Web Service Matchmaking Using Web Search Engine and Machine Learning

Web Services discovery that locates adequate services, has been studied very actively for better quality of service retrieval. Starting from conventional keyword matching, logic-based matching and combination of the methods with information retrieval approach have been proposed to enable better discovery performance. The combining method using term-similarity can overcome the decision failure when the keyword or the logic-based methods were applied, and it was shown that the methods outperform the existing methods. And researches to aggregate matchmaking variants by machine learning has been attempted, and it also improves the discovery performance. The approaches still suffer from fixed corpus set for term similarity calculation. In this research, we attempted to calculate the similarity based on search engine to reflect the current Web context. Tokenized terms are used for the matchmaking degree. Variants for the matchmaking from ontology and term similarity are aggregated using Support Vector Machine (SVM) with non-linear kernel function. Matchmaking test on the trip domain service discovery was conducted. Experimental result based on the standard measure of precision and recall rate for the top 1-20 services of matched result on the trip domain test set are shown.