Web API Recommendation With Graph Network And Semantic Information

As a lightweight Web application,Mashup can meet the complex requirements of various web applications by combining existing Web APIs,which reduces developers’development time and improves development efficiency.With the rapid growth of the number of Web APIs,how to select a suitable Web API for Mashup development from this huge resource library has become an urgent problem to be solved,and the data sparsity and cold start problems also bring challenges to the recommendation of Web APIs.To this end,this paper focuses on Web API recommendation.For a given Mashup requirement,the text description information of Mashup and the rich relationship in the graph network are used,and corresponding recommendation algorithms are proposed respectively in these two aspects.The main contents are as follows:Firstly,in terms of semantic information recommendation,this thesis proposes a nonnegative matrix factorization algorithm based on Mashup semantic information.Firstly,the text description of the mashup is preprocessed,then the semantic similarity between the mashups based on word embedding is calculated,and the improved K-means algorithm is used to cluster the similar mashups into the same category to get the cluster center-API call Finally,the non-negative matrix factorization method is used to obtain the cluster centerspecific matrix and the API-specific matrix,so as to obtain the predicted value.This method utilizes descriptive text information,which can alleviate the cold start problem of new Mashup to a certain extent.Using the data obtained from the ProgrammableWeb website,the proposed method is compared with the baseline method,and its performance on the four indicators of Precision@N,Recall@N,F1@N and NDCG@N is analyzed.It can be found that the proposed method is better than the baseline method.All baseline methods perform better.Secondly,in the aspect of graph network recommendation,this thesis proposes a biased random walk recommendation algorithm based on graph network.The Mashup-API graph network is first constructed,then perform a biased walk according to the types of different nodes in the constructed graph network,and then train the result of the walk with the SkipGram model to obtain the vector representation of the nodes to calculate the calling probability of Mashup to the Web API,so as to realize the application of Web API.recommend.This method not only utilizes Mashup’s historical invocation record information of Web API,but also utilizes Mashup and Web API’s structure information,which can improve the accuracy and compatibility of recommendation.Using the data obtained from the ProgrammableWeb website,the proposed method and the baseline method are compared experimentally,and it is found that the proposed method has better performance than all the baseline methods in recommendation.