Augmented latent Dirichlet allocation model via word embedded clusters for mashup service clustering

Abstract

AbstractAlong with the growing number of mashups, there have been many changes in their diversity as well. With the creation of a wide variety of mashup services, it has become difficult to select and recommend appropriate web‐based APIs or services to launch an application based on the principle of mashup technology. In keeping with the methods already in place, we realized that the input/output parameters and text description which are key features directly from WSDL specification documents are increasingly being used in the mashup service innovation process. The quality of these methods can be enhanced by capturing hidden topic content from WSDL documents through the LDA (latent Dirichlet allocation). In this paper, the performance as concerns the LDA model can be enhanced by using some auxiliary features rather than limiting it to some extent. The reason for this is that the word vectors achieved by the word2vec tool is of advanced quality than the word vectors of the word gained by the LDA model. So, we proposed to innovate the novel mashup service innovation and clustering method with better efficiency and higher accuracy through word2vec. Here developed an improved LDA model with a word embedded cluster (called AUG‐LDA) that has a major impact on word vector quality. In this proposed approach, the LDA's training process can be semimonitored by first acquiring the word vectors through the word2vec tool and then merging the word clusters with the K‐means++ algorithm. This will enable better representations for the distribution of mashup services. This can be obtained by crawling the dataset from the website calledwww.programmableweb.comto perform a comprehensive experiment and demonstrate the proposed method. It has been proven that there is a better increase by comparing our approach with different metrics and the clustering accuracy values obtained by other methods.