Semi-supervised Service Clustering And Tag Recommendation Approaches For Mashup Services

In the Web 2.0 era,the number of Web Services has grown rapidly,and the Mashup based on Web integration has increased dramatically since its rise.Mashup technology makes it possible to develop new Web Services faster and more easily,but the rapid growth of services makes the Mashup Services developers have to spend a lot of time managing all the existing services.In the field of service computing,it is a hot topic that how to analyze and process a large amount of service data properly through machine learning technology,improve the management efficiency of large number of Mashup services.Based on the survey of service tag recommendation methods,this thesis compares the Mashup service tag recommendation method based on supervised learning(SLMSTR)and unsupervised learning which is mainly service clustering(SC-MSTR)on two datasets published in 2016 and 2018.The experimental results of the two datasets show that the unsupervised SC-MSTR method has better performance on tag recommendation than the supervised SL-MSTR method.And it can be observed that the unsupervised SC-MSTR method always perform better than the SL-MSTR method when the number of candidate tags changes.Further analysis of the experimental results show that the SL-MSTR method relies on the similarity between the current Mashup service and the candidate tags,the SC-MSTR method relies on similarity of the current Mashup service and the Mashup service with existing real tag set.Since the latter is a measure of similarity of similar features,the recommended results are more accurate.In this thesis,the SL-MSTR method and the SC-MSTR method are considered to have great differences in principle.So it is expected to obtain further optimization of the recommended quality through the integration of the results of the two methods.Experiments show that the fusion method of the Mashup service tag recommendation(Fusion-MSTR)has greatly improved the quality of Mashup service tag recommendations.The main work of this thesis are as following:(1)The thesis studies and analyzes the traditional services clustering approaches,and introduces the idea of supervised learning.Then the S3C_PL-MSC approach is proposed on the basis of pseudo-labels which are obtained from the traditional Mashup services clustering results.The experimental results on a public Mashup services dataset show that the proposed S3C_PL-MSC approach made an optimization of the traditional service clustering approach on 8 metrics.(2)The thesis analyzes the data characteristics of Mashup services,and studies various traditional unsupervised Mashup service tag recommendation approaches.Then it considers the Mashup service tag recommendation problem as a preference ranking problem of candidate tags according to the memberAPIs,the WSDL documents and the existing tags of Mashup services.As a result,the thesis proposes the SL-MSTR.The experimental results on two public Mashup service datasets show that the proposed SL-MSTR in this thesis improves the performance of the traditional Mashup service tag recommendation approach on precision,recall and F1.(3)Based on the proposed Mashup service clustering approach(MSC),this thesis designs and implements a Mashup service tag recommendation approach,called SCMSTR.This thesis uses the voting mechanism to fuse the result of SC-MSTR with SLMSTR to obtain the Fusion-MSTR approach.The experimental results on the Mashup service dataset show that the Fusion-MSTR approach has a further improvement on the Mashup service tag recommendation performance.In general,the semi-supervised Mashup services clustering approach based on pseudo-labels optimized the traditional Mashup service clustering approach effectively.The Mashup service tag recommendation approach based on supervised learning improves the service tag recommendation performance.At the same time,this thesis adds the service clustering into the service tag recommendation,and then uses the voting mechanism to fuse SC-MSTR and the SL-MSTR,which further improves the performance of Mashup service tag recommendation.