Research On Web Service Recommendation Based On ProgrammableWeb

With the development of SOA(Service Oriented Architecture),an increasing num-ber of Web services have been published on the Internet,which leads to information overload.How to recommend suitable Web services for users becomes a research hotspot in service computing.Currently ProgrammableWeb has become the main in-termediary for service publishing and discovery.Thus in this paper we focus on how to recommend suitable Web services for users on ProgrammableWeb.The current Web service recommendation methods mainly focus on QoS(Quality of Service)or the functional information of services.These methods pay little attention to rich side information of Web services,such as topic information and compositional information.Furthermore,current service recommendation methods tend to focus only on the improvement of accuracy,while neglecting the recommendation diversity.This situation will lead to more long tail services generated on the service platform.To address the discussed issues,this paper has done the following work based on Pro-grammableWeb:1.To improve the recommendation accuracy,we propose an algorithm based on topic information and compositional information of services.The algorithm is based on the real-world data from ProgrammableWeb.The implicit user feedback is used to calculate the ratings,and the topic information and compositional information of service is integrated to the matrix factorization model to predict users' ratings.The experimental results show that the proposed algorithm outperforms the existing collaborative filtering algorithms on the recommendation accuracy.2.To improve the recommendation diversity,we propose a re-ranking algorithm that considers user bias.The algorithm considers the biases in user scoring habits and user interest distribution,and combines these two biases into the calculation of the threshold TR.Thus,the improved re-ranking model no longer improve the diversity according to the fixed threshold TR.Experiments show that compared with the clas-sical re-ranking algorithm,our methods can improve the diversity of Web service recommendation while maintaining acceptable levels of recommendation accuracy.