A model for anonymous spatio-temporal trajectory recognition

Abstract

The application provides a model for anonymous spatio-temporal trajectory recognition, comprising: an original trajectory graph construction module, configured to construct an original trajectory graph according to trajectory data generated by anonymous users within a period of time, the original trajectory graph comprising a plurality of nodes representing a plurality of check-in records of the trajectory data and a plurality of edges representing relationships between the nodes, and to perform embedding representation on the plurality of nodes of the original trajectory graph to obtain spatial information of the plurality of nodes; a multi-scale trajectory graph construction module, configured to construct a plurality of trajectory graphs of different scales based on the original trajectory graph according to a plurality of preset sampling rates; a trajectory spatial information extraction module, configured to extract spatial features of each node of each trajectory graph in the plurality of trajectory graphs based on the spatial information of the plurality of nodes, aggregate the spatial features of each node of each trajectory graph in the plurality of trajectory graphs, and obtain multi-scale trajectory spatial features; and a trajectory recognition module, configured to recognize a probability value of the trajectory data belonging to each known user according to the trajectory spatial features.

Description

Technical Field

[0001] This invention relates to the field of neural networks, more specifically to the field of anonymous trajectory information recognition, and more specifically, to a model for anonymous spatiotemporal trajectory recognition. Background Technology

[0002] The most relevant research area in anonymized trajectory recognition technology is trajectory-user linking, which is a technique for connecting trajectories to the anonymous users who generated them. Currently, there are two main types of solutions for anonymized trajectory recognition: traditional trajectory similarity-based research and deep learning-based models.

[0003] Traditional trajectory similarity research typically relies on mathematical statistics methods to find spatial information statistical features in trajectories, such as Deep Bayesian Networks (DBNs), Hidden Markov Models (HMMs), and Longest Common Subsequences (LCSS). However, traditional approaches require researchers to leverage domain knowledge and feature engineering to differentiate the spatial information in trajectories before applying mathematical statistics to analyze this data. This type of approach often fails to effectively address the linear indistinguishability and noise issues encountered in trajectory representations within large trajectory datasets, and accuracy remains a common challenge in pairwise point matching similarity calculations.

[0004] Compared with statistical methods and machine learning techniques, deep learning has received widespread attention due to its excellent nonlinear modeling capabilities and in-depth analysis of spatial relationships. Deep learning methods for anonymous trajectory recognition can be divided into trajectory sequence methods and trajectory graph methods. For example, the trajectory sequence recognition method shown in reference [1] (the method in this reference is abbreviated as TULER) is inspired by word embedding in natural language processing. This type of method regards the geographical information in the trajectory as words and the entire trajectory as sentences, thereby embedding two-dimensional spatial information into high-dimensional vectors, so that the temporal dependency information in the trajectory sequence can be learned in the deep learning process. Following this work, the method represented by reference [2] (the method in this reference is abbreviated as TULVAE) is based on generative networks such as variational autoencoders. It uses variational inference to mine the potential distribution of the trajectory to optimize the representation of spatial information in the trajectory, and applies a semi-supervised learning method to allow unlabeled trajectory sequences to participate in the training process. Taking reference [3] (the method in this reference is abbreviated as STULIG) and reference [4] (the method in this reference is abbreviated as TGAN) as examples, both methods enhance the original training set based on the above. The difference between the two methods is that the former STULIG uses a variational autoencoder with hierarchical latent factors, while the latter TGAN is developed based on generative adversarial networks. However, models that use generative networks lack good parallel computing capabilities and often have high computational complexity. If generative tasks are used to improve the performance of classification tasks, it will also lead to unacceptable time costs.

[0005] Subsequently, trajectory sequence-type methods also adopted attention methods and focused on spatial information analysis according to the order of the trajectory sequence. Reference [5] (the method in this paper is abbreviated as DeepTUL) designed an attention module based on historical trajectories to obtain contextual information for understanding human movement patterns. Although the anonymous trajectory recognition technology that treats trajectories as sequences makes good use of the sequential information involved in the trajectory, the spatial information is not fully utilized. The recent reference [6] (the method in this paper is abbreviated as GraphSAGE) learns the neighbor features of each node. However, this method requires a fixed number of spatial information samples to aggregate spatial information, which inevitably leads to the neglect of spatial information or the introduction of redundant information. At the same time, the user trajectory data in the actual scenario is extremely sparse, which will hinder the effective representation of trajectory spatial information. Therefore, the existing methods all neglect spatial information to varying degrees.

[0006] Based on the above analysis, although existing deep learning models have achieved satisfactory results, the following issues still limit their performance:

[0007] (1) Ignoring spatial information: The model ignores the unique spatial information in the trajectory. That is, a large amount of current work simply treats the trajectory as a special time series, representing the geospatial information in the trajectory as sequential data only according to the order of time. This trajectory representation technology determines the upper limit of the accuracy of anonymous trajectory recognition from the outset. This obviously violates the understanding of complex human movement patterns and lags behind the research methods in the field of transportation. Furthermore, spatial information-based recognition also ignores some spatial information.

[0008] (2) Sampling leads to information loss: Due to the limitations of the model structure, the model needs to sample or fill the trajectory based on a fixed length in order to be able to represent the trajectory in a standardized way for input into the model. However, in the implementation scenario, the length of the trajectory is always variable, because users can generate short trajectories for short shopping trips or long trajectories for commuting.

