| With the rapid development of global positioning technology and the popularization of mobile intelligent terminals,Location-based Social Networks(LBSNs)have gradually emerged and become important platforms for information dissemination,socializing,and sharing knowledge and experience.Most of the existing studies focus on hometown POI recommendation.they integrate different factors such as temporal and spatial factors,social networking factors and content factors to model user trajectory and preference mining.Compared with the recommendation of local interest points,the check-in data of users in out-of-town cities is extremely sparse or even completely absent(i.e.,the cold start problem),which brings great challenges to the analysis of user behavior and mining of user interest preferences.In order to solve the problems of data sparseness,cold start and local accessibility in cross-city POI recommendation tasks,this study proposes a cross-city POI recommendation model based on user preference migration.The main work of the paper is as follows:1.In order to solve the problems of data sparseness and cold start,this study constructs a user preference transfer module based on a two-layer attention mechanism.This paper builds a double-channel input in this module.First,the rich check-in records of a specific user in the source city and the check-in records of his similar users in the target city are used as input data.Then,self-attention fusion is performed on the check-in records within the city,and cross-attention fusion is performed on the check-in records between cities to obtain a user feature representation vector that integrates the user’s long-term interest preference(user’s own inherent preference)and short-term visiting interest(target city characteristics).In addition,this module designs the transfer loss function to make the distribution characteristics of users’ interests in the two cities to be consistent,so as to achieve the purpose of user preference migration.2.To solve the local accessibility problem in cross-city POI recommendation,this study constructs a location-aware POI popularity model based on graph attention network.First,the structure of the POI network is constructed according to the embedded representation and geographic location of the POI.Then the POI information is updated through the Graph Attention Network.The final output is the updated POI feature representation matrix.3.By combining the user feature representation vector and the target city POI feature representation vector through the operation of vector dot product,the user’s POI visiting probability in the target city is obtained,and then the cross-city Top-k POI recommendation is realized.4.This study conducts model training and evaluation experiments on the LBSN Yelp public dataset,and also designs related comparison experiments,ablation experiments,parametric analysis experiments and generalization performance verification experiments.This research also carries out study of specific case.The experiments show that the model designed in this paper achieves the best results on the two Top-K evaluation indicators(HR and NDCG),which verifies the performance of this model.Based on the above research,this paper designs a cross-city POI recommendation system based on user preference migration. |
