A Link Prediction Algorithm that Solves the Data Sparsity Problem in Service QoS Prediction

Abstract

Quality-of-Service (QoS) is an important aspect of services computing, and QoS prediction based on collaborative filtering (CF) has been extensively researched. With the dramatic increase in the number of Web services and the consequent sparseness of the User-Service matrix, searching a sufficient number of similar neighbors has become a critical challenge in CF-based QoS prediction. Nevertheless, the topic remains improperly investigated. In this paper, we modify two existing link prediction algorithms, WAA (Weighted Adamic-Adar) and WRA (Weighted Resource Allocation), into a novel QoS prediction approach that considers the proximity of users' locations. Our modified algorithms search for implicit neighbors by link prediction and build similar networks to increase the number of neighbors. Both algorithms effectively increase the coverage of the prediction algorithm. Meanwhile, we optimize the implicit neighbor search by incorporating location factors. The ability of the improved algorithms to solve the data sparsity problem is validated in experiments on public real world datasets. The new algorithms outperform the existing IPCC, UPCC and WSRec algorithms.