A decision making approach to relevance feedback in information retrieval: a model based on a soft consensus dynamics

Information retrieval (IR) can be regarded as a natural instance of multicriteria decision making (MCDM). Queries are formulated as selection criteria aggregated by means of appropriate operators. Retrieval is then performed as a MCDM process by evaluating the degrees of satisfaction of the criteria by each document, and then aggregating them. Another decisional instance in IR concerns the problem of improving retrieval performance by taking into account user indications on documents relevance. Relevance feedback mechanisms exploit user-system interaction in order to improve retrieval results by means of an iterative process of query refinement. In this process the main decisional issue is that of finding new concepts (terms) with which to expand-modify the initial query so that it better reflects the user's information needs. In this paper we introduce a relevance feedback mechanism based on a dynamical consensus model originally proposed in the framework of group decision making. In the relevance feedback context the consensual interaction highlights associations among the most significant terms in the relevant retrieved documents selected by the user. The resulting associative structure can then be used to expand the original query by including new terms which result strongly associated with those in the query