Research On Requirement-driven Intelligent Development Approaches For Mashup Services

Nowadays,Mashup service,as a lightweight service composition model,combine multiple Open APIs or remote services into a composite application to meet the complex requirements of developers,and has become a popular development approach.Mashup service is widely used in cloud computing,mobile Internet,service-oriented computing and other fields.Using the Open API,developers do not need to develop Mashup service from scratch,as long as the reuse of Open API on the Internet can greatly improve the development efficiency of developers.On the Internet,after users have used these services,they will put forward their requirements and suggestions based on their own experience.By taking advantage of this information,developers can better develop Mashup services.However,the current Mashup service development still has the following problems: 1)How to effectively mine meaningful user requirement information from the massive user-view-based service review data;2)How to efficiently discover suitable,high-quality Open API according to the changing requirements of users.In order to effectively solve the above problems,this paper proposes a requirement-driven intelligent development approach for Mashup services,which consists of two stages: user requirements elicitation and service recommendation.The specific work is as follows:(1)In the aspect of requirements elicitation based on user review data,requirement classification has always been the focus and difficulty of research.Aiming at this problem,this paper proposes an approach to obtain Mashup service requirements based on user review data.The main steps of the approach are as follows: firstly,nine different requirement categories(functional requirements and non-functional requirements)and one other category expressing user emotions are defined according to the application scenario;secondly,the Mashup service review data is crawled and preprocessed,and manually label the data,then generate highquality sentence embedding vectors through the BERT model,and then input to the RCNN(Recurrent Convolutional Neural Networks)model for effective mining.Finally,the topic model is integrated to form a feedback list and return it to the developers for the development and iteration of the Mashup services.(2)According to the requirements put forward by users,a deep learning-based Open API recommendation approach is proposed,which integrates multiple functions to accurately and effectively help developers get suitable Open APIs.The main steps of the approach are as follows: first,use the Ro BERTa model to generate high-quality sentence embedding vectors,then use the HDBSCAN(Hierarchical Density-Based Spatial Clustering of Applications with Noise)model to obtain Mashup clusters;secondly,extract the requirement keywords through the Text Rank and build a language model based on Word2Vec;then generate candidate Open APIs through a three-level similarity calculation structure;finally,these candidate Open APIs are supplemented with additional information to form a recommended list to feed back to service developers.In order to evaluate the two approaches proposed in this paper,this paper conducts experiments on a set of datasets constructed by scholars in the professional field and a wellknown Mashup development platform——Programmable Web(PWeb).The experimental results show that those approaches in this paper have better performance to a certain extent.