Research On Web Service Recommendation Based On Heterogeneous Information Network

With the development of SaaS technology,the development of service-based software system(SBS)has become the main method to generate software on the Internet.In order to put all kinds of service resources in the cloud,there are many public service registries,such as PWeb,Mashape,etc.There are a lot of services in these registries.At the same time,these service resources have obvious heterogeneous characteristics,including the protocols they follow(such as SOAP,REST,etc.)and the description methods they use(such as WSDL,Natural language text,etc.).How to recommend appropriate services for software system from massive heterogeneous services is a very challenging problem in service-based software development.The effect will directly affect the efficiency of software system development and the quality of service reuse.In response to the above facts,the key issue to be resolved in this paper is “How to use heterogeneous information network for effectively recommending services in the face of the large number of heterogeneous and semantically sparse service resources over the Internet”.Focusing on this key issue,the contributions of this article are mainly in the following aspects:(1)A RGPS-based service clustering method is proposed.Different types of service description documents are uniformly converted into feature vector forms based on the RGPS requirements meta-model framework,and then,the service feature vectors are converted into potential features by using the BTM model,and the services are used to describe the potential features for service clustering.This method can extend the feature description information of semantic sparse from RGPS multi-dimensional and lay a foundation for topic model clustering.(2)An integrated service recommendation method based on heterogeneous information network(HIN)is proposed.The heterogeneous information network is used to uniformly model the heterogeneous resources of public service registry,and the similarity between objects is mined by using the meta-path and word embedding integration.Two different ways can more accurately measure the semantic similarity between objects in different types of meta-paths,and use BPR algorithm to optimize the combined weights of different paths.The effective use of the service clustering method in the method helps to effectively organize the services in the registry and improve the recommendation efficiency.Through a large amount of experiments,it is proved that the contribution of different meta-paths in the similarity calculation is different,and the optimal combination of meta-paths is given for the requirements in different application scenarios.(3)A network representation learning based service recommendation method is proposed.In order to fully understand the implicit information in service,the heterogeneous network embedding and word embedding methods are used to extract the structural features and content features of service and vectorized representation respectively.The collaborative joint learning is used to represent the SBS and Service objects in the heterogeneous information network.Service recommendation based on the distance between SBS and Service in vector space.This method mines the potential vector representation of heterogeneous objects through network embedding,which helps to more accurately match the services that meet the user's needs.In this thesis,the key issue of “distributing heterogeneous service resources on the Internet,how to organize services to effectively recommend services” is explored,and based on topic-oriented service clustering,a heterogeneous information network-based service recommendation framework is proposed for software developers in multiple scenarios through a multi-strategy approach.At the same time,multi-dimensional analysis and characterization of services based on RGPS helps to alleviate the semantic sparseness of service descriptions.It helps to discover the full range of implicit information of services by using heterogeneous information networks to model heterogeneous service resources.