Approximately Solving Aggregate k-Nearest Neighbor Queries over Web Services

Abstract

In this paper, we propose a procedure for solving aggregate k-nearest neighbor (k-ANN) queries, which is effective in building location-based services (LBSs) by mashing up several Web services. In order to get the exact results of a k-ANN query, the Cartesian product has to be computed toward data of information sources, each of which is located at a different Web site. However, this requires a large amount of communication, because the entire data of at least one information source are sent to the other site. To reduce the cost, representative query point (RQP) is introduced to represent a set of query points and a k-nearest neighbor (k-NN) query with RQP as a key is requested to disseminate information of data objects. Although it computes an approximate result, the experimental results with synthetic data on precision and recall length ratio show that a k-NN query result with the minimal point of sum distance as RQP is allowable for a sum k-NN query result and a k-NN query result with the minimal point of max distance as RQP is allowable for a max k-NN query result regarding some combinations of k and the number of query points.