[0009] Therefore, existing deep learning models for trajectory recognition have two main problems: first, they tend to ignore spatial information in the trajectory, resulting in a failure to understand human movement patterns; second, they rely on fixed trajectory lengths for recognition, leading to poor recognition performance and a lack of flexibility when faced with diverse trajectories.

[0010] It should be noted that the background information presented here is only for illustrating relevant information about the present invention to aid in understanding the technical solutions of the present invention, and does not imply that the relevant information is necessarily prior art. In the absence of evidence indicating that the relevant information was disclosed before the filing date of this invention, the relevant information should not be considered prior art.

[0011] References:

[0012] [1] Gao, Q., Zhou, F., Zhang, K., Trajcevski, G., Luo, X., Zhang, F.: Identifying human mobility via trajectory embeddings. In: IJCAI. vol. 17, pp.1689–1695 (2017).

[0013] [2] Zhou, F., Gao, Q., Trajcevski, G., Zhang, K., Zhong, T., Zhang, F.: Trajectory-user linking via variational autoencoder. In: IJCAI. pp. 3212–3218 (2018).

[0014] [3] Zhou, F., Liu,

[0015] [4] Zhou, F., Yin, R., Trajcevski, G., Zhang, K., Wu, J., Khokhar, AA, 2021c. Improving human mobility identification with trajectoryaugmentation. GeoInformatica 25, 453–483.

[0016] [5] Miao, C., Wang, J., Yu, H., Zhang, W., Qi, Y., 2020. Trajectory-user linking with attentive recurrent network, in: AAMAS '20, Auckland, New Zealand, May 9-13, 2020, International Foundation for Autonomous Agents andMultiagent Systems. pp. 878–886.

[0017] [6] Hamilton, WL, Ying, Z., Leskovec, J., 2017. Inductiverepresentation learning on large graphs, in: NIPS 2017, December 4-9, 2017, Long Beach, CA, USA, pp. 1024–1034. Summary of the Invention

[0018] Therefore, the purpose of this invention is to overcome the shortcomings of the prior art and provide a model for anonymous spatiotemporal trajectory recognition.

[0019] The objective of this invention is achieved through the following technical solution:

[0020] According to a first aspect of the present invention, a model for anonymous spatiotemporal trajectory identification is provided, the model comprising:

[0021] The original trajectory graph construction module is used to construct an original trajectory graph based on the trajectory data generated by anonymous users over a period of time. It includes multiple nodes representing multiple check-in records of trajectory data and multiple edges representing the relationships between nodes. The module also embeds multiple nodes of the original trajectory graph to obtain the spatial information of multiple nodes.

[0022] The multi-scale trajectory map construction module is used to sample nodes from the original trajectory map at multiple preset sampling rates to construct multiple trajectory maps at different scales.

[0023] The trajectory spatial information extraction module is used to extract the spatial features of each node in each trajectory graph based on the spatial information of multiple nodes, and to aggregate the spatial features of each node in each trajectory graph in multiple trajectory graphs to obtain multi-scale trajectory spatial features.

[0024] The trajectory recognition module is used to identify the probability value of the trajectory data belonging to each known user based on the multi-scale trajectory spatial features.

[0025] In some embodiments of the present invention, the trajectory spatial information extraction module includes:

[0026] Multiple graph attention networks are used, each graph attention network is used to determine the attention value between each node in each trajectory graph and any other node in all other trajectories based on the spatial information of the nodes in each trajectory graph, and update the initial features of each node according to the attention value;

[0027] The multi-head self-attention mechanism layer is used to determine the relation weight between each node and any other node in each trajectory graph based on the initial features of each node in each trajectory graph using the multi-head self-attention mechanism, and obtain the spatial features of each node in each trajectory graph based on the relation weight.

[0028] The global average pooling unit is used to perform global average pooling on the spatial features of each node of each trajectory graph in multiple trajectory graphs to obtain the graph space features of each trajectory graph. The graph space features of multiple trajectory graphs have the same dimension.

[0029] The aggregation unit is used to stitch together the graph space features of multiple trajectory maps to obtain multi-scale trajectory space features.

[0030] In some embodiments of the present invention, each of the plurality of check-in records includes a timestamp and a geographical location. The original trajectory map construction module constructs the original trajectory map by: obtaining multiple grid areas obtained by dividing the surface space into average sizes according to a preset geographical space; using a grid area where the geographical location of each check-in record in the trajectory data is located as a node to represent the check-in record; and connecting the grid areas corresponding to each check-in record sequentially to form multiple edges according to the chronological order of their timestamps in the trajectory data to obtain the original trajectory map.

[0031] In some embodiments of the present invention, the multi-scale trajectory map construction module includes:

[0032] The node sampling unit is used to sample nodes from the original trajectory map multiple times according to multiple preset sampling rates and a preset sampling method for each sampling rate, to obtain multiple node sets. Among them, the multiple sampling rates include sampling at 100%.

[0033] The edge construction unit is used to determine whether there is an edge relationship between any two nodes in each node set according to the preset edge construction rules, and to establish an edge between the two nodes with an edge relationship, so as to obtain a trajectory map of one scale under each sampling rate.

[0034] In some embodiments of the present invention, the edge construction unit includes a preset edge construction rule: when any two nodes meet either a preset time condition or a preset spatial distance condition, it is determined that there is an edge relationship between the two nodes. The preset time condition includes that the timestamps of the check-in records of the two nodes are adjacent, and the preset spatial distance condition includes that the distance between the geographical locations of the two nodes is less than a preset threshold.

