Next Point-of-interest Recommendation Based On Sequential Influence And Attention Mechanism

With the rapid development of location-based social networks,people’s choice of location has become more and more abundant,but it has also become more difficult to choose.As an important tool to help users make decisions and actions,the next point of interest(POI)recommendation research has great value,but the road to obtain these values is not smooth,and the next point of interest recommendation also faces many aspects question.On the one hand,user preferences are often determined by various factors such as time,space,timing,and user preferences are not only affected by short-term behavioral trajectories,but also by long-term habits and behavioral patterns that users develop.How to comprehensively consider the joint impact of the two is an urgent problem to be solved.On the other hand,the check-in data of most users is very sparse and the check-in frequency of users is inconsistent,which causes the next POI recommendation task to face data sparse and cold start problems,and most of the existing next POI recommendation methods are limited by the recurrent neural network structure,there are gradient problems and long-term dependency problems.In view of the above limitations,this paper proposes two methods.Firstly,the existing research cannot comprehensively consider long-term and short-term preferences,cannot fully utilize the time,space and time sequence information in the sequence,and it is difficult to mine the potential behavior intention of users.Different trajectories and time slices are used for feature embedding,and the long-term and short-term preferences of users are modeled separately by calculating time similarity and constructing geographic matrices,so as to fully exploit the potential impact of long-term and short-term preferences on user decision-making.Explore the user’s potential behavioral intentions,eliminate the limitations of the original check-in sequence,and achieve the next point of interest recommendation that is more in line with the user’s needs.Secondly,in view of the common gradient problem,long-term dependency problem,data sparse problem and cold start problem in the existing next-point-of-interest recommendation models,Use the advantage of parallel computing of Temporal convolutional network(TCN)to learn user preference features and model users’ short-term preferences.At the same time,the attention mechanism is integrated to highlight the impact of key features on users’ long-term preferences.Through this combination method,the efficiency of model learning is improved,and the impact of data sparsity and cold start problems on user behavior prediction is effectively alleviated,so as to achieve a more accurate recommendation of the next point of interest.Finally,the two next-point-of-interest recommendation methods proposed in this paper are experimentally verified on the real and effective NYC dataset and TYK dataset,and extensive experimental comparison and analysis with existing methods are carried out to verify the effectiveness of the next point of interest recommendation method proposed in this paper.