Research On Web Service Recommendation Based On Collaborative Filtering And Characteristic Of User And Service

In recent years,with the development of Internet technology,web service has caused serious concern.How to select and recommend web service to users is an important problem in the field of service computing. With the increasing growth of users and web service,it is becoming more difficult to recommend service to users.The main research and contributions of this thesis is to recommend web service in a large number of similar service by user preference and the functional and non-functional attributes of service.At present,the recommendation techniques include content-based recommendation, association rule-based recommendation,collaborative filtering recommendation and so on. Collaborative filtering is one of most successful technologies applied in recommender system in multiple domains.Collaborative filtering appears in B2C e-commerce,and has made a great progress.Business company can recommend products such as music,books to users by user interests.Although collaborative filtering recommendation method have some advantages in the field of web service,it still has some problem.The main problems are new user problem,new item problem and sparsity.With the increasing growth of web service,these problems are worth to pay more attention to.This paper is focus on these problems.In this paper, firstly,we propose a collaborative filtering method based on user preference about Quality-of-Service of web service.In the original method,the similarity between users relies only on the information of user rating score,but it is not very accurate.Eg.two users choose a same web service,one prefer the response time of web service,another prefer the security. The similarity of users in our method relies on the user rating score and user preference.Our method has a better performance in experiments.Secondly,we propose a method based the professional degree of users to address new user problem and new item problem.The professional degree is computing by the involving degree and accuracy of rating score of users.The recommendation information for new user is computed by both similar users and professional users.Because professional users can give more accurate score of new items,new item problem can be alleviated. We made experiments to analyze and summary these problems.Finally,for the sparsity problem,we propose a user similarity transfer method and analyze the reasons for lack of similar users.We create a trust model between two similar users based on the common rating set for user similarity transfer.Experiments show that the method can alleviate sparsity problem.