[0035] In some embodiments of the present invention, the trajectory recognition module includes:

[0036] The trajectory embedding layer is used to process multi-scale trajectory spatial features using recurrent neural networks to obtain fused trajectory spatial features.

[0037] The linking unit is used to represent the fused trajectory space features using a multilayer perceptron to obtain the final trajectory space features, and then uses the Softmax function to obtain the probability value of the trajectory data belonging to each known user based on the final trajectory space features.

[0038] According to a second aspect of the present invention, a training method for a model for anonymous spatiotemporal trajectory recognition is provided. The training method includes: acquiring a training set comprising multiple samples, each sample comprising trajectory data generated by a known user over a period of time and a corresponding user label sequence, the user label sequence comprising multiple user labels, each user label indicating whether the corresponding trajectory data belongs to the known user; training the model described in the first aspect of the present invention using the training set, obtaining the probability value of the trajectory data belonging to each known user based on the input trajectory data, updating the model parameters with the optimization objective of minimizing the difference between the probability values ​​of each known user and the user label sequence, thereby obtaining the trained model.

[0039] According to a third aspect of the present invention, an anonymous spatiotemporal trajectory recognition method is provided, the method comprising: acquiring trajectory data generated by an anonymous user over a period of time; processing the trajectory data using a trained model obtained by the method described in the second aspect of the present invention to obtain probability values ​​of the trajectory data belonging to each known user; and identifying the known user to which the trajectory data belongs based on the probability values ​​of each known user.

[0040] According to a fourth aspect of the present invention, an electronic device is provided, comprising: one or more processors; and a memory for storing executable instructions; wherein the one or more processors are configured to implement the steps of the method described in the second or third aspect of the present invention by executing the executable instructions.

[0041] Compared with the prior art, the advantages of the present invention are as follows:

[0042] This invention's model samples the original trajectory map at multiple preset sampling rates, constructing multiple trajectory map representations for a single trajectory data point. Based on these multiple trajectory maps, spatial feature extraction and recognition are performed, improving the trajectory recognition accuracy of the model's algorithm. Furthermore, the trajectory data input to the model consists of multiple check-in records generated by anonymous users over a period of time. Since the trajectory length of different users over a period of time is always variable, the model can learn and recognize features from trajectories of arbitrary length, avoiding the loss of information features and improving the model's trajectory recognition capabilities in real-world scenarios as well as enhancing the flexibility of model applications. Attached Figure Description

[0043] The embodiments of the present invention will be further described below with reference to the accompanying drawings, wherein:

[0044] Figure 1 This is a schematic diagram of a model structure for anonymous spatiotemporal trajectory recognition according to an embodiment of the present invention;

[0045] Figure 2 This is a schematic diagram illustrating the data processing principle of the original trajectory map construction module according to an embodiment of the present invention;

[0046] Figure 3 This is a schematic diagram illustrating the principle of constructing multiple trajectory maps at different scales according to an embodiment of the present invention.

[0047] Figure 4 This is a schematic diagram illustrating the overall framework of the model and the complete process of processing trajectory data according to an embodiment of the present invention. Detailed Implementation

[0048] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are merely illustrative and are not intended to limit the invention.

[0049] As mentioned in the background section, existing deep learning models for trajectory recognition have two main drawbacks. First, they tend to overlook spatial information in the trajectory, resulting in a failure to adequately understand human movement patterns. Second, they rely on fixed trajectory lengths for recognition, leading to poor recognition performance and a lack of flexibility when faced with diverse trajectories.

[0050] To address the aforementioned problems, the inventors propose a model for anonymous spatiotemporal trajectory recognition. This model identifies the known users to whom a trajectory belongs based on trajectory data generated by anonymous users over a period of time. The inventors found that the causes of the first problem include: 1. Existing methods treat trajectories as time series for trajectory recognition; 2. In some methods based on spatial information for trajectory recognition, the highly random nature of user-submitted check-in records in real-world scenarios often results in sparsity among the check-in records in the acquired trajectory data, hindering the effective representation of trajectory spatial information. Therefore, these two points often prevent existing models from fully utilizing and learning spatial information during trajectory recognition, thus neglecting spatial information. To address the first problem, the model of this invention samples the original trajectory map at multiple preset sampling rates, constructing trajectory maps of multiple scales for a single trajectory data point. This allows for the full display of trajectory spatial information through multiple trajectory maps, and spatial feature extraction and recognition are performed based on these multiple trajectory maps, improving the model's trajectory recognition accuracy. Regarding the second issue, this invention does not standardize the trajectory data of the input model. That is, the trajectory data of the input model consists of multiple check-in records generated by anonymous users over a period of time. The trajectory length of different users over a period of time is always variable, which enables the model of this invention to perform feature learning and recognition on trajectories of arbitrary length, avoid the loss of information features, improve the trajectory recognition capability of the model in real-world scenarios, and enhance the flexibility of model application.

[0051] In summary, according to one embodiment of the present invention, see Figure 1This diagram illustrates the structure of a model for anonymous spatiotemporal trajectory recognition. The model shown comprises four parts: an original trajectory graph construction module, a multi-scale trajectory graph construction module, a trajectory spatial information extraction module, and a trajectory recognition module. The original trajectory graph construction module constructs an original trajectory graph based on trajectory data generated by anonymous users over a period of time. This graph includes multiple nodes representing multiple check-in records of the trajectory data and multiple edges representing relationships between nodes. The module embeds the spatial information of each node in the original trajectory graph. The multi-scale trajectory graph construction module samples nodes from the original trajectory graph at preset sampling rates to construct multiple trajectory graphs at different scales. The trajectory spatial information extraction module extracts spatial features from each node in each of the multiple trajectory graphs based on the spatial information of the multiple nodes, aggregating the spatial features of each node in each trajectory graph to obtain multi-scale trajectory spatial features. The trajectory recognition module identifies the probability value of the trajectory data belonging to each known user based on the multi-scale trajectory spatial features. To better understand this invention, specific embodiments are described in detail below regarding the model structure, model training, and application scenarios.

