Research On The Tecnology Of QoS Data-driven Context-aware Web Service Collaborative Recommendation

As the Internet plus grows,Web service as an indispensable digital glue of software development,business cooperation and business model innovation,has become a main technology in supporting Web application interoperability,distributed application systems construction and communication and data exchange between programs.With the increasing number of Web services in the network,service recommendation driven by QoS data has become a hot research topic in the area of service computing.On one hand,service recommendation can help service users to find the reliable Web services that they are really interested in,thus improving users‘ experiences;on the other hand,it will promote service providers provide quality ensured Web services to service users,so as to achieve a win-win situation for both users and service providers.This paper relys on the machine learning theory and Web service technology,improves service recommendation effects and user experiences by analyzing QoS data,timing information and location information,and propose noval prediction models and recommendation algorithms.Firstly,in order to alleviate the prediction error caused by the QoS rang differences among users and services,Gauss normalization mechanism of QoS data is introduced,and a collaborative prediction method considering the variation range of QoS data is proposed.This method proposes QoS evaluation matrix,user mean value matrix and service mean value matrix to analyze the fact that there is a large difference among users‘ and services‘ QoS data.Gauss normalization theory is introduced to maping the QoS data to a unified interval and makes the neighbors have the same dimension in calculation,thereby reducing the prediction error caused by QoS data in different ranges.Experimeanal results and case study show that Gauss normalization mechanism can solve the influence of QoS variation range on the prediction results effectively,and improve the prediction accuracy significantly.Secondly,in order to predict the QoS in the next period of time with history data accurately,and to support the online dynamic evolution of Web services,a prediction method called QARSPre based on timing analysis is proposed.The method introduces QoS attribute tensor,matrix and sequence to model the temporal QoS data,and uses collaborative filtering method to calculate the missing items in QoS attributes matrix for constructing QoS time series.On this basis,a prediction algorithm based on the principle of ?larger weight on the nearer data but smaller weight on the farther? is desined to make prediction.Experimental results demonstrate that QARSPre can obtain good prediction performance in the time varying service recommendation scenario.Thirdly,in order to solve the data sparsity and cold start problem in service recommendation model,this paper analyzes the geographical location information of Web service users,and a service recommendation model ULMF based on explicit QoS and implicit user location information is proposed.This paper observes that ?the closer user location,the more similar of QoS? based on real-word dataset analysis.ULMF uses the latitude and longitude information to calculate user geographical distance and qualitatively analyze the relationship between user location and QoS.A two-way filtering strategy based on geographical distance and similarity is designed for neighbor selection.To make use of the wisdom of the nearest neighbors,selected geographical users are integrated into the matrix facoriztion model.Experimental results illustrate that ULMF significantly improves the prediction accuracy,especially in the cold start users‘ senarios.Then,in order to realize context-aware personalized Web services recommendation,this paper analyzes different geographical contexts of provider,autonomous system and country,a location aware Web service recommendation model called Geo MF is proposed.A hierarchical bottom up neighbor search algorithm is designed based on the structure of the geographical neighbor tree,which can select high quality neighbors by effectively using geographical contexts.In addition,the model can effectively solve the problem of data sparsity and cold start problem by modeling neighobrs from the point views of users and services.Experimental results show that geographical neighbors can provide significant information,and the incoperation of geographical neighbors‘ wisdom can make GeoMF receive high prediction accuracy in both warm-start and cold-start scenarios,and obtains better performance than existing Web service recommendation methods.Moreover,GeoMF have excellent convergence performance in different dataset densities.Finally,in order to meet the objective requirement of QoS data driven context aware Web service recommendation platform,this paper designs and implements a Web service registration,monitoring,predicting and recommendation platform prototype system.To accumulate data for QoS data driven Web service recommendation studies,this platform realizes the Web service online registration,long-term monitors of Web services key QoS attributes.Moreover,the palatform implementes a variety of Web service personalized recommendation algorithm proposed in this paper to predict the unknown Web service QoS for users and supports unified,efficient,safe and personalized Web service selection recommendation for software developers.In addition,this platform analyszes and judgments the quality changes of invoked Web service of the distributed application systems based on monitored QoS data,thereby providing a technical support for the online evolution of application systems.