Refining Aggregation Functions for Improving Document Ranking in Information Retrieval
Résumé
Classical information retrieval (IR) methods use the sum for aggregating term weights. In some cases, this may diminish the discriminating power between documents because some information is lost in this aggregation. To cope with this problem, the paper presents an approach for ranking documents in IR, based on a refined vector-based ordering technique taken from multiple criteria analysis methods. Different vector representations of the retrieval status values are considered and compared. Moreover, another refinement of the sum-based evaluation that controls if a term is worth adding or not (in order to avoid noise effect) is considered. The proposal is evaluated on a benchmark collection that allows us to compare the effectiveness of the approach with respect to a classical one. The proposed method provides some improvement of the precision w.r.t Mercure IR system.