[0052] I. Model Structure

[0053] (1) Original trajectory map construction module

[0054] According to one embodiment of the present invention, before describing the original trajectory map construction module, the trajectory data of the input model is first described. The trajectory data consists of multiple check-in records generated by an anonymous user over a period of time, arranged in chronological order. Each check-in record is automatically generated in the background after the anonymous user opens the mobile application. Each check-in record includes at least the user's latitude and longitude and a timestamp. That is, each check-in record includes a timestamp and the corresponding geographical location expressed in latitude and longitude.

[0055] According to an embodiment of the present invention, in the original trajectory map construction module, the method of constructing the original trajectory map includes: obtaining multiple grid regions obtained by dividing the surface space into multiple grid regions of a preset geographic size; using a grid region where the geographic location of each check-in record in the trajectory data is located as a node to represent the check-in record; and connecting the grid regions corresponding to each check-in record sequentially to form multiple edges according to the chronological order of their timestamps in the trajectory data to obtain the original trajectory map. Specifically, when converting all check-in records of the overall trajectory data into grid regions for node representation, if the geographic locations of multiple check-in records are in the same grid region, that grid region can be used to represent multiple check-in records. The technical solution of this embodiment can achieve at least the following beneficial technical effects: First, the method of using divided grid regions to replace check-in records for representation merges check-in records with similar spatial locations, reducing the spatial size of the original spatial information representation; second, grid region representation is simpler and faster than other representation methods; and finally, directly using grid regions as nodes facilitates subsequent modeling and analysis of trajectory map structures at multiple different scales.

[0056] According to one embodiment of the present invention, the method of dividing the land surface space includes: dividing the land surface space equally according to a longitude of 0.01 and a latitude of 0.01 to obtain multiple grid areas; or dividing the land surface space equally according to a longitude of 0.02 and a latitude of 0.02. It should be understood that the size of the geographical space can be preset according to actual needs, for example, the division can also be based on a longitude of 0.03 and a latitude of 0.03, and the present invention is not limited thereto.

[0057] According to one embodiment of the present invention, pre-trained embedding representations are performed on the grid regions corresponding to all check-in records of the trajectory data, converting the grid regions into vectors. Since the grid regions are used as nodes, the unique attribute of a node is the vector corresponding to that grid region, and the vector serves as the spatial information of that node. When the geographical locations of multiple check-in records are in the same grid region, the nodes corresponding to the multiple check-in records are in the same grid region, that is, the nodes of the multiple check-in records have the same vector. The embedding representation can be word embedding, where the vector of each grid region is treated as a word vector. Illustratively, the CBOW model of Word2vec is used to embed the grid regions to obtain low-dimensional dense vectors. The technical solution of this embodiment can achieve at least the following beneficial technical effects: Since the above-mentioned partitioning rules generate tens of thousands of grid regions, the present invention uses word embedding to embed the grid regions, transforming them from a one-hot code representation with tens of thousands of dimensions into a low-dimensional and dense vector. Word embedding further compresses the representation size of the original spatial information of the trajectory, thereby improving the efficiency of subsequent data processing.

[0058] illustrative, see Figure 2 This is a schematic diagram illustrating the data processing principle of the original trajectory map construction module. First, it acquires trajectory data generated by an anonymous user. This trajectory data includes multiple check-in records, with timestamps ranging from 1 to... These are: First check-in record: (position 1, timestamp 1), Second check-in record: (position 2, timestamp 2), ..., ... Individual check-in records: (location) timestamp In the diagram, the first layer is the trajectory data map, which is a discrete representation of the geographical locations of all check-in records of the trajectory data in the surface space, generated from the trajectory data, and denoted as ( ),and The first layer records the latitude and longitude values ​​of each geographical location. Taking this trajectory data map as an example, the trajectory data includes 9 check-in records, so the discrete representation includes the latitude and longitude values ​​of 9 geographical locations. Then, the surface space is divided into evenly, revealing the grid areas where the geographical locations of each check-in record in the trajectory data map are located. The second layer is the original trajectory map, which is based on the surface space after dividing it into grid areas. It uses the method described above to represent the grid areas of each check-in record in the trajectory data, denoted as... In the original trajectory graph, based on the chronological order of the timestamps of all check-in records in the trajectory data, the corresponding grid regions of each check-in record are connected sequentially. Additionally, since the second and third check-in records belong to the same grid region, there is an edge connecting the grid regions corresponding to these two records, ultimately forming nine edges. The third layer is the vectorized result of embedding the original trajectory graph, denoted as... Since there are 9 check-in records, the vectorized version It includes vectors from 9 grid regions.

