Attentional Neural Factorization Machine for Web Services Classification via Exploring Content and Structural Semantics

Due to the rapid development of Web 2.0, a lot of Web services emerge over the Internet. How to efficiently manage Web services through classification is very important for Web service discovery. Although there have been a lot of works on Web services classification, they still have drawbacks. On one side, the feature extraction from the service repository is insufficient, which will influence the classification accuracy no matter what classification model is used. On the other side, the extracted features are used with the same weights and they lack depth fusion when training the classification model. In real-world application scenarios, different feature interactions often have different predictive capabilities, and not all feature interactions contain useful or positive information for estimating the target. In addition, both low-and high-order feature interactions are usually underlain real-world data. To solve the problems above, this paper proposes a novel Web services classification approach via fully exploring and integrating the content and structural semantics of Web services. In the proposed approach, the content representation is explored with the BERT-based document embedding model, and the structural representation is explored with the Node2vec network embedding model. Finally, attentional neural factorization machine is used for both the deep fusion of features and Web services classification. A set of experiments are done on real-world datasets crawled from Programmable Web. And solid experimental results show that the proposed approach outperforms the state-of-the-art approach and the other baselines.

I. Introduction

Web services are self-contained, modular, distributed, dynamic applications that can be described, published, located, or accessed over the network to create products, processes, and supply chains. Mashups are a new type of Web service that is composed of Web APIs. Mashups leverage content from different data sources to create new services. Hereafter, Mashups and Web APIs are also referred to as Web services in this work. Due to the advantages of Mashup technology, more and more developers and ordinary users tend to use Mashups to solve their business requirements. The great demand from users stimulates developer users to publish Web APIs on the Internet. Thus, a plentiful number of Web services are published on the Internet. Take the well-known ProgrammableWeb platform as an example, which hosts over 24,000 APIs and 8,000 Mashups. With such a huge number of Web services on the Web, how to manage them efficiently by means of web services classification according to their functionality is of great importance for further tasks, like service discovery or selection, service composition, and service recommendation [1]. However, there are two main challenging issues for accurate Web services classification. The first one is how to extract sufficient multi-dimensional features from the service repository since functional features of Web services are very limited by extracting from service descriptions. The second one is how to train an effective classification model with the extracted features since complex relationships may exist among different features.