Embedded Learning Recommendation And Point Of Interest Prediction In Heterogeneous Information Networks

In the field of information filtering system application,recommendation system can be regarded as a major direction of research and development,which plays an increasingly great role in improving user experience and corporate efficiency.The mainstream business recommendation mainly adopts matrix decomposition mode and deep learning mode,and business promotion methods based on memory network and integrated learning have become popular applications in recent years.Many areas are widely exist in heterogeneous information,detailed describe the complex link between different subjects,and heterogeneous information network(HIN)as a modern large-scale commercial recommender systems involved in the heterogeneous data types of a natural and common data representation,can help effectively hidden in the data mining model,regularity,opens a new direction for the personalized recommendation system.The major existing problems of recommendation system are:recommended structure similar,it is difficult to say Suggestions,the high-level semantics as well as the integration of heterogeneous information application and most recommended practice follow the route of similar nodes associated connections,difficult to get hidden in the structure of the user and the project information,interest recommended results do not conform to the current location information.The main contributions of this paper are as follows:1)Considering that most of the similar type shortcomings of recommendation results are due to excessive reliance on similarity or strong connection,a heterogeneous social network recommendation algorithm is proposed at the level of weak relationship.First,set a minimum value as the trust lower limit between users,remove the node connection less than this value;Secondly,a full relationship heterogeneous information network(UI-HIN)is established,which contains three types of relationship: user-user,project-project and user-project.Finally,the optimal trust path selection algorithm(BTP)is proposed to select from the UI-HIN path to ensure that unexpected projects can be obtained based on the targeted level of projects and the diversity of project types.2)In order to solve the problem of cold start caused by sparse data,a high order embedded learning framework of heterogeneous information network was constructed.Firstly,the BTP algorithm was used to construct the full relation heterogeneous information network,and the importance factor of multi-task shared feature based on polygraph attention was defined.The semantic information of nodes was screened out,and the network nodes were effectively represented by the interaction structure.Secondly,the characteristics of the sampled nodes are aggregated to the nodes to be predicted by the hierarchical propagation mechanism of heterogeneous information network.Finally,an influence recommendation algorithm for high order information representation model is proposed to implement recommendation.It effectively improves the accuracy of rating prediction,as well as the pertinence,novelty and diversity of recommendation generation,especially in the application scenarios with sparse data,and has a good recommendation effect.3)To solve the problem that the sparsity of user check-in data in time and space leads to the homogeneity of recommendation results and fails to meet current needs due to location switching,firstly,the high-order embedded learning framework of heterogeneous information network is used to calculate the distance of interest points--geographical social attributes.Secondly,the prediction model of user preference activity range is established on time scale.Finally,the model of interest point--geographic natural attribute is established,which contains the information of interest point’s geographical location,surrounding,consumption level,representative project,comment heat and so on,so as to recommend users’ actual location on the basis of guaranteeing the type of interest point.4)Personalized movie recommendation system is designed based on heterogeneous information network high order embedded learning framework.According to the social network of each user to customize the required movie list,users can search by themselves or by the system analysis to generate recommendations,and in the system,also implemented recommendations based on different algorithms,and the effect evaluation;In addition,users can regard users with similar interests as friends of the system,and the system will recommend users according to their characteristics,so as to expand the selection range of recommendations and avoid repetition and singleness of recommendations.