Web Services Classification And Recommendation Based On Attention Neural Network

In recent years,with the large-scale application of Internet technology,the number of Web services increases significantly,but their functions tend to be homogeneous.This practical problem makes it difficult for developers to quickly search their desired web services according to their functional development needs.Service recommendation is becoming the key to automatically and effectively find the right service from a large number of candidates and overcome the problem of information overload.Researchers have proposed a variety of service recommendation models to improve the quality of service recommendation.Previous studies have shown that it is effective to build a recommendation model by integrating multiple service characteristics.In this paper,we take this forward a step,and think that these studies have effectively improved the recommendation performance to a certain extent,but they only model all features with the same importance,and ignore the fact that not all features are equally useful and predictive,some useless features may even introduce noise and have adverse effects on the recommendation results.In this paper,we propose a new service recommendation strategy,which uses attention mechanism and Attentional Factorization Machine to distinguish the importance of each feature from multi-dimensional features to meet this challenge.Experimental results show that our model has better performance and interpretability.Our main contributions are summarized as follows:1?Existing function-based service clustering techniques have some problems,such as the sparse document semantics,unconsidered word order and the con-text information,so the accuracy of service classification needs to be further improved.In this case,this paper proposes to use attention mechanism to further refine and filter the functional semantic information in the text.A new web service classification method is proposed by combining the document level LDA topic vector representation with the Bi LSTM hidden state vector of a single word in the text.Specifically,it uses Bi-LSTM to automatically learn the feature representation of Web service.Then,the topic vector of web service text is obtained by offline training,and the generation process of web service function representation vector is enhanced by attention mechanism,and the weight value of each word in Web service text is obtained.Finally,the enhanced Web service feature representation is used as the input of the Soft Max neural network layer to perform the classification prediction of Web service.It can provide fundamental support for further service research.2?To some extent,the existing research effectively improves the recommendation performance,but they only model all features with the same importance,and ignore the fact that not all features are equally useful and have prediction ability.Some useless features may even introduce noise and have adverse effects on the recommendation results.To solve this problem,this paper proposes a web service recommendation model(AFMRec)based on Attentional Factorization Machine.In this model,firstly,the valuable service features hidden in the original data set are extracted and transformed into the input format of the Attentional Factorization Machine.Then,Attentional Factorization Machine is used to model multidimensional information,such as functional similarity between mashups and services,tags,popularity of Web APIs,etc.,in order to predict the recommendation score between Mashup and candidate services.The comprehensive experiments on the crawled dataset show that the proposed method significantly improves the accuracy of recommendation compared with a variety of basic algorithms.