Research On User Geographic Point-of-interest Recommendation Model Based On Deep Learning

In recent years,point-of-interest(POI)recommendation system has become a research hotspot in many recommendation systems.The research object of the recommendation system is user's POI check-in data in location-based social network(LBSN),its data source is rich and the amount of data is huge,so it has certain research significance.For the users of LBSN,POI recommendation system can mine and analyze the user's information and behavior data,gain an in-depth understanding of the user's preferences for check-in behavior,and encourage users to explore new locations and become familiar with the new cities.For the LBSN platform,the POI recommendation system can solve the problem of information overload caused by massive data,and help the owners(merchants or enterprises)of geographic interest points to discover potential users,so as to implement precise marketing strategies for users.From a social perspective,by analyzing the POI recommendation results,it can provide reference suggestions and open up research directions for exploring complex academic issues such as user behavior and the nature of social interaction.Therefore,from the perspective of user's check-in sequence behavior,this paper applies deep learning methods to the research of POI recommendation,and constructs a personalized POI recommendation model.The main work of this paper is as follows:1.Aiming at the problem of sparse check-in data,this paper analyzes the check-in data set and combines the interest point ID,geographic coordinates and POI category information in the data set to construct the POI collaboration matrix,position approximation matrix and category correlation matrix,which are used to capture POI association and alleviate the sparsity of check-in data.2.In order to solve the problem that the current research on check-in data information modeling method ignores the joint influence between multi-dimensional information and does not well represent the interest points and user characteristics,a POI and user association characteristics matrix based on collaborative filtering algorithm is proposed.It integrates coding information,geographic information,text information(POI category),social relationship information of interest points in the check-in data.On this basis,the hidden factors are obtained by matrix facotrization,and feature vectors of users and POIs are calculated to provide a general and effective representation of user and POI features for the recommendation model.3.To solve the problem that a single neural network model cannot simultaneously learn user's fine-grained interest preferences(user's long and short-term interest preferences),this paper proposes an MF-ADNN model.The Recurrent Neural Network(RNN)with attention mechanism is applied to the user's sequence behavior modeling,and different neural network models are constructed to learn the user's long-term and short-term interest features respectively,so as to solve the learning problem of interaction between multiple features and make sequence context feature representation more precise.At the same time,the training method of the model is determined by combining the time influencing factors,so as to predict the user's POI check-in results in different time periods.4.Design the model optimization experiment,use the pre-processed data to train the model,and determine the parameters that make the model performance reach the optimal through a series of hyper-parameter selection experiments.On this basis,the comparison experiment of POI recommendation model is designed,and the MF-ADNN model is compared with the five most representative models in the field of matrix factorization and deep learning POI recommendation algorithm,so as to analyze the experimental results to verify the effectiveness of MF-ADNN model.5.Based on the ionic framework,Baidu Map API and BI technology programming to implement the POI recommendation visualization prototype system,the recommendation results are displayed and analyzed.The system provides POI recommendation function for ordinary users,and provides the function of observing the future time POI checkin characteristics for LBSN platform.