IMPROVED RANK AGGREGATION

Abstract

Multiple rankings of same object based on multiple criteria poses a problem of choice to rank that object at a position closest to all the rankings. Choosing a ranking for a list of such objects ranked previously is called rank aggregation. The aggregated ranking is analysed by computing the Kendall tau distance or Spearman Footrule distance. The ranking chosen by minimizing Spearman Footrule distance, is NP-Hard even if number of ranking lists greater than four for partial lists. In context of World Wide Web, the results generated by the multiple search engines may be collected together and an aggregated result can be produced using rank aggregation methods or in simple words the meta-search engine comes into existence. However, the use of existing search engines reveal that none of them have been effective in generating the results up to the quality and reliability desired by the end user, the reason being many. The application of fuzzy logic techniques to minimize the aggregated Spearman Footrule distance obtained from rank aggregation when applied in meta-searching have been studied at length, and the improvements in the existing Shimura’s technique, viz. Enhanced Shimurasquare and Enhanced Shimura-sqrt have been proposed to achieve the extent that is better to Borda’s Method, a benchmark for comparison of aggregated rankings as well the other common techniques. In addition, the improvements in existing heuristics based positional methods have also been proposed. A new heuristics have also been put forward for future work. A series of experiments on real and benchmark data have been conducted to validate the proposed improvements. The platform used for experimentation was matlab framework. A comparative analysis of the previous and the proposed rank aggregation methods on six search engines and twenty one search queries have been presented. The different aggregation methods have been compared on the basis of aggregate spearman footrule distances. The computation time have also been measured and reported. It has been revealed by the experiments that our proposed soft computing method, Enhanced Shimura Square performs best in terms of effective performance and computational efficiency.