Optimized Collaborative Filtering Prediction Methods Of QoS And Web Service Recommendation

The open Internet environment has promoted the continuous innovation and development of Web services.Faced with many Web services with nearly the same functions,people have difficulty in choosing them,and then the uneven Quality of Service(QoS)has become the entry point for the precise choice of Web services.Therefore,the recommendation of Web services based on QoS data shows important research value.In view of the problem of sparse QoS data and personalized difference of users in the collaborative recommendation of Web services,the following work is completed.Firstly,the advantages and disadvantages of different similarity calculation methods and the different data environments based on users' collaborative filtering algorithm are analyzed and summarized.A similarity improvement scheme is introduced which combines the Pearson correlation coefficient and Tanimoto Coefficient to synthesize the advantages of both and common call number threshold is also introduced to adjust the combination,which can compensate for the inadequacy of Pearson correlation coefficient in measuring the similarity of user structure and reduce the degree that the algorithm is inaccurate in the case of sparse data.Secondly,when selecting the nearest neighbor group,there is a problem that the traditional Top-K algorithm or threshold filtering method has not fully utilized the information provided by the existing user-service QoS matrix and lacks the adaptability and flexibility.Therefore,in order to make full use of the users' or services' similarity calculation results,a dynamic generation threshold method is proposed,and the generation of each threshold is dynamically adjusted according to the user or service,which help to find the candidate users with strong correlation with the target users who can participate in the QoS prediction process.Then,the optimized collaborative filtering algorithm is presented combining it and the improved similarity calculation method proposed in the previous section.Thirdly,according to the problem that the acceptable range of QoS data of different users is not fully considered in the traditional collaborative forecasting method,which causes the situation that the users' personalized needs cannot be satisfied,A QoS collaborative prediction method based on user preference scope is proposed,which can extract users' preference information from raw QoS data by using normalized rules for cost and benefit type attributes,then calculate similarity based on Euclidean method,and select similar nearest-neighbors by using the negative filtering of Top-k algorithm,and complete the prediction of missing QoS.Finally,in view of the problem that the existing service recommendation system has insufficient attention to service performance,and lacks of integration of existing resources,technologies and research steps of Web services.Therefore,a one-stop platform for Web services integrating Web services registration,service QoS collection,model prediction and service recommendation is designed and implemented,which covers the collection and performance monitoring of services,and applies collaborative filtering technology with better results to the recommendation selection of web services to improve users' satisfaction with the recommended results.