Towards both Local and Global Query Result Diversification

Abstract

Query result diversification is critical for improving users’ query satisfaction by making the top ranked results cover more different query semantics. The state-of-the-art works address the problem via bi-criteria (namely, relevance and dissimilarity) optimization. However, such works only consider how dissimilar the returned results are to each other, which is referred to “local diversity”. In contrast, some works consider how similar the not returned results are to the returned results, which is referred to “global diversity”, and however need a user defined threshold to predicate whether a result set is diverse. In this paper, we extend the traditional bi-criteria optimization problem to a tri-criteria problem that considers both local diversity and global diversity. For that, we formally define the metrics of global diversity and global-and-local diversity. Then, we prove the NP-hardness of the proposed problems, and propose two heuristic algorithms, greedy search and vertex substitution, and sophisticated optimization techniques to solve the problems efficiently. To evaluate our approach, we perform comprehensive experiments on three real datasets. The results demonstrate that our approach can indeed find more reasonably diversified results. Moreover, our greedy search algorithm can significantly reduce the time cost by leveraging the critical object, and then our vertex substitution algorithm can incrementally improve the objective value of results returned by greedy search with extra time cost.