Research On Recommendation Methods Based On Heterogeneous Network Representation Learning

With the rise of location-based social network,it is more and more easy to obtain location information,which makes the increasing amount of data cause serious information overload.In the complex data,it is difficult for users to make a quick choice.In order to alleviate the choice difficulties faced by users,location-based personalized point-of-interest recommendation has attracted more and more attention.Compared with traditional recommendation,location-based recommendation adds user's location information,which is not only valuable information but also a restriction of user's activities.It has more environmental awareness and needs to be more personalized.Due to the limitation of the user's location,the sign in matrix is extremely sparse,which brings severe challenges to the traditional recommendation technology,and it is difficult to provide users with high-quality recommendations.At the same time,the location-based social network is a typical heterogeneous network,and the existing research is mainly focused on a single type of node for the study of homogeneous network,which will lose a lot of valuable information in heterogeneous network.In order to alleviate the problems of sparse check-in data,heterogeneity and low recommendation accuracy of traditional algorithms,this paper discusses the method of interest points recommendation based on heterogeneous network representation learning.First of all,this paper proposes a location-based heterogeneous random walk point-of-interest recommendation algorithm for the location network's sign in sparsity,heterogeneity and network representation learning technology.The heterogeneous information network is constructed by fully considering the internal relations among the nodes in the location network.The feature representation of learning nodes can effectively alleviate the problem of data sparsity.At the same time,the heterogeneous random walk strategy can effectively solve the problem of node heterogeneity in the network.According to the feature representation of the node,the user's preference list of interest points is generated and recommended.Then,by analyzing the impact of sign in attribute information such as time and semantics on user's sign in behavior,the above method is improved,and a heterogeneous random walk algorithm based on meta path is proposed.Adding attribute nodes to the construction of heterogeneous information network can better describe the real world.The introduction of meta path in the swimming strategy can better constrain the direction of the walk,and make it more practical.Finally,the proposed method is tested and analyzed on foursquare and Gowalla datasets to verify the effectiveness of the algorithm and improve the quality of recommendation.