Web API Recommendation Based On Multi-feature Fusion And Knowledge Graph Representation

Web API is a kind of application programming interface that can be used by application program to realize the service functions of storage,message and calculation.It is so easy to access,develop,compose and extend that it plays an important role in the construction of Mashup based service system.With the rapid increase of Web APIs on the Internet,developers are faced with the problem of how to choose the most suitable target from many similar APIs to build Mashups,which brings many challenges to developers.Building a recommendation system based on API can help to alleviate this problem,so a lot of work based on collaborative filtering,link analysis and topic model has been proposed;however,most of the current work only uses a single information dimension or algorithm model,and lacks in-depth mining and application of multi-dimensional features.Therefore,this thesis proposes a Web API recommendation algorithm from two perspectives: multi feature features mining and fusion,Knowledge Graph(KG)modeling and expression,in order to enhance the recommendation effect by using multidimensional features.The specific work of this paper is as followsBased on the Dempster Shafer Evidence Theory,a multi-feature fusion Web API recommendation method is proposed.Firstly,by analyzing the influencing factors of API participating in Mashup process,it can be divided into three evidence dimensions: text feature,nearest neighbor feature and API related feature;data fusion operator is used to fuse the sub features of each evidence dimension;Shannon Entropy is used to quantify the weight of each evidence dimension;finally,the data fusion operator is used to fuse the sub-features of each evidence dimension,based on the evidence of three dimensions,this paper evaluates the correlation between the requirements of API and Mashup,so as to realize API recommendation.The experimental results show that the Non-machine learning recommendation method can get satisfactory results.Starting from the method of machine learning,based on the modeling and expression of Knowledge Graph,a Web API recommendation method with embedded semantic representation enhancement is proposed.Firstly,according to the relationship between API and Mashup,such as recommendation,recommended,tag,attribute and so on,the triples are extracted and the Knowledge Graph is constructed;the shallow vector expression of API and Mashup is obtained by using knowledge representation learning Trans algorithm of Knowledge Graph;in order to enhance the effect of Knowledge Graph expression learning,the bidirectional gating loop unit is constructed;in order to get the depth vector representation of Mashup and API,a text coding module based on Bidirectional GRU neural network(BiGRU)and Attention Mechanism(Attention)is proposed.In this paper,the text coding module and trans series algorithm are combined for alternate training to achieve deep and shallow collaborative expression learning.Finally,cosine similarity is used to quantify the correlation between target Mashups and candidate APIs to achieve API recommendation.Experiments show that this method can get better recommendation effect.