An efficient method for the reverse top k search

Abstract

The reverse top-k search problem is a variant of the top-k search problem. Suppose that there are a set of objects O and a set of weight vectors W in the d-dimensional space. Then, given a query object, the reverse top-k(RTOPk) search is to find a subset W′ of W such that, when the query object is inserted into O(or already included in O), the query object is one of the top-k objects for each weight vector in W′. Many studies on the RTOPk search have been conducted in the last few years. However, the existing methods for the RTOPk search only focus on how to efficiently process evaluations of the weight vectors, which check whether the weight vectors are included in the result without efficient pruning methods. In this paper, we propose an efficient method for the RTOPk search, that overcomes drawbacks of existing methods. In contrast to existing methods, the proposed method prunes each unnecessary weight vector without traversing objects, and filters out each unnecessary object without considering the weight vectors. In addition, the proposed method evaluates weight vectors by preserving search contexts for multiple traversals to tree-based indexes of weight vectors and objects. The experimental results based on synthetic datasets and a real dataset show that the proposed method is significantly more efficient than existing methods.