A Research For Service Recommendation Based On Multi-level Relations

Nowadays,as the quick development of Internet and Cloud Computing,more and more services are deployed on the Internet.As a result,it prompts the researches of service computing.Among them,service composition is a technique to compose existing Web services and create new services,and mashup is one of its implementation.The technology can reuse existing services more quickly and easily,which respond to the personal needs of users.With the increasing number of services,it has been an urgent problem how to recommend appropriate service for combining.Since the composition of services tends to follow particular patterns,we can mine the existing compositions to extract multiple relationships for recommendation.Currently there are already some recommendation approaches for the completion of service compositio n,but they are more or less restricted to some problems.These problems are mainly in the following aspects:(1)Cold-start problems,unable to recommend new services or new service composition.(2)Lack of richness,the recommended information comes from single source.(3)Lack of integrity,do not consider the overall cooperation.To solve the above problems,we propose an approach to gather and mine existing service compositions in order to model the deep relationships between them and to design effective recommendation algorithms.Our approach leverages the multi-source informa t io n hidden in the provenance,to provide more accurate and complete solutions for creation of service composition.The main work of this paper is reflected in the following aspects:1.We propose a multi-level recommendation model which combines the frequent pattern mining method and topic model mining for step-by-step mashup creation,to solve the problem that the new service cannot be recommended.2.We propose a multi-source relationship mining methods to mine the knowledge base,and establish a unified model of multi-level relationship,providing recommendation for complete mashup creation.In this study,we further come up with package recommendat io n,which leverages the relationship between the recommended services,so as to enhance the integrity of recommendation.3.We expand the multi-level recommendation model by increasing input and output information match,as a consideration to the collaboration between services,and further enrich the model.Experiments on real datasets show the efficiency of our approaches.