Context-based QoS Prediction Research Of Web Service Recommendation

With the development of IT techniques in various cross-platform and cross-industry of different business,Web services have been developed to meet users' personalized requirements in their daily life.However,with the rapid increase of Web services,users are gradually facing the challenge of service overload problem.How to select candidate services with optimal attributes and functions and composite these unit services has become an essential issue in current research of service computing.Personalized recommendation technology exploit user's historical information to create leatent preference,then the unknown candidate services that meet users' requirements would be recommended to solve the problem of information overload.When more perfectly capable Web services have been developed on the Internet,users turn to pay more attention on QoS(Quality-of-Service)such as response time,accessibility,and success rate.The user's intuitive perception is generally determined by the non-functional quality of Web service,which will have greate impact on the Web applications that are developed by these Web services.In addition,users have different non-functional requirements of Web services in different context.How to recommend Web services with satisfying non-functional performance attract more attention in current service computing research.Therefore,this thesis designs several dynamic QoS prediction models by introducing collaborative filtering technology and combining different contextual information such as context at server-side,geographic location,dynamic time series and trust relationship to recommend high-quality Web service for target users.The main contribution of the thesis is presented as follows:(1)To study the impact of server-side contextual information on QoS prediction of service invocation,a novel prediction algorithm WSFNIMF based on server-side context clustering is proposed in this thesis.We obtain the word vector which represents the context of Web service by analyzing the WSDL file at server-side.Based on the distance between the word vectors,a service clustering is used to find the neighbor set of target service.At the same time,the historical QoS similarity is also considered to get the Top-K neighbor set of target user.Then we can predict the unknown QoS by combining two neighbor sets with the matrix factorization model and improve the accuracy of QoS prediction.The experimental results show that our proposed WSFNIMF algorithm can effectively improve the accuracy of QoS prediction in large-scale data and generate a high-quality personalized Web service recommendation approach.(2)In order to deal with the unreliable information provided by the untrustworthy users in same location region,a location and reputation aware QoS prediction algorithm LRMF is proposed in this thesis.We analyze the historical QoS records to design an iterative user reputation computing algorithm and discover the location region based on users' location information to find these untrustworthy users in the same geographical region.The geographical region information and users' QoS reputation score are integrated into the matrix factorization model to generate our recommendation system.The experimental results show that our LRMF algorithm can effectively improve the accuracy of QoS prediction and avoid the unrealizable QoS history records provided by untrustworthy users.(3)We propose a dynamic time-aware QoS prediction method in Web service recommendation framework based on collaborative prediction algorithm.This method is proposed to predict the QoS value of the Web services in the future time intervals based on users' historical QoS records at different time slots.In order to meet the non-functional requirements of target user along with time,service invocation time series are considered as a dynamic factor in the collaborative filtering model.And users who have similar QoS invocation records with the target user at same time interval are regarded as neighborhood and considered into a matrix factorization model,which is called TMF.When the prediction results of missing QoS values are obtained at first step,a curve smoothing method is used according to the correlation between the predicted values at current time series and other values at the nearby time intervals.The experiments prove that the accuracy of the QoS prediction is improved when time series are considered in our framework.(4)Finally based on the trust relationship between users in social networks of recommendation system,we propose a novel collaborative service recommendation approach in this thesis.We first compute the trust relationship between users based on their latent correlation in the historical service invocation records.Based on the computed trust degree,the strong trust relationship and the weak trust relationship are redefined by setting certain threshold.By integrating the latent trust relationship between users and the matrix factorization method,a trust-aware QoS prediction algorithm LTMF is proposed to generate an efficient and reliable personalized Web service recommendation approach in this thesis.