Research On Point Of Interest Recommendation Algorithm Based On Multi-dimension Scenario Information

Users are prone to the problem of "information overload" because they cannot quickly get the right information.Recommendation system is a very popular technology which can provide users with travel destination suggestions.it can not only help users make quick decisions and reduce choice anxiety,but also has great commercial value in advertising and business decisions.Traditional POI recommendation systems tend to delineate the user’s activity range when tapping into the influence of geographic factors,which makes it almost impossible for content outside the activity range to be recommended.This paper propose a novel user similarity calculation method that uses Gaussian distribution to model the hotspot areas of user activities.Second,we design a dynamic division of similar user groups method to maximize the influence of users’ social relationships on user’s POI preferences.Finally,we propose a region transfer recommendation algorithm for generating POI recommendation lists.It cleverly combines geographic and social influences in POI recommendation to improve the accuracy of POI recommendation and the diversity of POI recommendation.The next POI recommendation recommend the next place.Compared with traditional recommendation,it is more complex because it needs to consider multiple contextual information trends at the same time.Most of the existing research treats the next POI recommendation as a sequential prediction problem and ignores the graph structure information that is naturally present in the POI data.Our method constructs a check-in transfer graph,a distance graph,a category transfer graph,and a user social relationship graph from the user’s historical check-in data.This modeling approach can make full use of the natural graph structure information in location-based social network data,and provides four different perspectives for information mining in graph convolutional network.Transformer has excellent performance in mining users’ longterm and short-term check-in preferences.We use a multi-layer perceptron to decode the information delivered by the Transformer encoder and output the results.