| The popularity of mobile devices such as mobile phones,the use of global positioning systems such as GPS and Beidou satellite navigation,and the rapid development of Web 2.0 technology have enabled millions of users to share their locations on location-based social networks.The Point-of-Interest recommendation system is the core function of these location-based social network platforms,and how to provide accurate and effective Point-of-Interest recommendations for users has become an urgent problem to be solved.The existing Point-of-Interest recommendation algorithms mainly have the following problems: spatio-temporal feature embedding is prone to sparse encoding;The spatiotemporal connection between non adjacent access points of interest is ignored;The issue of user interest drift has not been taken seriously;The cold start problem limits the performance of most algorithm models.In response to the above issues,this article proposes two Point-of-Interest recommendation algorithm models to improve existing Point-ofInterest algorithms,and designs an Point-of-Interest recommendation system to carry out interaction between users and the system.The main research content of this article is as follows:(1)Aiming at the problem of interest drift,this paper proposes a point-of-interest recommendation algorithm model based on spatio-temporal attention,using graph neural network to connect the weight information between points of interest to explore the short-term intention of the user,so as to deal with the drift of user interest.question.This paper also uses a discretization embedding module to interpolate and embed spatiotemporal feature information to solve the problem of sparse coding;use bidirectional attention to learn and capture the spatiotemporal correlation between non-adjacent interest points;propose a temporal preference matching The mechanism can achieve the effect of preliminary pruning after a simple calculation,and effectively deal with the impact of the increase in the number of classifications.In this paper,a comparative experiment was conducted with the baseline model on multiple data sets,and the results showed that the recall rate of the model increased by 12% to46%,and there was a good improvement in the input sequence of interest points of different lengths.(2)This paper designs a dialogue-based two-channel point-of-interest recommendation system enhancement model for the cold start problem,and uses conversational recommendation system and clustering to fill user sequences with insufficient sequence length,replacing the traditional zero-fill method.The results show that predicting the filled sequence can increase the recall rate by about 2%.Afterwards,the ablation experiment proves that the dialogue mechanism and the clustering module have a certain effect on the performance of the original model.(3)This paper also designs a system based on a map development platform.After the system is implemented,the operation shows that the system performs well,can collect user trajectory information,and extract user behavior characteristics to recommend the next point of interest for the user. |
