Predicting QoS for Web Service Recommendations Based on Reputation and Location Clustering with Collaborative Filtering

Abstract

Nowadays, numerous web services with equivalent functionality have become available on the Internet. The Quality of Service (QoS) for web services seems to have an essential role when it comes to selecting the best web services. However, evaluating the user-side efficiently in terms of quality of web services has become a critical research topic. Predicting the QoS values of web services and the credibility of the values published by different users are major challenges in this area. A commonly used technique to predict QoS values of web services is collaborative filtering (CF). To address these critical challenges, a personalized QoS predicting technique is proposed for web services which depends on the reputation and location-based CF approach. Firstly, a set of untrusted users is identified through the Dirichlet probability distribution on the basis of the user's reputation, followed by processing the unreliable data contributed by untrusted users. Secondly, the users are clustered based on their geographic information to improve the neighborhood similarity computation. Finally, the similarity weights of neighboring users are used to predict unknown QoS values in each cluster. It has been observed that the proposed model realized a more favorable performance in terms of accuracy and efficiency as compared to other existing approaches. According to the matrix densities from 10% to 90%, the measures of MAE and RMSE for the response time attribute range from 0.47 to 0.30 and from 1.26 to 0.95, respectively, and the measures of MAE and RMSE for the throughput attribute range from 15.64 to 7.58 and from 50.50 to 34.15, respectively.