Research On Mashup Tag Recommendation Algorithms Based On The Topic Model

Web service was created to provide an Internet-based solution for the different Internet users.Due to its many benefits such as low cost and loose coupling,etc.,Web services have been a favorable choice for developers when creating new Internet-based applications.People can incorporate different Web services and various Web resources to construct more complicated Web application called Mashup.In recent years,with the development and popularity of the Internet technologies,enormous and multifunctional Web services emerged,which raises new challenge of managering these resources efficiently to achieve the automatic Web service classification and improve the retrievaling efficiency.It has been proved that annotation can address these problems to some extent.For example,Programmable Web,the most active Web service registering platform,allows users to associate the Web service with several relevant tags.In general,tags are able to reveal the semantic information of the Web services and this can help users to understand,locate,classify and manage the Web services.However,people prefer munual annotation at moment,which is time-consuming and tedious.The automatic annotation methods are desired.In this thesis,we propose two topic model-based approach for Mashup service recommendation,including similarity-based and prediction-based Mashup tag recommendation algorithm.For the similarity-based Mashup tag recommendation algorithm,we have done the following work:1,We propose a method for Mashup tag recomme ndation based on a topic model.The model simultaneously takes the description documents for Mashups and Web Application Programming Interfaces(APIs)as well as the composition relationships between them into account.Based on the model,our approach firs t selects the most similar APIs of the target Mashup.Subsequently,those chosen similar APIs and composed APIs of this Mashup are combined into a single APIs set.We select several most important APIs from this APIs set based on a weighted Page Rank algori thm.Finally,tags of these important APIs are recommended to the Mashup.We also design an algorithm to rank tags recommended according to their topic relevance with the target Mashup.The experimental results on a real world dataset collected from Programmable Web prove that our approach obviously outperforms other tag recommendation methods.For the prediction-based Mashup tag recommendation algorithm,we have done the following work:1,We propose a novel Topic-Sensitive approach based on Factorization Machines for mashup tag recommendation.We exploit various types of relationships as features.Factorization Machines is utilized to model the pair-wise interactions between all features and predict adequate tags for mashups.In this approach,we first obtain the latent topics of all tags,the description documents for mashups and APIs based on a novel probabilistic topic model.Then,a multi-relational network by mining various relationships from the Web service data is constructed.Various auxiliary information are subsequently extracted from the network to train the Factorization Machines.The proposed model is evaluated on three real-world datasets.Experimental results show that it outperforms several state-of-the-art methods.