Research On Next POI Recommender System Based On Deep Learning

With the popularity of the Internet and the maturity of geolocation technology,location-based social networks(LBSN)recommendation services have become a reality.In addition to economic development,tourism is favored by more and more people.Personalized POI(Point-of-Interest)recommendation based on user location has become a popular research topic.POI points of interest are recommended as an important branch of the location social network.Through data modeling,the user's historical behavior check-in records are analyzed,and the user's personalized preferences are learned to help the user find the most interesting personalized place.Point-of-interest recommendation is very different from traditional item recommendation,mainly manifested in sparse data;the user's check-in data samples are all positive samples,and the negative samples are lacking;user's interest preferences will change with time;the user's check-in points of interest will be at a certain time.There is spatial aggregation inside,so the recommendation of points of interest has great challenges.In response to the above problems,this paper proposes a new personalized Next POI recommendation framework based on deep neural networks.The framework first uses Embedding to convert the relevant information and attributes of users and locations into a vector form.This method allows the network to accurately learn the user's location and location information,avoiding the shortcomings of traditional matrix decomposition techniques.Then,through the LSTM neural network,the feature information of the position data is extracted,Finally,a user-location cross-attention mechanism is introduced to dynamically model the personalized time check-in sequence.which can dynamically obtain the user's preference for location.Aiming at the environment with context information,the fusion time,category and geographic location context information is proposed to improve the accuracy of recommendation.The main work of this article is as follows:1)This paper proposed and designed the Next POI recommendation model(L-attention)based on the LSTM user-place cross-attention mechanism,which used deep neural networks to dynamically model user preferences,successfully embedding,LSTM and cross Combining attention mechanisms.It solved the timing problem that the traditional recommendation method did not solve and the attention problem that was not considered under the deep recommendation.2)In this paper,based on the Next POI recommendation model based on the LSTM user-location cross-attention mechanism,the context information of the POI location was successfully fused,and an in-depth POI recommendation model incorporating context information was proposed.3)The effectiveness of the model in this paper was verified through a large number of experiments on the public Foursquare check-in dataset.First,compare the different deep recommendation algorithm models.The experimental results showed that the accuracy of the algorithm model in this paper was higher than the other algorithms under different sparsity.Secondly,by adjusting the different parameters of the model,this paper verified the influence of different parameters on the model,and finally by increasing the number of neural network layers to analyze the impact of the scale of the neural network on the algorithm.4)This article designed a user-based Next POI recommender system platform UNP(User Next POI).For users with clear requirements,use the non-personalized recommendation function module,and users without clear requirements,used the personalized recommendation requirement module,in which the personalized recommendation module incorporated the Next POI recommendation algorithm proposed in this paper.