An Optimization Method for Proportionally Diversifying Search Results

Lin Wu, Yang Wang, John Shepherd, Xiang Zhao

Abstract: The problem of diversifying search results has attracted much attention, since diverse results can provide non-redundant information and cover multiple query-related topics. However, existing approaches typically assign equal importance to each topic. In this paper, we propose a novel method for diversification: proportionally diversifying search results. Specifically, we study the problem of returning a top-k ranked list where the number of candidates in each topic is proportional to the popularity degree of that topic with respect to the query. We obtain such a top-k proportionally diverse list by maximizing our proposed objective function and we prove that this is an NP-hard problem. We further propose a greedy heuristic to efficiently obtain a good approximate solution. To evaluate the effectiveness of our model, we also propose a novel metric based on the concept of proportionality. Extensive experimental evaluations over our proposed metric as well as standard measures demonstrate the effectiveness and efficiency of our method.