Research On Heterogeneous Data Fusioin Based Service Discovery And Recommendation

With the rapid development of service computing and mobile Internet,the research objects of service computing has expanded from the original Web services to various types of services,such as API and mobile App services.The scale of service data continues to expand and data sources are increasingly diversified,which not only brings more service choices for users and new opportunities for the research of service computing,but also poses new challenges to traditional service mining.To deal with the heterogeneous service data,it is necessary to propose new service mining methods to learn how to effectively integrate different service data sources,how to construct complex rela-tionships between service objects,and how to fuse multiple view of features.This work focuses on social media information based service discovery,heterogeneous information network based service generation,service recommendation through multi-view feature fusion,and deep learning based service recommendation.The main contributions are summarized as follows.Considering the limited discovery performance caused by the single source of data in tradi-tional service discovery research,we introduce the social media information of services and propose a service discovery approach integrating different social factors.This approach proposes four so-cial factors that are semantic similarity,popularity,activity,and decay factor according to social information of API services in Twitter.To integrate all the social factors measured in different ways,a weight learning algorithm is proposed to learn the contribution of each factor.Finally,the service discovery is implemented for a given query by a weighted linear model combining all the social factors.Introducing social media information that contains rich direct feedback and collective knowledge greatly enriches the semantics of services,and can further meet both users’functional and non-functional requirements of services.Experimental results in our work prove that leveraging social media information can effectively improve the performance of service discovery.Considering the heterogeneous service data and complex relationships among the data,we pro-pose a heterogeneous information network based recommendation method for service composition.The method constructs a heterogeneous information network for services and proposes a similarity measurement based on different meta-paths between service objects.Combined with collaborative filtering,the recommendation model for service composition is trained under the BPR framework.The way of expressing service data as a unified heterogeneous information network effectively integrates the heterogeneous data.The utilization of similarities calculated under meta-paths with different semantic information can improve the quality of service recommendation and model inter-pretability as it considers users’ multiple demands.Experiments based on a real dataset verify the effectiveness of the service recommendation method based on heterogeneous information network.To alleviate the information overload problem caused by the increasing number of services,we propose a multi-view feature fusion based service recommendation method.The method mod-els different feature interactions for different types of services,and builds the category feature,multi-view features,and the interactions between them into a unified tensor structure to obtain the full-order interaction information.A tensor decomposition technique is applied to learn the weight parameters corresponding to the high-dimensional feature interaction tensor.which effectively re-duces the time and space complexity.Meanwhile,a new regularization is designed to further im-prove the recommendation performance.The proposed method effectively solves the integration problem of multiple types of features and the training problem with high-dimensional feature in-teraction structure.Experiments based on real datasets show the effectiveness of the multi-view feature fusion based service recommendation method.Deep learning enables automatic feature learning of heterogeneous data,mapping different data into the same latent space to obtain a uniform representations.To apply the powerful ability of learning representation to heterogeneous service data,we propose a neural network based method for service recommendation.The method constructs a hierarchical neural network with two lay-ers of attention mechanism to model the interactions between multi-view features to predict users’preferences on services.The neural network includes:a pair-wise view interaction layer modeling the interactions between features from different views,which can effectively alleviate the redun-dant information generated by the feature interactions in the same view,a feature-level attention mechanism designed to learn more accurate view representations,and a view-level attention mech-anism proposed to learn the contributions of different view interactions.Through the synchronous training of the neural network,the parameter of the nonlinear model are learned to select the infor-mative features and feature interactions.Experiments based a real dataset prove the effectiveness of the proposed method.