Recommendation Algorithm Of The Next Point Of Interest Based On Long Short-term Memory Network And User Behavior Preference

Location-based social networks(LBSNs)are already closely related to our lives.Users of the website can check-in through mobile network devices or location-based services to record their current location information.Users will hope that the LBSNs website will recommend points of interest(POIs)which they may be interested in based on their past check-in experience to them.The next POI recommendation aims to analyze the user based on the user’s current location and the user’s past check-in data,so as to recommend the next point of interest that the user may be interested in.The research on the user’s next POI recommendation has attracted the interest of many researchers,but many early POI recommendation algorithms only considered the influence of geographical location,but did not take into account the influence of spacetime factors.There are many early POI recommendation algorithms only consider the user’s long-term preferences,but do not consider that the influence of the long-term trajectory on the user’s current preference may be small.In recent years,recommendation algorithms based on deep learning recurrent neural network(RNN)and its variants: long short-term memory network(LSTM)and gated recurrent unit(GRU)have shown good performance in solving the problem of next POI recommendation.In order to better solve the problem of next POI recommendation,this paper proposes a POI recommendation algorithm that uses RNN and LSTM,and integrates attention mechanism to model user preferences to achieve a better recommendation result.Specifically,this model uses a combination of long-term and short-term preferences structure.The model is composed of a network that models long-term preferences of users and a network that models short-term preferences.Three prediction tasks are added to the long-term model to improve prediction accuracy,which are the time influence module,geographical influence module and category and evaluation module.At the same time,the recommendation algorithm proposed in this paper also adds an attention mechanism to analyze the user’s historical preference,taking into account the evaluation of POI and the user’s own evaluation of POI categories.The impact of user preferences on user choices,and the structure of modeling the user’s long-term and short-term preferences separately and combining them also greatly improves the accuracy of recommendation.In this paper,four recommendation algorithms that perform well in POI recommendation are used as base-lines,and the short-term preference module is removed as the control group of the ablation experiment.After several rounds of experimental results,it is shown that the Long and Short term Preference Model combined with Contextual information(LSPMC)proposed in this paper is based on the data sets of Foursquare and Yelp websites,and the evaluation indicators of several commonly used POI recommendations,such as Precision and Recall are better than other comparison algorithms.In terms of Recall,LSPMC has a 20%-50% improvement compared to other comparative algorithms.