Web Service Classification Method Based On Heterogeneous Graph Attention And Reinforcement Learning

In recent years,with the rapid growth of Web services on the Internet,it has become a great challenge to discover services from a large number of Web services and efficiently classify them,and provides appropriate web services for users.Most of the existing web service classification methods focus on the service description text and tags,but ignore the network structure information implied between the words in the web service description text and the web service description document itself.Moreover,web service classification technology based on existing functions has some problems,such as text semantic sparsity,word order and context information between words are not considered comprehensively,and so on At present,most of the proposed methods are display structure（artificial model）to learn tasks,which will bring some inherent noise and affect the classification results.To solve these problems,this paper combines the network structure information and semantic structure information of web service documents,and selects the appropriate neural network model to mine the description documents from two different angles,The main contributions of this paper are as follows:1.This paper proposes a flexible HIN（Heterogeneous Information Network）framework for modeling web service description text.It can integrate various types of additional information and capture their relationships to solve semantic sparsity and implicit network structure information.Then,a heterogeneous graph attention network（DAHGCN）model based on two-level attention mechanism is used,which includes node level and type level attention.Attention mechanism can learn the importance of different adjacent nodes and the importance of different node（information）types to the current node to solve the problems of word order between words and context information.The experimental results show that our model is superior to the most advanced methods in benchmark data set compared with other methods.2.This paper proposes an information distillation model based on reinforcement learning.Firstly,reinforcement learning is introduced to eliminate redundant information.Then,the information after distillation is classified by classification network.According to the prediction results,the reward is determined and fed back to reinforcement learning for gradient updating.At the same time,the classification network is updated reversely according to the classification results.This method can effectively alleviate the inherent noise caused by the introduction of attention mechanism,and can also learn the classification task by identifying important words or task related structures without displaying the structure annotation（artificially constructed model）.The experimental results on real programmableweb datasets show that our proposed model achieves better results on benchmark datasets than other methods.