Web Service QoS Prediction under Sparse Data via Local Link Prediction

Client-side Quality-of-Service (QoS) evaluation of Web services is a critical factor in selecting the optimal Web service from a set of functionally equivalent service candidates. And collaborative filtering (CF) method becomes an important way for automatic QoS evaluation. Traditional CF approaches for this problem predict the QoS values by employing historical QoS information, but their performance may suffer from the sparsity of data such the increase of the failure rates and decrease of the accuracy. In this paper, we investigate the data sparsity problem in QoS value prediction. By constructing the user similarity weighted network, we use the modified local link prediction methods to find implicit neighbors to alleviate the lack of neighbors and low successful rate in predictions caused by the sparse data. We also consider the user location proximity and weak tie affection carefully in link prediction. The experiment results on the public dataset validate that via local link prediction, the accuracy and the successful rate both increase compared to the traditional user-based CF approach.