Web Service Recommendation Based On Composition Patterns Mining

In recent years,with the popularity of software-as-a-service and the development of Web service technology,more and more enterprises and organizations have published their own data and services on the Internet in the form of Web services,and provided the Web services to customers and the third-party developers to achieve commercial values.As the number of Web services grows and their functions become more abundant,users can develop more powerful applications or value-added services(such as Mashups)by effectively combining Web services with different functions.Service composition has become a very popular way of software development.However,in reality,developing software by combining Web services is still confronted with challenges.On one hand,users lack understanding of the combined patterns in the Web services ecosystem and may not be able to come up with meaningful Web service portfolio requirements.On the other hand,faced by a large number of Web service resources on the Internet,it is difficult for users to quickly find Web services that meet their needs.In order to solve the above problems,this thesis studies composition patterns mining of Web services and composition patterns-aware Web service recommendation method.First,considering that each tag of a Web service embodies the function summary and description of the resource,the composition pattern of the Web service is discovered by mining the tag cooccurrence relationship in the Web service compositions.These composition patterns reflect which service function compositions are meaningful and valuable.Then,composition patterns are combined with the latest recommendation technology to automatically recommend the appropriate web service for the Mashup developer.The main contributions of this paper are as follows:(1)A Web service composition pattern discovery method(EWACP)based on association rule mining has been proposed.The method combines the tags of the Web service and the Web service composition records in the Mashup.First,the TF-IDF algorithm is used to extract the keywords in the Web service description document as new tags to extend the original tags.Then,Stanford Core NLP is used to perform tags lemmatization and the synonyms of tags are consolidated using Word Net.Finally,the FP-growth-based Web service tag association rule mining algorithm is used to obtain strong association rules between Web service tags.By filtering the tag association rules,the Web service composition patterns that can accurately reflect the Web service function association relationship are obtained.Experimental results show that the EWACP method has high accuracy and effectiveness.(2)A composition patterns-aware Web service recommendation method(EWACPDeep FM)has been proposed.The method first combines the tags among the Mashups,the Web services and the Web service composition records in the Mashups,and uses the EWACP method to obtain the composition patterns between the Mashups and the Web services,and then uses the Jaccard similarity coefficient and the historical call times of the Web service to obtain the co-occurrence and popularity of Web services,and finally uses the deep factorization machine model to train multi-dimensional feature information including the composition patterns,Web service popularity,co-occurrence,Mashup tags,Web service tags,to learn the potential link relationships between Mashups and Web services,and recommend Top-N Web services for the target Mashup.The experimental results show that the EWACP-Deep FM method is outperforms other methods in terms of precision,recall and F-measure.