Research On Web Service Recommendation Method Combining Network Location

In the Web 2.0 era,service-oriented computing is widely used in academic and industry.As the number of Web services continually growing,it becomes difficult for users to select the high-quality Web services that meet users' needs while facing a large number of Web services.Therefore,how to recommend the superior Web service has become a hot issue in service computing.In the Web service recommendation system,Quality of Service(QoS)is usually used to measure the performance of Web services.Due to the limited invoking records,predicting the missing QoS value has been the key for a valid service recommendation.Collaborative filtering is a widely used method for QoS prediction.Among them,neighbor-based collaborative filtering methods and matrix factorization methods are most effective.Although these two methods have achieved successfully in QoS prediction,there are still some problems that need to be solved.On the basis of analyzing and summarizing existing researches,this paper improves the neighbor-based collaborative filtering methods and matrix factorization methods,and proposes two novel service recommendation methods:(1)We proposed a service recommendation method based on network location and random walk.In the similarity calculation phase,we utilize random walk model to mine the similarity between users and the similarity between services,and then combines the network location information to optimize the similarity calculation method for target users and target services.In the neighborhood screening phase,users and services are divided according to network location relationships,both relevant users and services with abnormal QoS call records are eliminated.Through the above improvements,three optimized QoS prediction methods are proposed,named as UL-RW,SL-RW and HL-RW.We utilize the proposed methods to achieve more effective Web service recommendation.(2)We proposed two kinds of service recommendation models based on matrix factorization.In this paper,two kinds of matrix factorization models,LFM(Latent Factor Model)and PMF(Probabilistic Matrix Factorization),are respectively improved.First of all,we take the invoking record and the network location of the user into cluster users by the hybrid attribute clustering method.Then,the baseline prediction model is established according to the clusteringresults.Moreover,because the country information of service is an important factor that affects the preference of users,we take the country of service as the implicit feedback information to get the user's preference through training implicit feedback factor analysis,so as to improve the accuracy of QoS prediction.Finally,the personalized reference prediction model and implicit feedback factor are integrated to LFM and build an improved recommendation model,called LA-LFM.In addition,since the PMF model is further improve by introducing the probability model based on the LFM.PMF alleviate the over-fitting problem to some extent.Therefore,similar as LA-LFM,LA-PMF is proposed by integrating the personalized reference prediction model and implicit feedback factor.In this paper,a large number of experiments on the two above algorithms are carried out on a real data set of Web services,and compared with the famous service recommendation algorithm.Experimental results show that the proposed method can accurately predict QoS in the situation of data sparsity,leading to high quality service recommendations.