Research On Improved POI Recommendation Model Based On Gating And Mixture Of Experts Model

With the development of economy and society,Location-based Social Network(LBSN)has a profound impact on people’s lives.This network has accumulated massive data through user check-in,and the Point of Interest(POI)information contained in the data has become a hot research topic at present,which has been paid more and more attention by scholars.POI recommendation system can make recommendations according to the user’s historical sign-in data.However,because the user’s sign-in data contains a lot of irrelevant information,redundant parameters are input and the information utilization rate is low during model training,and the output recommendation results are not accurate and focused.In view of the above problems,Firstly,this paper deeply studies the application of traditional recommendation model and recommendation model based on deep learning in POI recommendation system,Then,we focus on one of the POI recommendation models-DAN-SNR,which uses self-attention mechanism to model POI sequence influence and user social influence,and establishes two parallel channels to capture short-term user preference and long-term user preference and social influence respectively.Two innovations are put forward for this model: first,a G-DAN-SNR model based on gating mechanism is constructed,which introduces gating mechanism before the fusion of long-term and short-term interest preference features of DAN-SNR model through self-attention mechanism,and adds Kalman algorithm to the gating mechanism to filter out the optimal features by weighting feature parameters;Secondly,MG-DAN-SNR model--Aiming at the POI single input and single output problem of DAN-SNR model,on the basis of the above-mentioned improved G-DAN-SNR model,the prediction layer is added into the Mixture of Experts(MOE)to optimize the prediction results again.Through simulation experiments,the two improved models proposed in this paper can effectively predict and recommend POI.Compared with the relevant reference models,the accuracy of the new model has been further improved,indicating the effectiveness of the two improved models.Finally,this paper implements the application of the two improved models in POI recommendation system through VUE and My SQL database.