A Research Of Sequential POI Recommendation Based On User Preferences

In recent years,the rapid development of mobile Internet and the widespread popularity of smart devices have led to unprecedented development of location-based social networks.Location based social networks can link virtual communities with the real world,so that users can share their favorite locations on the Internet through mobile devices,helping other users discover interesting locations.Therefore,the point-of-interest recommendation system has become a hot research direction.The point-of-interest recommendation system mainly uses the user's historical check-in data to obtain travel preferences.However,in the existing research on POI recommendation,users' travel preferences are often regarded as static rather than dynamic.In addition,previous studies only considered the user's own check-in data when learning travel preferences,and ignored the influence of the user's unvisited POI.Therefore,the traditional POI recommendation model often has limitations when portraying the impact of users' historical check-in records on future travel.In response to the above problems,we will separately propose a next POI recommendation algorithm that integrates dynamic spatio-temporal preferences and a next POI recommendation algorithm based on user's local and global preferences.The main content and innovations of this article are as follows:1)The research focuses on users' dynamically changing travel preferences,and proposes a POI recommendation model ST-NHP that can integrate users' dynamic spatiotemporal preferences.In order to model the travel preferences of users over time,we propose a novel self-attentive continuous-time POI recommendation model for capturing the evolving demands of users over time.The model first learns spatial preferences through the user's history check-in data and the corresponding geographic area,then uses the attention mechanism to capture the user's periodic preference for POI.In the ST-NHP model,the user's dynamic preferences in space and time are simultaneously integrated to make the final point of interest recommendation.Finally,experiments on two public datasets can verify the effectiveness of the ST-NHP model.2)In order to obtain the user's personalized local preferences and global spatiotemporal preferences,we propose a multi-dimensional graph attention network model LSTU-GAT.The model can learn local personalized preferences from the user's own historical check-in data,and explore the influence of the user's unvisited location from the global spatio-temporal graph network,while allowing users to selectively learn from their social friends.Finally,we conduct extensive experiments on tow real-world datasets to demonstrate that LSTU-GAT is effective for next POI recommendation problems.