Web Service Recommendation Method Based On The Network Location

Along with the arrival of the Web2.0era and rapid development of thetechnologies, more and more Web services with the same or similar function areprovided for service users in the network by different service providers, which havelead to much inconvenience for the users because they usually have to spend a lot oftime and energy to find the exact service that meet their own needs. Besides, even ifsome users find what they want, it is always not the optimal. Therefore, it has becomemore crucial to come up method which helps in the selection of optimal service froma large number of candidate services with similar functions.Quality of service (QoS) is used to describe and evaluate the non-functionalcharacteristics of Web service. In the QoS-based Web recommendation service, it isnecessary to predict the missing value. There has been some research on collaborativefiltering method with Web service recommended, but they rarely consider the locationaware information of users and Web services in the prediction of QoS property values.In fact, location information of the user or the Web service has a significant impact onQoS values, such as response time, throughput, failure rate. In addition, with the rapidincrease of Internet users and Web services, the users only call a very little part of theWeb services among the candidates. The current Web service recommendationalgorithm is lack of accuracy. Moreover, its performance and efficiency is not high inthe large-scale sparse data situations. In the view of problems including lowprediction accuracy of Web services QoS under the circumstance of the currentlarge-scale sparse data as well as the bad scalability and cold start, this article haveput forward two novel collaborative filtering algorithms based on location aware. Themain research and innovations are as follows:(1) A kind of collaborative web service QoS prediction with user networklocation-based regularization is put forward. With analysis on network locationawareness attribute of users, by adding a regularization term to improve the matrixfactorization model, the algorithm can predict the missing values of QoS for activeusers, and then sequence the candidate services which meet the needs of usersaccording to the QoS value. Eventually web service with the best performance can berecommended for active users. After a number of experiments are conducted, the experimental results show that, the prediction of the QoS missing value is better thanother existing recommendation algorithm in the prediction accuracy by adding userlocation information to improve matrix factorization model. At the same time, themass data sparsity and scalability problems is well solved as well as the the cold startproblem in collaborative filtering algorithm because the time complexity of thealgorithm shows a linear correlation of the data size.(2) A Collaborative Web Service QoS prediction Via Network Location-BasedNeighborhood Integrated Matrix Factorization. The algorithm factorizes and practicesmatrix of users and web service neighborhood model both based on locationawareness. Two models are obtained by associating it with matrix factorization. QoSvalues are predicted through the adjustable parameters by combining the results oftwo models finally. Then comes the sequence of the candidate services according tothe QoS value. What the experimental results tell us is that, the model proposed in thisarticle is not only of higher prediction accuracy than other methods, but also solve theproblems of large-scale sparse data and poor scalability with the traditionalcollaborative filtering algorithm because of linear correlation between the time-spacecomplexity of the algorithm and the data size.(3) The feasibility of the proposed recommendation algorithm is validatedthrough experiments by using the large QoS data of public release set of realenvironment. The data set contains1,974,675QoS records, which are obtained bycollecting the call information of339users distributed in30countries on5825Webservices among73countries. It has been the largest open QoS data released in realenvironment up to now.