[0059] (2) Multi-scale trajectory map construction module

[0060] Since check-in records are actively provided by users' software usage, the time intervals between check-in records can be large, resulting in random and sparse check-in records, which hinders the effective representation of trajectory spatial information. Therefore, according to an embodiment of the present invention, the multi-scale trajectory graph construction module is provided with a node sampling unit and an edge construction unit. The node sampling unit is used to sample nodes from the original trajectory graph multiple times according to preset sampling rates and sampling methods preset for each sampling rate, to obtain multiple node sets, wherein the multiple sampling rates include sampling at 100%; the edge construction unit is used to determine whether there is an edge relationship between any two nodes in each node set according to preset edge construction rules, and to establish an edge between the two nodes with an edge relationship, to obtain a trajectory graph of one scale under each sampling rate. The technical solution of this embodiment can at least achieve the following beneficial technical effects: the sampling of the present invention includes complete sampling at 100%, which preserves complete spatial information. In addition, in order to avoid the sparsity of check-in records affecting the recognition accuracy of the model, some nodes in the trajectory are deleted according to the sampling rate, and the trajectory graph is represented again, which can help the model better learn the features of the trajectory, thereby improving the recognition ability of the actual trajectory.

[0061] According to one embodiment of the present invention, in the node sampling unit, the sampling rate can be set to 50%, 33%, or 25%, etc. Illustratively, three sampling rates are preset, namely 100%, 50%, and 33%. When sampling nodes at 50% and 33%, interval sampling is performed based on the timestamps of the check-in records. For example, for the 10 nodes from node 1 to node 10, the preset sampling method for 50% includes: sampling every two nodes in chronological order according to the timestamps of the check-in records, achieving interval sampling of multiple nodes corresponding to the overall trajectory, resulting in node 1, node 3, node 5, node 7, and node 9. The preset sampling method for 33% includes: sampling every three nodes in chronological order according to the timestamps of the check-in records, achieving interval sampling of multiple nodes corresponding to the overall trajectory, resulting in node 1, node 4, node 7, and node 10. Since grid areas are used to represent the nodes of the check-in records, the sampling node represents the corresponding grid area, and the trajectory map is constructed based on the sampled grid area.

[0062] It should be understood that the sampling rate and sampling method in the above embodiments are merely illustrative and can be set according to actual needs. For example, only two sampling rates can be preset, namely 100% and 50% respectively, or 100% and 33% respectively, or four sampling rates can be preset, namely 100%, 50%, 33%, and 25% respectively. The sampling method can also be random sampling; this invention does not limit the sampling rate and sampling method.

[0063] According to an embodiment of the present invention, in the edge construction unit, the preset edge construction rule includes: when any two nodes satisfy either a preset time condition or a preset spatial distance condition, it is determined that an edge relationship exists between the two nodes. The preset time condition includes that the timestamps of the check-in records corresponding to the two nodes are adjacent, and the preset spatial distance condition includes that the distance between the geographical locations corresponding to the two nodes is less than a preset threshold. Adjacent timestamps mean that in a series of nodes sampled at a 100% sampling rate, for example, the timestamps corresponding to nodes 1, 2, 3, 4, and 5 are 1:20, 1:30, 1:50, 2:00, and 2:30 respectively, then the timestamps corresponding to nodes 1 and 2 are adjacent, the timestamps corresponding to nodes 2 and 3 are adjacent, the timestamps corresponding to nodes 3 and 4 are adjacent, and the timestamps corresponding to nodes 4 and 5 are adjacent. Furthermore, in a series of nodes sampled at a sampling rate of 50%, for example, the timestamps corresponding to nodes 1, 3, and 5 are 1:20, 1:50, and 2:30 respectively. Therefore, the timestamps corresponding to nodes 1 and 3 are adjacent, and the timestamps corresponding to nodes 3 and 5 are adjacent. This embodiment's technical solution can achieve at least the following beneficial technical effects: by considering the temporal order in the trajectory and the proximity in geographic space, the grid areas (also called nodes) representing check-in records are connected into a check-in-oriented graph (Construction of Check-in Oriented Graphs, or COGs, also known as trajectory graphs) structure, enabling these constructed check-in-oriented graphs to represent the geographic spatial information of the trajectory.

[0064] According to an embodiment of the present invention, in the edge construction unit, for a trajectory data, firstly, it is determined whether a preset time condition is met: in terms of time, if the timestamps of two check-in records are adjacent, then the grid regions (also called nodes) corresponding to the geographical locations of the two check-in records have an edge relationship. Therefore, according to the chronological order of the timestamps of all check-in records in the trajectory data, the grid regions corresponding to each check-in record are sequentially connected to form multiple edges. Secondly, it is determined whether a preset spatial distance condition is met: in terms of space, if the geographical locations of the grid regions corresponding to the geographical locations of two check-in records are adjacent, then the two grid regions have an edge relationship. The proximity of geographical locations can be determined by calculating the Euclidean distance between the two grid regions, and then manually setting a threshold to determine whether the calculated Euclidean distance is less than the preset threshold. If the Euclidean distance is less than the preset threshold, then the two grid regions have an edge relationship. Therefore, connecting two geographically adjacent grid regions yields one or more edges. Finally, combining all the edges constructed in both time and space aspects, a general non-Euclidean graph structure is obtained. The method for determining the preset spatial distance condition is as follows:

[0065] ,

[0066] in, Represents two grid regions (i.e., grid areas) and grid area There is an edge relationship between them. 0 indicates a grid area and grid area There is no boundary relationship between them. Represents grid area Longitude of location Represents grid area Longitude of location Represents grid area Latitude of location Represents grid area Latitude of location This indicates a preset threshold.

