Diversifying Top-k Service Retrieval

As more and more applications are based on SOA (service oriented architecture), effective service discovery is an urgent requirement for such service applications in the cloud environment. Some existing work focus on taking service content as text and using document search technique to implement service discovery. However services are designed to implement some specific functions or objectives, which leads to the underlying "topic" or "semantic" of services. In this paper, we study the top-k service retrieval problem from both the text perspective and the semantic aspect, which is to find a set of k services that can best answer a query and the result set is to balance between the content relevance and the topic diversity among the returned services. Both service content and service topic are considered to identify the candidate services. We propose the objective function which is sub-modular, and we design the search algorithm with a approximation guarantee of factor 1 - 1/e for the (best-first) greedy search algorithm. Experiments on a large TREC benchmark and services collection show the effectiveness of our approach.