Research On Point-of-Interest Recommendation Based On Deep Learning

With the development of location-based social networks(LBSNs),Point-of-Interest(POI)recommendation has attracted lots of attention.Based on the LBSNs,users are able to share their relevant visiting experience via check-in records.Based on the previous check-ins of a user,POI recommendation systems can give the user a list of locations that the user may be interested in.The check-ins of in the POI recommendation systems are mostly implicit feedback,so it is difficult for the recommendation system to directly mine user preferences from the data.The most of the existing studies focus on recommending POIs to users based on their recent check-ins.However,the recent check-ins may contain some daily check-ins that users are not really interested in.If a model treats the recent check-ins equally,it is non-trivial to capture the actual preference of users.Last but not least,the check-ins in the POI recommendation system is highly sparse and sequential.Due to its great ability of nonlinear expression and the ability to handle multi-source heterogeneous data,Deep learning achieves success in many fields,including recommendation systems.Therefore,this paper focuses on solving related problems in the POI recommendation systems based on deep learning technology.The major contributions of our work are summarized as follows:(1)To address the issues of the implicit feedback and mining users' actual preferences from users' recent check-ins,this paper proposes an attention-based deep learning framework(ADPR),which consists of a latent representation method and a deep convolutional neural network employing the attention mechanism.To generate representations for users' personal preferences and POIs,this paper proposes a latent representation method incorporating the geographical influence and the categories of POIs.Then,this paper uses the generated latent representation of users and POIs as the input to the attention-based deep convolutional neural network for the task of POIs recommendations.This paper conducts the extensive experiments to compare ADPR with the state-of-the-art approaches,and the experimental results show the effectiveness of ADPR.(2)To address the sparsity and the issue of mining user preference from sequential check-ins,this paper proposes a POI recommendation system based on generative adversarial networks(POI-GAN).The POI recommendation system uses a generator to generate fake samples and uses adversarial learning to train the network,which can alleviate the impact of data sparsity.Both the generator and the discriminator use LSTM to learn the sequence of check-ins,and modify the user's current preference based on the sequence information.In addition,the network uses multi-task learning.During training,the generator also predicts the user's current location to incorporate the geographical influence.This paper conducts the extensive experiments to compare POI-GAN with the state-of-the-art approaches,and the experimental results show the effectiveness of POI-GAN.