Fusion of Multimedia Document Intra-Modality Relevancies using Linear Combination Model

Abstract

Information retrieval systems that search within multimedia artifacts face inter-modality fusion problem. Fuzzy logic, sequential and linear combinational techniques are used for inter-modality fusion. We explore asymptotically that Fuzzy logic and sequential techniques have limitations. One major limitation is that they only address fusion of documents coming from different modalities, not document relevancies distributed in different modality relevancy spaces. Linear combinational techniques fuse document relevancies distributed within different relevancy spaces using inter-modality weights and calculate composite relevancies. Inter-modality weights can be calculated using several offline and online techniques. Offline techniques mostly use machine learning techniques and adjust weights before search process. Online techniques calculate inter-modality weights within search process and are satisfactory for general purpose information retrieval systems. We investigate asymptotically that linear combination technique for inter-modality fusion outperforms fuzzy logic techniques and workflows. We explore a variation of linear combination technique based on ratio of average arithmetic means of document relevancies. Our proposed technique smoothes the effect of inter-modality weights and provides a moderate mechanism of inter-modality fusion.