Publication details

Rank Aggregation of Candidate Sets for Efficient Similarity Search



Year of publication 2014
Type Article in Proceedings
Conference 25th International Conference on Database and Expert Systems Applications (DEXA 2014 )
MU Faculty or unit

Faculty of Informatics

Field Electronics amd optoelectronics, electrotechnics
Keywords Similarity Search; Metric Space; Approximation; Scalability
Description Many current applications need to organize data with respect to mutual similarity between data objects. Generic similarity retrieval in large data collections is a tough task that has been drawing researchers’ attention for two decades. A typical general strategy to retrieve the most similar objects to a given example is to access and then refine a candidate set of objects; the overall search costs (and search time) then typically correlate with the candidate set size. We propose a generic approach that combines several independent indexes by aggregating their candidate sets in such a way that the resulting candidate set can be one or two orders of magnitude smaller (while keeping the answer quality). This achievement comes at the expense of higher computational costs of the ranking algorithm but experiments on two real-life and one artificial datasets indicate that the overall gain can be significant.
Related projects: