Research On Urban Spatial Perception Based On Multimodal Trajectory Data

In today's society,various intelligent devices contribute a lot of trajectory data,such as vehicle GPS,mobile phones,bus cards and so on.These trajectory data contains abundant human behavior information,and can be used in many applications through multi-modal trajectory data mining.The user's trajectory data contains rich semantic information,which reflects the changes of city to a certain extent.Mining user's trajectory data plays an increasingly important role in personalized recommendation,urban planning,security and other fields.The researches on trajectory data have attracted more and more attention from relevant scholars.With the development of urbanization,user trajectory data is increasing.How to mine useful information from massive trajectory data faces enormous challenges! And human trajectories are diverse,how to explore human social relationships and mobile models from it faces many problems! Up to now,some achievements have been made in this field at home and abroad.The main contents of this research are mobile model clustering,abnormal trajectory data detection,trajectory prediction and so on.A large part of the researches focus on online social networks and single-source trajectory data.There are few researches on capturing social relationships and mining multi-modal trajectory data.And building human mobile models in an accurate manner remains a challenging issue.Based on the above problems,this paper proposes a user social relationship learning and urban physical space learning framework,and reconstructs the user's offline social structure and urban physical space structure.The main work is as follows:Firstly,in terms of user social space perception,this paper uses WiFi technology to passively locate and collect user's trajectory data.To deal with the stability of wireless signals,this paper proposes a solution using sliding windows.And the paper identifies users with MAC addresses,filtering mechanism is proposed to efficiently identify users,analyzes the impact of different numbers of WiFi detectors on positioning accuracy.In addition,this paper proposes a method to detect user's cooperative location and form user's social context information,this method can get more detailed social information and accurate interactive information of users at the individual level.Then,this paper uses the skip-gram model to learn the context information containing rich user social relationships,modifies the negative sampling strategy,reconstructs the user social structure using the principle of cosine similarity,uses t-SNE to project high-dimensional word vectors to 2D space for offline social relationship interpretation.The paper also defines social distance between users to visually check the social structure,defines the user's social group to verify the user's social structure using DBSCAN density clustering and F-Measure algorithm..Secondly,Cellphone CDR Data,Smartcard Data,Taxicab GPS data,and Bus GPS data of the Chinese City Shenzhen are used to study urban physical space.To express better the local differences in point distribution patterns,this paper divides Shenzhen into thousands of blocks according to the length of 500 m.And trajectory data is used to explore the human mobile model every one-hour slot of the day after de-duplication,sorting,denoising,etc.The paper establishes a human mobility model by urban region pairs with high transit mobility for multi-source data using the PNPoly algorithm.Based on the dependence and spatial correlation between the trajectory data in the physical space,POIRank proposed in the paper quantifies the influence of different functional areas on cities.Experiments show that hotspot areas generated by POIRank algorithm conform to POIs' classification in the physical world according to Urban Master Planning of Shenzhen(2010-2020).And this paper proposes the novel notion of urban physical space representation learning based on the POI pairs co-context events detected to build human mobile model between POIs at different time of the day,mine similar POIs and cluster and evaluate the experimental results.The experimental results show that this paper can effectively reconstruct real user social structure and physical space structure,social relationship interpretation and physical space interpretation can be accurately expressed.