An Effective Fusion Technique Based on Signal to Noise Ratio

Abstract

Information retrieval (IR) is the process of retrieving information that is relevant to the users' needs. Data fusion in IR usually combines the various retrieval schemes (strategies) to enhance the performance of the IR systems. Though, the fusion operation successfully combines the merits of all participating member strategies, some of them may affect it's performance. Hence, it is essential to eliminate role of the worst performing members. This paper focuses on a method, which effectively combines the best functioning schemes. In the proposed approach, assignment of the final relevance score to the documents is based on the principle of signal to noise ratio. An algorithm is used to separate the best performing (information bearings) schemes from the worst (noise) ones. The presented method is tested in three test document collections namely ADI, MED, and CISI and it has been identified that, the proposed approach results in significant improvement over the existing Comb-functions for combining scores