[0067] illustrative, see Figure 3 This is a schematic diagram illustrating the principle of constructing multiple trajectory maps at different scales. Figure 2 Taking the original trajectory graph obtained as an example, in order to enable the model to handle trajectory data of different lengths, a multi-scale sampling method is used (i.e., sampling nodes in the original trajectory graph at multiple preset sampling rates). In the figure, three preset sampling rates are 100%, 50%, and 33%, respectively. When sampling nodes at 50% and 33%, interval sampling is performed based on the timestamps of the check-in records. After sampling, the edge construction rules described above are used to build edges for each node in each sampled node set, thus obtaining three trajectory graphs at different scales. Among the three trajectory graphs obtained from top to bottom on the right, the first trajectory graph is obtained by constructing edges after sampling at 100%. Compared to the original trajectory graph, it has 6 more edges because it considers both time and space, unlike the original trajectory graph which only considers time. The second trajectory graph is obtained by interval sampling at 50% and edge construction; the third trajectory graph is obtained by interval sampling at 33% and edge construction; thus obtaining multiple check-in-oriented graphs at different scales.

[0068] (3) Trajectory Spatial Information Extraction Module

[0069] According to one embodiment of the present invention, the trajectory spatial information extraction module includes multiple graph attention networks, a multi-head self-attention mechanism layer, a global average pooling unit, and an aggregation unit. Each graph attention network (GAT) is used to determine the attention value between each node in each trajectory graph and any other node based on the spatial information of the nodes in each trajectory graph, and to update the initial features of each node according to the attention value. The multi-head self-attention mechanism layer is used to determine the relation weight between each node and any other node based on the initial features of each node in each trajectory graph using a multi-head self-attention mechanism, and to obtain the spatial features of each node in each trajectory graph according to the relation weight. The global average pooling unit is used to perform global average pooling processing on the spatial features of each node in each of the multiple trajectory graphs to obtain the graph spatial features of each trajectory graph, wherein the graph spatial features corresponding to the multiple trajectory graphs have the same dimension. The aggregation unit is used to concatenate the graph spatial features of the multiple trajectory graphs to obtain multi-scale trajectory spatial features.

[0070] The technical solutions of the above embodiments can achieve at least the following beneficial technical effects: First, for each scale of trajectory map constructed based on the original trajectory map, since no fixed number of nodes are sampled, the amount of spatial information contained in the trajectory map is arbitrary. To process arbitrary amounts of spatial information and maintain low time complexity as much as possible, this invention uses a graph attention network to embed the spatial information of each node into a low-dimensional vector space for effective application in real-world scenarios. Second, the graph attention network updates the spatial feature representation of nodes by calculating the attention values ​​between nodes, thus capturing the spatial relationships between different check-in records. Furthermore, a multi-head self-attention mechanism is adopted to better represent the spatial relationships between check-in records, effectively analyze the spatial information and spatial interactions between nodes, and this mechanism can process multiple attention heads in parallel, improving the model's ability to represent node spatial relationships and increasing running speed. Then, global average pooling is used to aggregate the embedding vectors of different numbers of nodes into a unified dimension. This invention uses global average pooling to achieve information processing of trajectory data of arbitrary length, improving the accuracy of anonymous trajectory recognition. Finally, the graph space features of all trajectory maps after multi-scale sampling are aggregated to better capture the multi-scale information of the trajectory.

[0071] According to one embodiment of the present invention, in the multi-head self-attention mechanism layer, the spatial features of each node of each trajectory graph are obtained according to the relation weights as follows:

[0072] ,

[0073] in, Indicates the first Spatial characteristics of each node This indicates the number of network layers with a multi-head self-attention mechanism. This indicates the network layer number of the multi-head self-attention mechanism layer. This represents the activation function sigmoid. Indicates the first The node belongs to the node. A trajectory map The nodes in Indicates the input number of the first... A trajectory diagram, This indicates the first layer of the multi-head self-attention mechanism. The first layer of the network is determined by the layer. The node and the first The relationship weights of edges between nodes , , express function, express function, This represents the attention score. Indicates the first Initial characteristics of each node With learnable parameters Multiply, Indicates the first Initial characteristics of each node With learnable parameters Multiply, This indicates the first layer of the multi-head self-attention mechanism. The first layer of the network is determined by the layer. The node and the first The edge relationships between nodes, The first layer representing the multi-head self-attention mechanism layer Learnable parameters of a network layer Indicates the first Each node belongs to the trajectory graph. The nodes in.

[0074] According to one embodiment of the present invention, the global average pooling processing method is as follows:

[0075] ,

[0076] in, Indicates the first A trajectory map Graph space features, This indicates taking the average. Indicates the first A trajectory map The spatial characteristics of the first node, Indicates the first A trajectory map The spatial features of the second node, Indicates the input number of the first... A trajectory map The spatial characteristics of the last node.

[0077] According to one embodiment of the present invention, the polymerization unit is processed as follows:

[0078] ,

[0079] in, Representing multi-scale trajectory spatial features, This represents the graph space features of stitching together all trajectory graphs. This represents the total number of trajectory maps constructed after multi-scale sampling.

[0080] (4) Trajectory recognition module

[0081] According to one embodiment of the present invention, the trajectory recognition module includes a trajectory embedding layer and a linking unit. The trajectory embedding layer is used to process multi-scale trajectory spatial features using a recurrent neural network to obtain fused trajectory spatial features. The linking unit includes a multilayer perceptron and a softmax function. The multilayer perceptron is used to represent the fused trajectory spatial features to obtain the final trajectory spatial features, and the softmax function is used to obtain the probability value of the trajectory data belonging to each known user based on the final trajectory spatial features.

