Research On Point Of Interest Recommendation Method Based On Recurrent Neural Network

Point of interest recommendation plays an important role in location-based social network services.It can not only help users find attractive points of interest in massive amounts of contextual information,meet their personalized needs,but also dig out user behavior habits for businesses.So as to bring huge economic benefits to businesses.Current research on point-of-interest recommendation focuses on situational information such as geographic location,social relationships,and POI categories to improve the accuracy of recommendations.However,in the actual recommendation process,the original data set recommended by POI is very sparse,and context information cannot be fully utilized,resulting in poor performance of POI recommendation.In addition,with the rapid development of the national economy and communication technology,users' personal preferences have become one of the main influencing factors of POI recommendations.The current research on user behavior and habits is mainly based on traditional recommendation methods such as collaborative filtering and matrix decomposition.When the method is used to model user behavior,user preferences are not fully captured,personalization is not prominent,and model efficiency is low,which severely reduces the accuracy and personalization of POI recommendations.Therefore,how to mine the potential preferences of users based on multiple heterogeneous types of contextual information,and integrate contextual information such as geographic location,POI category,and user check-in data into the POI recommendation model,is of great importance to improving the accuracy and personalization of POI recommendation.Important impact.In response to the above problems,this paper mines the POI context information based on the powerful nonlinear learning ability of the recurrent neural network,and uses the attention mechanism to model the user's long-term and short-term preferences to improve the accuracy and personalization of POI recommendations,So as to provide people with efficient and real-time smart services.The main research contents of this paper are as follows:A POI recommendation method fused with contextual information is proposed.In order to dig deeper into the contextual information of POI,a POI static feature extraction method based on symmetric matrix decomposition was designed to capture the geographic location and POI category features in LBSN,and the improved CBOW(Continuous Bags-of-Words)model was used to extract user comments.The semantic features in the information realize the hidden vector representation of POI in geographical,category,semantic and temporal feature space.Using the improved structure of the recurrent neural network,the gated recurrent unit learns the nonlinear interaction between different feature vectors of the POI,reproduces the scene when the user visits the POI,and improves the accuracy of the POI recommendation.Propose a POI personalized recommendation method that introduces attention mechanism.Based on the POI recommendation method that integrates contextual information,the long-and short-term attention mechanism is introduced to construct a user preference feature extraction model.Based on the user's historical POI check-in sequence,this model uses similarity calculation and a high-order Markov chain model to model the sequence relationship between check-in POIs.Then,the long-and short-term attention network is used to adaptively model the user's long-term and short-term preferences,and capture the sequence relationship in the user's sign-in data,so as to realize the personalized recommendation of POI.A large number of experiments are carried out on public data sets at home and abroad to verify the effectiveness of the method proposed in this paper.By analyzing the parameters of the model in this paper,the optimal model parameters are determined.The accuracy and recall rate of this method are compared with other mainstream POI recommendation methods on the data set.The experimental results show that the method in this paper has better recommendation performance.