Clustering Search Applied to Rank Aggregation

Several practical applications require joining various rankings into a consensus ranking. These applications include gathering the results of multiple queries in information retrieval, deciding the result of a poll involving multiple judges and joining the outputs from ranking classification algorithms. Finding the ranking that best represents a set of rankings is a NP-hard problem, but a good solution can be found by using met heuristics. In this paper, we investigate the use of Clustering Search (CS) algorithm allied to Simulated Annealing (SA) for solving the rank aggregation problem. CS will clusters the solutions found by SA in order to find promising regions in the search space, that can be further exploited by a local search. Experimental results on benchmark data sets show the potential of this approach to find a consensus ranking, achieving similar or better solutions than those found by other popular rank aggregation strategies.