Research On Sequence POI Recommender System Based On Attention Mechanism

In location-based social network(LBSN),people can publish geographic location information by check-in and use recommender services.Therefore,LBSN develops rapidly.Among them,Point of Interest(POI)recommendation as a key point in LBSN has attracted extensive research.POI recommends analyzing users' personalized preferences through historical check-in sequences.It can recommend suitable POIs for users.In real life,the user's check-in behavior usually occurs in a sequence.The recommender system recommends the next possible POI for the user through the user's historical check-in record.However,in the user's historical behavior data,not all behaviors can provide a role for future recommendations.The data weights that can affect the next behavior are also different.Moreover,the time-space interval information in the user's check-in sequence may provide a key factor for the user's next behavior.However,the existing POI recommender algorithms do not consider the effective use of information about the time and space interval between user check-in locations.It can not accurately express user preferences.In view of the above problems,this paper conducts the following research:(1)An attention-integrated matrix factorization POI recommender algorithm is proposed.On the basis of matrix decomposition,user attention is added to better analyze the user's attention value for different locations.Thereby,the user's preference can be acquired more accurately and the recommendation can be made more accurately.The attention mechanism can assign different weights to user check-in locations and effectively capture the relationship between user check-in locations.Experimental analysis is carried out on the real data set.The experimental results show that the model outperforms other comparison algorithms.(2)A POI sequence recommender algorithm integrating spatiotemporal network and self-attention mechanism is proposed.It incorporates the temporal and spatial interval information between user check-in information into a network of gated recurrent units.A self-attention mechanism is used to assign weights to check-in locations to obtain the user's weight sequence.Finally,the POI is matched by the time interval and spatial interval between the check-in location and the candidate location.It recommends a POI sequence containing three consecutive locations to the user.Tested and validated on real datasets,the experimental results show that the model outperforms previously proposed state-of-the-art models.(3)A point of interest recommender system platform is designed.Function development is carried out according to different needs of users.For users who need POI sequence recommendation,use the POI sequence recommender model that integrates spatiotemporal network and self-attention mechanism to recommend.Use the matrix factorization POI recommender model of fusion attention to recommend users who are recommended by ordinary POI.In this way,it can accurately recommend POIs to users to meet the personalized needs of users.