A relevance feedback model based on soft consensus dynamics

In this paper a model of relevance feedback for an information retrieval system (IRS) is proposed. Relevance feedback mechanisms aim at enhancing the effectiveness of IRSs through a process of query refinement. They provide users with an interactive and iterative procedure based on an analysis of the effective relevance of documents retrieved by a query. Adapting a recent proposal in the context of group decision theory, we define a dynamical model of relevance feedback based on the so-called soft consensus paradigm. Starting from a set of significant terms in the relevant documents selected by the user (among those retrieved by the query) the soft consensus dynamics generates a fuzzy associative pattern among the significant terms and thereby suggests a structured extension of the original term set in the query. Moreover, by grouping together terms referring to the same concept the associative pattern generated by the soft consensus scheme can be regarded as a fuzzy thesaurus for the relevant documents