| With the rapid development of location-based services,location information is accumulated as a potential resource.With the rapid updating of the Internet,social networks have become a platform for users to publish and exchange information.The combination of social networks and location-based services is called location-based social networking(LBSN).In a location-based social network,users publish location information on social networks in the form of check-ins.The sign-in form of the social network anytime and anywhere effectively reduces the distance between the user's virtual world and the real world,and is greatly favored by users.The long-term signing behavior of users leaves a wealth of historical location information on social networks.By collecting user history check-in information and data mining,users can be aware of the user's behavior patterns and user location preferences,and can provide users with better location-based location.Services such as smart transportation,quick dating and more.At the same time,in a location-based real-time advertising system,the merchant can decide which type of advertisement to serve for the user by predicting the location of the user.This thesis applies the check-in data of social network user history to research the problem of user location prediction.The main research results are as follows:1.This thesis has analyzed the user's time-based sign-in rule,illustrated the necessity of sub-trajectory by example,and proposed a user trajectory similarity measure method based on POI embedding.A sign-in of a social network user is regarded as a word,and a sign-in sequence of a user's sub-trajectory is regarded as a sentence.Through neural network model training,a POI is embedded and expressed as a real-valued vector.The sub-trajectory vector is further represented,and then the user sub-trajectory similarity is measured.2.For the problems of RNN and LSTM modeling user check-in sequences,GRU is used to model the user's check-in sequence,and at the same time,the time,place,and user are integrated,and the STU-GRU position prediction model is proposed.At the model embedding layer,the time encoding scheme is improved.Taking into account the impact of social relationships on location prediction,social relationships are redefined,and the position prediction of social relationships based on the STU-GRU model is proposed.At the same time,taking into account the possibility of novelty exploration by users,the next position is regarded as a binary problem,and neural network position prediction and fusion social position prediction are integrated into a unified model.Finally,through experiments,the proposed method improves the position prediction accuracy. |
