Inverse square rank fusion for multimodal search

Abstract

Rank fusion is the task of combining multiple ranked document lists (ranks) into a single ranked list. It is a late fusion approach designed to improve the rankings produced by individual systems. Rank fusion techniques have been applied throughout multiple domains: e.g. combining results from multiple retrieval functions, or multimodal search where several feature spaces are common. In this paper, we present the Inverse Square Rank fusion method family, a set of novel fully unsupervised rank fusion methods based on quadratic decay and on logarithmic document frequency normalization. Our experiments created with standard Information Retrieval datasets (image and text fusion) and image datasets (image features fusion), show that ISR outperforms existing rank fusion algorithms. Thus, the proposed technique has comparable or better performance than existing state-of-the-art approaches, while maintaining a low computational complexity and avoiding the need for document scores or training data.

Rank level fusion in multibiometric systems

“…Performance of our proposed methods (as shown in Fig. 1) are compared with the following state-of-the-art methods: Highest Rank [4], Borda Count [4], Logistic Regression [4], Mixed Group Rank [10], Inverse Rank Position [18], Reciprocal Rank Fusion [20], Nonlinear Weighted [14] and Inverse Square Rank [21].…”

“…Other rank level fusion techniques are Bucklin Majority Voting [12], Bayesian approach [13], Nonlinear Weighted Rank [14], Markov Chain based fusion [15], Fuzzy Rank based fusion [16] and Neural Network based fusion methods [17]. Various other rank level fusion techniques proposed in the fields of image, document and information retrieval are: Inverse Rank Position [18], Leave Out [18], Reciprocal Rank Fusion [20] and Inverse Square Rank [21]. Recently, Zhang et al, [22] proposed a graph based fusion technique in which rank lists are modeled as graphs and multiple graphs are merged and re-ranked by link analysis of the fused graph.…”

Scalable Multimodal Search with Distributed Indexing by Sparse Hashing

“…Inverse Square Rank (ISR) [24] combines the inverse rank approach of RR and RRF with document frequency weighing:…”

“…After searching multiple indexes (one for each feature space), each returning a rank, one needs to merge the individual ranks into a single final rank. Rank aggregation techniques have been applied to solve multiple problems: combining results from multiple textual retrieval functions [10], combining the results of multimodal multifeature queries [24], federated search and expert search [20]. Our proposal is to use rank-based aggregation of multimodal ranks, as they have a low computational complexity.…”

Learning to reformulate long queries for clinical decision support

“…The shared task captured the interest of many research teams, and ran again the following year (Roberts, Simpson, Voorhees, & Hersh, ). Among the many approaches proposed, automatic query expansion techniques were found to be very effective for the task (e.g., Balaneshin Kordan, Kotov, & Xisto, ; Choi & Choi, ; Mourao, Martins, & Magalhaes, ). Some expansion techniques relied on medical ontologies; others were based on Pseudo Relevance Feedback (PRF), a method in information retrieval to expand queries by selecting m number of good terms from k number of top‐ranked documents.…”