Web Service Recommendation Approach Based On BTM And GPU-DMM Topic Model

With the prevalence of Service-Oriented Architecture(SOA),the number of Web services is increasing rapidly,which meanwhile results in many functionally equivalent services existed on the Internet.Such a situation makes it hard for users to select the desired Web services with respect to a given application environment.Previous studies show that service recommendation technologies can effectively address this problem by accurate matchmaking between user requirements and relevant services.In the past,the probabilistic topic model has been widely investigated to mine functional semantics from textual service descriptions and user requirements,which are then used for functional properties matching and services recommendation.However,traditional topic models(e.g.,LDA)perform poorly on sparse and short texts,so does to model Web services which usually have limited number of words.To address the aforementioned problems,this paper seeks to develop and adopt the adapted topic model to efficiently elicit semantics from short service descriptions,which are then integrated into a depth factorization model with incorporating multidimensional helpful features for Web service recommendation.Our contributions are summarized as follows:1?We proposed a Web service clustering approach based on the BTM(Biterm Topic Model)model.It first removes the noisy features from Web service descriptions,and then learns the functional semantics of services based on the word co-occurrence-based BTM topic model.The training process is done by the Gibbs sampling,which finally generates the topic distribution of each Web service.Finally,it uses K-Means algorithm to cluster Web services based on the JS similarity measurement.Compared with methods such as LDA and TF-IDF,the proposed approach can achieve better performance in terms of the purity,entropy and F-measure.The results also demonstrate that our method can effectively solve the data sparseness problem,which paves the way for related research problems such as short text modeling and service recommendation.2?We proposed a Web service recommendation approach based on the word2 vec and probabilistic topic model.First,the semantic word vectors can be obtained from the external English Wikipedia corpora,which enable to calculate the word similarity between words.Then,the DMM model training with the GPU promotion strategy can be enhanced by the previously known word similarities,i.e.,the friend words can be used to infer more accurate semantics of the current sampling word.Finally,we proposed an approach for service recommendation based on the deep factorization model with combining various features,including the similarity between Web APIs,the similarity between Mashups(obtained by the revised DMM model),the co-occurrence of Web APIs and the popularity of Web APIs.Experimental results on a realworld dataset demonstrated the effectiveness of the proposed approach against several state-ofthe art baselines.