Service clustering by leveraging a context-sensitive approach

Service technology has gained increasing popularity in recent communication software applied in many domains. With a growing number of services that share same or similar functionalities, clustering services help improve both service composition and mashup creation. To achieve service clustering, utilizing probabilistic topic model to extract and characterize the service description documents as corresponding topics is an available scheme. However, unlike short text in social networks, the descriptions of published services possess higher dimensionality and sparse functional information. With traditional LDA (Latent Dirichlet Allocation) model to implement topic extraction makes topics unclear. To address that challenge, we conduct a context sensitive approach to generate context sensitive vector for merging the words with similar context before loading to LDA model, referred to as CV-LDA (Context Vector LDA). Through F1- Measure of clustering and topic perplexity analysis in the real-world dataset, it is shown that the proposed approach outperforms traditional LDA model in service clustering. © 2016 SERSC.