A semantic matching approach addressing multidimensional representations for web service discovery

Abstract

In recent years, discovering appropriate web services has become increasingly difficult as the number of services has grown rapidly. With the goal of improving discovery performance through accurate text matching, this study developed a service discovery method that constructs a neural matching network based on multidimensional service representations. Specifically, we performed data processing and adopted three methods called term frequency-inverse document frequency, Word2Vec, and ELMo to generate multidimensional representations for capturing the word frequency, static context features, and dynamic context features of each keyword. Based on these features, we calculated the cosine similarity of pairs of keywords to construct a multidimensional similarity matrix. We then implemented convolution, pooling, and optimization operations to construct a neural matching network that has a direct impact on the accuracy of service discovery. Finally, for a given query, target services are retrieved by ranking candidate services according to the scores predicted by the matching network. The proposed method was evaluated through multiple comparisons and the experimental results demonstrate the effectiveness of optimal web service retrieval.