Service Recommendation from the Evolution of Composition Patterns

Abstract

A service ecosystem, consisting of various kinds of services and mashups, evolves over time. Existing works on the evolution of service systems focus on either evaluating the impacts of services' changes on the usage of services and the stability of the whole ecosystem, or discovering co-occurrences between services, but fail to disclose any knowledge about the evolution of service composition patterns. Based on our previous work of SeCo-LDA, through scrutinizing the dependencies between different service co-occurrence topics, this paper reveals the latent service composition trends in a service ecosystem. We derive topic dependencies and describe it as a directed topic evolution graph, where four topic evolution patterns are identified. A novel methodology, named Dependency Compensated Service Co-occurrence LDA (DC-SeCo-LDA), is developed to calculate the directed dependencies between different topics, build the topic evolution graph. The evolution trend of service composition could be disclosed by the graph intuitively, and dependency compensation could be adopted to improve the performance when making service recommendation. Experiments on ProgrammableWeb.com show that DC-SeCo-LDA can recommend service composition more effectively, i.e., 2% better in terms of Mean Average Precision compared with baseline approaches.