A Research On Approaches To QoS Prediction Of Web Service Based On Matrix Factorization

Web services have become an interoperable technology to connect amongheterogeneous applications on the Internet, which are published by differentcompanies through many standard protocols. With the rapid development of Internetand web services technology, the needs of the enterprises change constantly, so thenumber of web services has become more and more. In the face of many candidateweb services, which have similar or equivalent functionality, it has a critical value toselect the optimal one to build web application from them. Due to the instability ofInternet environment and the discrepancy of service user’s network infrastructure, theaffirmatory QoS by the provider is always various for different service users. And it isinfeasible to acquire the personalized QoS record through invoking all web servicesby service users. Therefore, it is a crucial challenge to acquire the QoS valuesaccurately for select appropriate services to build web application.This paper discusses the difficult of web services QoS prediction problem underthe condition of historical data sparseness and research status of QoS predictionquestion is expatiated. We propose two models and combine the advantages of twoapproaches through model integration to improve the final prediction accuracy.Firstly, this paper briefly introduces the knowledge of web services QoSprediction problem, including web services, QoS, many kinds of recommendationtechniques and similarity algorithm. Secondly, we take the baseline estimates modeland similar neighbors into consideration and extend the PMF model withregularization of similar neighbors. Thirdly, we extend the baseline estimates modeland further a more abundant model was proposed, which combines contextinformation through mapping them to the latent feature space. Further more,considering the advantages of two approaches, we propose a fusion of above twomodels. Finally, we prove the effectiveness of our model through experimentalverification, especially under the condition of data sparseness.