Web Services Clustering Based on HDP and SOM Neural Network

Discovering appropriate, user-desired Web services from massive Web services quickly and accurately for users to build valuable Web applications has become a significant challenge. Web services clustering has been proved to be an effective way for Web service discovery. Some recent research works exploit additional information (such as, tags or word clusters information) to argument LDA-based topic modeling for better service clustering. Although they definitely boost service clustering via mining more implicit topics and semantic correlation of service document, their performance still can be improved due to inherent disadvantage of LDA topic model and adopted clustering algorithms. To address this problem, we propose a Web services clustering method based on HDP topic model and SOM neural network. This method, firstly uses Word2Vec to expand the original Web services description document from Wikipedia English corpus and applies HDP topic model to model the expanded Web services description document to obtain the document-topic vector. At last, it employs SOM algorithm on the document-topic vector to achieve Web services clustering. The comparative experiments are performed on ProgrammableWeb dataset. The experimental results show that the proposed method respectively achieves significant improvements of 486%, 54.0%, 35.8%, 47.1%, 39.0%, 28.6%, and 9.4%, compared with other seven service clustering methods.