Research On Web Service Recommendation Algorithms Based On Random Walk

In recent years, with the extensive application of service-oriented computing and architecture, more and more Web services are deployed on the Internet. Many Web services are with the same or similar functionality, but their quality of service(QoS) is quite different. Faced with a large number of candidate Web services, users have to spend a lot of time and efforts to identify the service with satisfactory functionality and high quality. Therefore, recommending appropriate high-quality Web services to users has attracted increasing attention from researchers.Conventional Web service recommender systems usually employed collaborative filtering methods, which are based on the services' history invocation records, to predict quality of services for the target users, so that high quality service could be discovered and recommended. In reality, with the rapid growth in the number of Web services and service users, a user usually invokes a few web services. As a result, the service invocation and QoS data could be extremely sparse, thus decreasing the accuracy of the collaborative filtering-based service recommendation methods in predicting service QoS. In addition, traditional Web service recommendation methods failed to obtain the confidence of QoS prediction and service recommendation. QoS of Web services is usually affected by the user and service location. Based on this observation, a number of research work considers exploiting user or service location information to predict the service QoS better. To some extent, these work can relief the problem casued by the sparse QoS data. However, in the case of the sparse distribution of the user and the service location, the QoS prediction accuracy of the location-based Web services recommendation method will be limited. In order to combine the advantages of collaborative filtering and location based mehods, we proposed a Web service recommendation method based on random walk model. The main contributions of this paper are as follows:(1) A random walk model for Web service recommendation method is proposed. The method combines users' location information and collaborative filtering techniques to predict QoS of Web services. The method consists of two stages. In the first stage, the users' location information and the similarities between users are exploited to build a user network. The second stage performs a random walk algorithm in the network, each walk as the goal with target user and service is computing a QoS prediction value, until the termination condition is satisfied. Finally, we aggregate the predicted QoS values to calculate the final result. Our method can effectively address the data sparsity and cold-start issues, improving the coverage rate of and the accuracy of service recommendation. And the method allows us to define and measure the confidence of the recommendation results. In order to evaluate the performance of the proposed method, a series of comprehensive experiments were conducted on a real Web service data set, and a comparison was made with the existing collaborative filtering methods. The results show the effectiveness of the proposed method.(2) By extending the previous random walk-based Web service recommendation method, a Web service QoS interval prediction method is proposed. In reality, the QoS of a Web service is often uncertain and dynamic, and it is easy to be affected by user location, network state and service workload. Predicting Web Service QoS by single point value cannot fully reflect the real situation, and its credibility may be unsatifactory. Therefore using interval values to estimate web services' QoS is desirable. In this paper, we propose a QoS interval prediction method for Web services, which can be used to find the maximum confidence interval of QoS as the prediction result under the condition of the given QoS interval length. The experimental results on real data valiated the performance of the proposed QoS interval prediction method.