Mirror, Mirror, on the Web, Which Is the Most Reputable Service of Them All? - A Domain-Aware and Reputation-Aware Method for Service Recommendation

With the wide adoption of service and cloud computing, nowadays we observe a rapidly increasing number of services and their compositions, resulting in a complex and evolving service ecosystem. Facing a huge number of services with similar functionalities, how to identify the core services in different domains and recommend the trustworthy ones for developers is an important issue for the promotion of the service ecosystem. In this paper, we present a heterogeneous network model, and then a unified reputation propagation URP framework is introduced to calculate the global reputation of entities in the ecosystem. Furthermore, the topic model based on Latent Dirichlet Allocation LDA is used to cluster the services into specific domains. Combining URP with the topic model, we re-rank services' reputations to distinguish the core services so as to recommend trustworthy domain-aware services. Experiments on ProgrammableWeb data show that, by fusing the heterogeneous network model and the topic model, we gain a 66.67% improvement on top20 precision and 20%~ 30% improvement on long tail top200~top500 precision. Furthermore, the reputation and domain-aware recommendation method gains a 118.54% improvement on top10 precision.