Research On Point-of-Interest Recommendation Based On Representation Learning And Topic Transition Behavior Mining

With the development of Location Based Social Networks(LBSN),people often share their experiences about Point of Interests(POI)by "check-in".Location-based social networks contain a wealth of information,and it has become a hot research topic to analyze and mine the information to recommend potential POI for users.The current POI recommendation research has the following deficiencies in POI preference modeling:The traditional LBSN-based POI recommendation often does not consider its rich auxiliary information,and only the implicit feedback is used to make the data sparsity problem more obvious.The user check-in sequence association model based on Markov chain modeling mostly stays at a lower-level of POI,but the user's transfer behavior actually reflects its higher-level preference characteristics.The user check-in sequence based on Markov chain modeling mainly uses the first-order Markov chain.Actually,the subsequent stroke only depends on the current position,which reduces the performance of the model to a certain extent.In fact,the user's subsequent POI may be due to a historical check-in behavior not far from the current moment.The main work of this paper includes the following aspects:(1)This paper proposes to model the user's POI preference based on Heterogeneous Information Network(HIN)representation learning and attention mechanism.First,the use of heterogeneous information network modeling data aims to maximize the use of auxiliary data outside the user-POI,it makes the semantic information contained in the data more abundant.Then,the representation learning based on the heterogeneous information network can make the nodes in the network represented as real-valued vectors.Since different metapaths have different semantics,the attention mechanism in the neural network is used to more advanced fusion of the representation vectors,and finally the neural network is used to model the user's POI preferences.(2)This paper proposes a multi-order Markov model based on topic transfer to model user's transfer preference between POIs.First,we use the word vector representation learning method in natural language processing to obtain the embedded representation of POI.The obtained similar POI are in a similar position in the embedding space,and then we can cluster according to the semantic information and classify the POI into different topic.Then we can use the transfer records of individual users and all users at different POI to get the transfer probability between different topic.On this basis,the multi-order Markov chain model is used to model the transfer preferences of users.(3)In order to verify the effectiveness of the algorithm,the method proposed in this paper has conducted a lot of experimental research on the real check-in record data set.The experimental results show that the multi-order Markov method based on topic transfer and the heterogeneous network-based learning with attention mechanism can effectively capture the user's check-in behavior characteristics,and also can better model the user's transfer preference and POI preference.The experimental results also prove the effectiveness of the proposed method.