API Recommendation Via Random Walk On Knowledge Graph And Deep Content Matching

With the rapid development of the information society,single-function Web APIs have become increasingly unable to cope with complex business needs.Mashup has become one of the key tools to address current challenges and an important driver of API economic prosperity to meet business needs by integrating multiple-featured APIs.The API recommendation research for Mashup aims to provide API support and improve development efficiency by recommending API combinations based on application development requirements.In recent years,the Web API ecosystem has accumulated a large amount of knowledge that can be used to enhance the API recommendation model,but research in this area is still limited.To solve this problem,this thesis proposes a technical framework for API recommendation tasks,which combines the recommendation strategy based on knowledge graph and the algorithm model based on depth content matching.In terms of knowledge graph model,this paper designs a simple knowledge graph schema to encode the context specific to API combinatorial development,and uses graphical entities to model development requirements.Then,this article uses Random Walks with Restart(RWR)Xo evaluate the potential correlation between development requirements and Web API based on the knowledge graph.In addition,this paper proposes query-specific weighting strategies to enhance knowledge graph construction.In terms of the deep learning model,on the one hand,this paper uses Word2Vec and CNN to perform vector representation and feature extraction on the text description information of historical data.On the other hand,this paper adds the multi-layer perceptron to jointly train model parameters.Then feature matching is performed,and the actual used API record is fitted according to the matching result.Finally,this paper combines the two models to achieve a powerful API recommendation model.The innovation of this thesis is that it effectively combines the history call record information of the API with the text description information,and combines the graph model with the neural network,which provides a comprehensive and effective solution for API recommendation.The experimental results show that the proposed method is far superior to some of the latest baseline methods,and has achieved significant results in reducing computational overhead and suppressing the negative Matthew effect in API recommendations.