Diversified keyword search based web service composition

Abstract

To assist system engineers in efficiently building service-based software systems, the keyword search based service composition approach on service connection graphs (scgraphs) has been proposed recently. However, due to the ambiguity of keywords, a keyword query may represent a bunch of different user requirements. Thus the current approach that only returns the composition with the optimal Quality of Service (QoS) cannot guarantee to hit the spot. In this paper, in order to satisfy the various possible requirements underlying a given keyword query, we formally introduce the top-k diverse service composition problem, and present a novel diversified keyword search approach on scgraphs to address it. Specifically, we firstly propose an All-Then-Diversify (ATD) algorithm that enumerates all potential compositions by searching a scgraph and then derives the top-k diverse subsets by deriving the maximal independent sets of a similarity graph. Then, due to the possibly large number of potential compositions, we present a Pop-And-Diversify (PAD) algorithm that only maintains a similarity graph of the top compositions that have been found so far during the search and computes its maximal independent sets incrementally until convergence, thereby reducing unnecessary computation overheads. Moreover, we propose two composition similarity measurements w.r.t. the categories or descriptions of services respectively. Lastly, the experimental results on ProgrammableWeb.com demonstrate that, our approach outperforms another state-of-the-art composition diversification approach on both metrics of density and redundancy, and meanwhile, improves the efficiency of diversification significantly.