Modeling and exploiting tag relevance for Web service mining

摘要

Web service tags, i.e., terms annotated by users to describe the functionality or other aspects of Web services, are being treated as collective user knowledge for Web service mining. Since user tagging is inherently uncontrolled, ambiguous, and overly personalized, a critical and fundamental problem is how to measure the relevance of a user-contributed tag with respect to the functionality of the annotated Web service. In this paper, we propose a hybrid mechanism by using Web Service Description Language documents and service-tag network information to compute the relevance scores of tags by employing semantic computation and Hyperlink-Induced Topic Search model, respectively. Further, we introduce tag relevance measurement mechanism into three applications of Web service mining: (1) Web service clustering; (2) Web service tag recommendation; and (3) tag-based Web service retrieval. To evaluate the accuracy of tag relevance measurement and its impact to Web service mining, experiments are implemented based on Titan which is a Web service search engine constructed based on 15,968 real Web services. Comprehensive experiments demonstrate the effectiveness of the proposed tag relevance measurement mechanism and its active promotion to the usage of tagging data in Web service mining.