Web Service Classification And Recommendation Based On Graph Attention Mechanism

More and more companies publish their business data online in the form of API,and the number of Web services is increasing day by day.In this context,how to quickly find API services that meet the needs of developers and users from a large-scale service database has become a challenging problem.Due to sparse features,difficult modeling and difficult collection of real quality of service data,the existing service recommendation methods have some problems,such as inaccurate text representation,unable to calculate the importance between service nodes,and failing to consider the meta-path semantics of heterogeneous service information networks.Therefore,from the perspective of service representation,this paper proposes the following web service classification and recommendation methods:(1)We propose a service recommendation method integrating contract composition,attention network representation and Deep FM quality prediction.The first step of this method is to classify web services based on graph attention network.Firstly,the service relationship network is constructed according to the service composition relationship and shared annotation relationship of Web services;Then,the self-attention mechanism is used to calculate the attention coefficients of different neighbor service nodes in the service relationship network.By weighted summation of the attention coefficients of neighbor services and their own features,the services can be embedded with high quality,and the representation results can be classified effectively by using Softmax.In the second step,we combine the high-quality representation results with multi-dimensional Qo S attributes to build a quality of service information matrix,mine the complex interaction between features by using Deep FM,and predict and sort the call scores of Web services.(2)We propose a service recommendation method combining heterogeneous graph attention network representation and Fi Bi NET quality prediction.The first step of this method is to use heterogeneous graph attention network to enhance the classification effect of Web services.Firstly,we use composite service information,atomic service information and their respective attribute information to build a heterogeneous information service network,and define meta-paths according to different semantic information;Then the service similarity matrix is constructed by using the commuting matrix and the similarity measurement technology based on meta-path;Next,we design a service double-layer attention model to calculate the node-level attention and meta-path-level attention of the service respectively,so as to obtain the node-level representation and meta-path-level representation of the services.After fusion,we generate a more expressive service feature embedding,and use Softmax to effectively classify the Web services.In the second step,we use Fi Bi NET to dynamically learn the importance of features and complex feature interactions,and predict and sort the call scores of Web services.Experiments on Programmable Web datasets show that the proposed methods are superior to other mainstream baseline methods in terms of performance indexes such as Precision,Recall,F1,DCG and AUC,and achieves better classification accuracy and recommendation effect.