| In recent years,with the development of modern information technology such as Internet,big data and artificial intelligence,more and more software systems,storage resources,and computing power are delivered in the form of services which can be accessed by end users in an easy way.As an emerging software asset,services are rising progressively.As a new computing paradigm,service computing provides a distributed computing infrastructure for application integration and collaboration intra-enterprise and inter-enterprise.It advocates building distributed applications through the composition of services,which has greatly changed the way of design,architecture,deliver and use for software applications and has receiverd extensive from the research community and the industry.With the popularity of service computing,the number and types of services on the Internet are rapidly increasing,and the data sources are increasingly diversified.The "Service Internet" based on services and their associations is being formed.The increasing scale of service Internet and the complexity of internal interaction bring many challenges to the research of service computing.From the perspective of systems science,the structure of a system affects its function.The Service Internet is also an artificial complex system whose(overall/global)structure also affects the functionality of the services in it.Due to the lack of appropriate representational means,researchers carry out the research on service computing from the perspective of the structure of the service Internet,which leads to a lack of adequate understanding of the structure,function and dynamic behavior of the service Internet.Complex network is investigated for using network model to the study the complex systems.By ignoring the details of the system implementation,it is more conducive to find the general rules of system structure,which provides a powerful tool for studying the internal structure of the system from the global perspective,and also provides a possible effective way to solve the problems related to services computing caused by the scale and complexity.Recently,many researchers have introduced complex networks into service computing and built a complex network model of service Internet—called service network,in which the nodes represent services and edges represent interaction structure between services.Based on service network,we reexamine the research on service computing.At present,the research of service network is still in infancy.Although some gratifying achievements have been made,it still stays in the construction of service network and structural analysis,and lacks relevant research on the guidance of service computing practice(clustering,discovery,recommendation,etc.)through structural analysis.At the same time,the structural analysis of service network is not comprehensive enough.According to the above problem,this paper will adopt the cross research method of complex networks and services computing take service network as the research object,analyze the structural characteristics of service network as the starting point,provide support for service clustering,service discovery and service recommendation as the goal,focus on the service network structure analysis based on complex network,service clustering method based on structural metrics,service discovery method based on the field label ontology,service recommended method based on the multi-dimensional information matrix and factoring machine.The specific research contents and achievements include:(1)Analysis of service network structure based on complex networkThe interconnection between massive services forms a super-large scale complex service Internet.Complex network is a powerful tool for analyzing the structure of complex systems,which can deepen our understanding of the system structure and provide us with a new way of thinking to understand and study the structure of the service Internet.In this paper,a structural analysis method of service networks based on complex networks is given.Programmable Web(PWeb),a real service library,is given as the carrier to make an empirical analysis.We construct the service network of PWeb,and introduce the structural parameters(degree distribution,degree centrality,medium number centrality,proximity centrality,PageRank value,network kernel,aggregation coefficient,etc.)in the complex network to analyze the structural characteristics of the service network.The experiment shows that the service network has obvious complex network characteristics such as no scale and small world.DC(degree centrality)is the best parameter to identify the importance of a service,but the difference between the parameters is not obvious.(2)Service clustering method based on structure similarity measurement and community miningService clustering reduces the search space of services by clustering similar services,thus effectively improving the performance of service discovery methods.Therefore,it has important significance to improve the performance of the service clustering method.The results show that the structure-related similarity index performs better in service clustering than the semantic-related similarity index.In view of this,this paper proposes a serviceclustering method based on structural similarity measurement and community mining.First,we construct the membership,dependency and similarity relationship among abstract services of various types of service network.Then,the similarity between services is evaluated based on the constructed service network.Finally,the community mining algorithm is introduced to identify the community structure in the service network,so as to realize the service clustering into clusters.The experimental results on PWeb data set show that the proposed method has betterperformance compared withsimilar work.(3)Service discovery method based on service clustering and domain tag ontologyHow to accurately and efficiently retrieve the required services has become a major problem faced by service consumers.Research shows that the service discovery method based on semantics has better performance than the traditional service discovery method based on syntax.Therefore,it is of great significance to improve the performance of semantic service discovery from the perspective of service network.This paper proposes a service discovery method based on service clustering and domain tag ontology.First,domain tag ontology of each service category is constructed based on the service category identified by the research content(2).Then,the concept of user story in agile development is introduced to describe the user's service requirements and service description information with the three elements of agile requirements.Finally,the service category is determined by calculating the similarity between user requirements and tag ontology,and then the discovered service is determined by calculating the similarity between user requirements and services within the category.The experimental results on PWeb dataset show that compared with similar work,our proposed service discovery method based on domain tag ontology can quickly respond to user requirements and has better performance.(4)Service recommendation method based on multidimensional information matrix and factorization machineAs Web API and Mashup applications continue to be published,a large number of Web API and Mashup applications appear with the same or similar functionality.How to quickly and efficiently find or recommend a Web API service set meeting the complex requirements of developers according to users' unstructured natural language expression is a hot research topic in the field of service computing.Existing researches show that the service recommendation method based on hybrid feature has better performance due to the combination of functional feature and non-functional feature.In this paper,a service recommendation method based on multidimensional information matrix and factorization is proposed.First,dependency syntax is used to mine the semantic information in Mashup and service description documents.Secondly,the similarity between mashups is calculated based on the content and the invocation relationship between mashups and services,and the content-based similarity of services is calculated.Then,Mashup similarity,Web service similarity,link information between Mashup and Web service,popularity of service and co-occurrence are fused to generate multidimensional information matrix.Finally,FM is used to simulate all interactions between input variables and multidimensional information to effectively predict the association between target mashups and services,thus improving the accuracy of Web services prediction and Mashup recommendation.Experiments on PWeb data set show that compared with similar work,the proposed method significantly improves the accuracy of service recommendation. |
PDF Full Text Request