Word Embedding-based Web Service Representations for Classification and Clustering

Web services classification and clustering rely on discriminative representations of service functional semantics. Most existing topic-based methods for Web service representation simply leverage word co-occurrence relations, which may cause sub-optimal results for some specific-tasks due to the sparse features in the short service descriptions. Despite sparse and discrete word statistical features, other information such as word-topic and word-tag relationships can be utilized to accurately represent Web services. In this paper, we propose a novel word embedding-based approach for representing Web services. First, we use Word2vec to generate word embeddings from service descriptions. Then, the word-topic and word-tag relationships are preserved for Web service representation generation via the topic-weighted and tag-weighted word embedding stages based on Gaussian Mixture Model and Bayesian rule, respectively. Experiments and validations on the real-world dataset demonstrate that our method outperforms the state of the arts by a wide margin on service classification and clustering.