Research On Personalized POI Recommendation Method Based On Knowledge Graph

The recommendation of Points of Interest is closely related to people's trips,it can provide a good choice for a trip,and the check-in sharing of Points of Interest is also popular content on social networks.Introducing knowledge graphs into recommendation systems can effectively improve the accuracy of the recommendation result.For now,Points of Interest recommendation systems still have problems with low recommendation accuracy,poor metapath design of knowledge graph,and insufficient fusion of user features.In response to these problems,the main research contents and innovations of this thesis include:(1)Aiming at the problem of low accuracy of Points of Interest recommendation,this thesis proposes a knowledge graph-based point of interest recommendation system framework,which aims to improve the feature extraction ability,and improve the accuracy of prediction.The system framework mainly includes preprocessing,training,recall,and prediction.The training section includes two parts: upstream training and downstream training.In the upstream part,a knowledge graph-based weighted multi-metapath learning method is proposed.In the downstream link,a long and short-term memory network with temporal and frequency gates is proposed,and combined with a graph attention network-based friend preference propagation.Finally,get recall and make recommendations based on the attributes of the Points of Interest.Through experiments on the Gowalla dataset,this thesis verifies the recommendation effect of the framework,compared with the classic Points of Interest recommendation model,this recommendation framework promotes both the recall rate and the precision rate.(2)Aiming at the problem of metapath design of knowledge graph,this thesis proposes a knowledge graph-based weighted multi-metapath learning method.The algorithm utilizes the Metapath2 Vec method to perform contextual prediction on the node sequences through metapath design,train the embedding vector of the knowledge graph,and give different weights to different metapaths in the training.By comparing the existing weightless algorithms through experiments,the weighted multi-metapath learning method proposed in this paper shows a better recall rate and accuracy rate,and improved the training effect of the knowledge graph.(3)Aiming at the problem of insufficient fusion of user features,this thesis proposes a long and short-term memory network with temporal and frequency gates based on user checkin features,and combine with a graph attention network-based friend preference propagation,extract features from multi-aspect.In the long-term and short-term network,utilizing temporal and frequency gates to calculate the time interval of user check-in and unit time check-in frequency,learning the time and frequency features of users.For graph neural networks,based on their similarity with social networks,cross-propagation of preferences among friends enhances preference representation.Through experiments,the results show that the algorithm proposed in this paper can capture user check-in features,promote the effect of preference propagation among users,and improve related indicators.Aiming at the problem of point of interest recommendation,this paper proposes a corresponding recommendation framework and algorithm,which can well solve the problems faced by the existing framework and algorithm,and can provide a certain reference for the related fields of recommendation systems.At the same time,the relevant research results on the recommendation of points of interest can also bring certain social effects to the system practice.