[0082] According to one embodiment of the present invention, the recurrent neural network preferably employs a bidirectional long short-term memory network (BiLSTM), but other recurrent neural networks such as a bidirectional recurrent neural network (Bi-RNN) can also be used; the present invention is not limited in this regard. Specifically, a bidirectional long short-term memory network (BiLSTM) is used to fuse trajectory representations at different scales. By combining a forward LSTM and a backward LSTM, BiLSTM processes the multi-scale trajectory space features, capturing the relationships between trajectory representations at different scales, resulting in the following fused trajectory space features:

[0083] ,

[0084] in, Represents the trajectory space features of the fusion. Indicates splicing, This represents the multi-scale trajectory space features of the input feedforward LSTM. This represents the multi-scale trajectory space features of the input backward LSTM.

[0085] According to one embodiment of the present invention, a multilayer perceptron is used in the linking unit to connect the trajectory data of anonymous users to each known user, and the fused trajectory space features are represented to obtain the final trajectory space features. The final trajectory space features are then used as input to a Softmax function, which calculates the probability value of the trajectory data belonging to each known user. The probability value is calculated as follows:

[0086] ,

[0087] in, This represents the probability that the trajectory data belongs to the i-th known user. This represents the user label of the i-th user. Indicates learnable parameters, This represents the fused trajectory space features obtained after processing the trajectory data of the i-th anonymous user through the model. This represents the learnable bias of a multilayer perceptron. This represents the anonymous user's ID. This represents the total number of anonymous users. This represents the fused trajectory space features obtained after processing the trajectory data of the kth anonymous user through the model.

[0088] Overall, the model structure and its overall data processing process are as follows: Figure 4 As shown, Figure 4 This is a schematic diagram illustrating the overall framework of the model and the complete process of processing trajectory data. The overall data processing procedure is as follows: Steps 1-4:

[0089] 1. Using the original trajectory graph construction module, an original trajectory graph is constructed based on the input trajectory data consisting of multiple check-in records. Each node in the original trajectory graph is embedded to obtain its spatial information. The processing procedure of the original trajectory graph construction module is as follows: Figure 2 The processing steps shown are the same, so they will not be repeated here.

[0090] 2. The multi-scale trajectory map construction module generates multiple trajectory maps at different scales based on the original trajectory map. The construction process and... Figure 3 The process shown is the same, and will not be repeated here. The three trajectory plots obtained at different scales will be denoted as follows: , , , , , These refer to the number of nodes (grid areas) in the first, second, and third trajectory maps, respectively. The dimension representing the spatial information of each node (grid area).

[0091] 3. The trajectory spatial information extraction module extracts multi-scale trajectory spatial features based on the spatial information of trajectory maps and nodes at multiple different scales. Figure 3 Taking three trajectory maps at different scales as an example, a graph attention network is constructed for each trajectory map, resulting in three graph attention networks, denoted as GAT1, GAT2, and GAT3. Each graph attention network is used to embed a representation of each node in each trajectory map, obtaining the initial representation of each node. A multi-head self-attention mechanism layer is then used to further process the initial representations of each node in each trajectory map, obtaining the spatial features of each node. In this way, the overall output of the spatial features of all nodes in each of the three trajectory maps is denoted as... , , , This represents the dimension of the spatial features of each node (grid region) in the trajectory graph after processing by the graph attention network and multi-head self-attention mechanism layer. A global average pooling unit is used to perform global average pooling on the spatial features of each node in each trajectory graph to obtain the graph spatial features of the trajectory graph. The graph spatial features corresponding to multiple trajectory graphs have the same dimension. An aggregation unit concatenates the graph spatial features of multiple trajectory graphs to obtain multi-scale trajectory spatial features.

[0092] 4. The trajectory recognition module obtains the probability values ​​of trajectory data belonging to each known user based on multi-scale trajectory spatial features. The trajectory embedding layer of the trajectory recognition module uses a recurrent neural network to process the multi-scale trajectory spatial features to obtain fused trajectory spatial features; a multilayer perceptron is used to represent the fused trajectory spatial features to obtain the final trajectory spatial features; and the Softmax function is used to obtain the probability values ​​of trajectory data belonging to each known user based on the final trajectory spatial features.

[0093] II. Model Training

[0094] According to an embodiment of the present invention, a training method for a model for anonymous spatiotemporal trajectory recognition is provided, the method comprising steps S1 and S2:

[0095] Step S1: Obtain the training set, which includes multiple samples. Each sample includes trajectory data generated by a known user within a certain period of time and the corresponding user label sequence. The user label sequence includes multiple user labels, and each user label is used to indicate whether the corresponding trajectory data belongs to the known user.

[0096] According to one embodiment of the present invention, the trajectory data of the sample can be obtained from the backend database of existing social networking service software, such as the backend database of any of the social networking service software such as Gowalla and Brightkite, to obtain the check-in records and trajectory data of multiple known users. A known user includes one or more trajectory data sets. The samples in the training set include a trajectory data set and a corresponding user label sequence. Each user label is represented by label 1 or label 0. Label 1 indicates that the corresponding trajectory data belongs to the known user, and label 0 indicates that the corresponding trajectory data does not belong to the known user. The number of dimensions of the user label sequence corresponding to a trajectory data set is equal to the number of known users in the entire dataset. For example, if there are 5 known users in the current dataset, namely known user 1, known user 2, known user 3, known user 4, and known user 5, and the trajectory data belongs to known user 2, then the user label sequence corresponding to a trajectory data set is [0,1,0,0,0].

