieval. Our contribution is particularly based on the belief that contextual retrieval is a decision making problem. For this reason we propose to apply influence diagrams which are extension of Bayesian networks to such problems, in order to solve the hard problem of user based relevance estimation. The basic underlying idea is to substitute to the traditional relevance function which measures the degree of matching document-query, a function indexed by the user. In our approach, the user is profiled using his long-term interests. In order to validate our model, we propose furthermore a novel evaluation protocol suitable for the personalized retrieval task. The test collection is an expansion of the standard TREC test data with user’s profiles, obtained using a learning scenario of the user’s interests. The experimental results show that our model is promising.