Point-of-Interest Recommendation Based On User Check-in Sequence

In recent years,with the rapid growth of location-based social network(LBSN)services,such as Foursquare,Gowalla,Dianping,etc.,more and more users share their favorite Points-of-Interest(POI)through a large number of geographical check-in behaviors.These online check-in data provide an opportunity to analyze the users' check-in behavior patterns.We can analyze and predict where the user will go next based on their historical check-in sequences.In addition,through the analysis of the users' check-in data,the merchants can recommend content to users by capturing their preferences,and improve the users' experience.However,existing POI recommender systems face many challenges,such as the lack of ability to capture the complex correlations among POIs,cannot fully utilize various information in the check-in data and unable to effectively model users' preferences.As for the above problems,this article exploits the effective information in the users' check-in sequences,so as to efficiently recommend personalized POIs for users.The work of this article mainly includes the following two aspects:(1)For the general POI recommendation,this paper proposes a context and preference aware model(CPAM)to incorporate both contextual influence and users' preferences into POI recommendation.CPAM model includes two key modules.Firstly,this paper uses Skip-Gram based POI Embedding Model(SG-PEM)to capture the contextual influence among POIs and learn the context-aware vector representation of POIs from visiting sequences.Then we calculate the users' preference for the target POI based on the learned embeddings with similarity metric.Secondly,for the implicit feedback information contained in the check-in data,this paper devises the Logistic Matrix Factorization(LMF)algorithm to model the users' personalized preferences for POIs.Finally,we unify these two components as CPAM model to perform personalized recommendation by leveraging contextual influence and users' preferences.Experimental results on two real-world datasets,show that the proposed model outperforms state-of-the-art baselines.(2)For the next POI recommendation,this paper proposes a Long Short-Term Memory network model SG-ALSTM(Skip-Gram and Attention based LSTM).The SG-ALSTM model and learns the feature context-aware vector representation of POIs through the POI embedding model SG-PEM,and then uses the learned vector representation of the POIs as input to train the attention-based LSTM network.Thereby it can predict the users' next choice of POI accurately.Experiments on two public realworld check-in datasets show that the effectiveness of SG-ALSTM model.