[0097] Step S2: Using the training set, train the model described in the above embodiment. Based on the input trajectory data, obtain the probability value of the trajectory data belonging to each known user. Update the model parameters with the optimization objective of minimizing the difference between the probability values ​​of each known user and the user label sequence to obtain the trained model.

[0098] According to one embodiment of the present invention, the training task of the model is a TUL (Trajectory-User Linking) task, which is a task that associates the trajectory data of anonymous users with known users. Therefore, the loss function is calculated using a multi-class cross-entropy loss function.

[0099] It should be noted that high time complexity is a significant factor limiting the training time and widespread application of anonymized trajectory recognition technology. Traditional anonymized trajectory recognition techniques often analyze geographical information in the trajectory sequentially, lacking sufficient parallel computing capabilities. Furthermore, some techniques utilize generative networks to assist in recognition, learning trajectory features through trajectory reconstruction, which undoubtedly introduces additional computational demands, further increasing computational complexity and trajectory recognition time. However, based on the model structure of this invention, the above embodiments can achieve at least the following beneficial effects: During model training, because the model structure implements a faster and more accurate anonymized trajectory recognition technique, the training time is only half that of traditional anonymized trajectory recognition techniques, and only one-tenth that of related techniques incorporating generative networks, enabling faster deployment in end-to-end application scenarios. In particular, the multiple graph attention networks and multi-head self-attention mechanism layers in the model alleviate the problem of high data computation requirements by performing parallel computation on multiple trajectory graphs and compressing them into a low-dimensional space.

[0100] III. Application Scenarios

[0101] According to an embodiment of the present invention, an anonymous spatiotemporal trajectory recognition method is provided. The method includes: acquiring trajectory data generated by an anonymous user over a period of time; processing the trajectory data using a trained model obtained by the training method of the above embodiment to obtain probability values ​​of the trajectory data belonging to each known user; and identifying the known user to which the trajectory data belongs based on the probability values ​​of each known user.

[0102] According to one embodiment of the present invention, the known users whose probability values ​​exceed a preset probability value can be identified as the recognition result. Alternatively, the probability values ​​of the known users can be sorted, and the known user corresponding to the first or more probability values ​​in the sorted list can be selected as the recognition result.

[0103] To verify the beneficial effects of the present invention, the inventors conducted the following experiments:

[0104] First, for the model of this invention (TULMGAT for short), three sampling rates are preset: 100%, 50%, and 33%. When sampling nodes at 50% and 33%, sampling is performed at intervals based on the timestamps of the check-in records. Based on the above three sampling rates, the following four variations of the model of this invention are obtained:

[0105] ①TULMGAT-O (also known as TUL-GAT): Only 100% sampling is used to obtain a trajectory map of one scale.

[0106] ②TULMGAT-T: Sampling is performed only at 100% and 50% to obtain two trajectory maps at different scales.

[0107] ③TULMGAT-TH: Sampling at 100%, 50% and 33% yields three trajectory maps at different scales, and the outputs of the global average pooling layer are stitched together for subsequent recognition.

[0108] ④TULMGAT-Add: Sampling is performed at 100%, 50% and 33% to obtain trajectory maps at three different scales. The outputs of the global average pooling layer are directly summed for subsequent recognition without using a recurrent neural network for further fusion.

[0109] Secondly, the trajectory datasets were obtained from two social networking service software programs, Gowalla and Brightkite. Training and test sets were generated based on the datasets from these two programs. Using the datasets obtained from each software program, four variants of the model of this invention were compared with existing methods. These existing methods include TULER-LSTM, TULER-GRU, BiTULER, TULVAE, TULAR, STULIG, GNNTUL, and TULRN. TULER-LSTM, TULER-GRU, and BiTULER are variants of TULER, which was the first model to solve the TULER task, utilizing various RNNs to learn user mobility and achieve classification.

[0110] Finally, the experimental results were compared using five evaluation metrics: ACC@1, ACC@5, Macro-P, Macro-R, and Macro-F1. Here, ACC@K represents the classification accuracy of user links (the proportion of correctly predicted users among the first K predictions), specifically ACC@1 represents the proportion of correctly predicted users among the first 1 prediction, ACC@5 represents the proportion of correctly predicted users among the first 5 predictions, and Macro-P, Macro-R, and Macro-F1 represent the macro-average precision, macro-average recall, and macro-average F1 score for all user categories, respectively.

[0111] Using a dataset obtained from the software Gowalla, four variants of the model of this invention were compared with existing methods. The experimental comparison results are shown in Table 1 below:

[0112] Table 1

[0113]

[0114] In Table 1 above, |U|=149 and |U|=247 represent the number of known users as 149 and 247, respectively. In Table 1, the optimal results are marked in bold. In particular, the TULMGAT-T and TULMGAT-TH variants demonstrate that the present invention has good anonymous trajectory recognition capabilities and higher accuracy compared to existing methods.

[0115] Using a dataset obtained from the Brightkite software, four variants of the model of this invention were compared with existing methods. The comparison results are shown in Table 2 below:

[0116] Table 2

[0117]

[0118] In Table 2 above, |U|=141 and |U|=223 represent the number of known users as 141 and 223, respectively. In Table 1, the optimal results are marked in bold. In particular, the three variants TULMGAT-O, TULMGAT-T and TULMGAT-TH demonstrate that the present invention has good anonymous trajectory recognition capabilities and has lower training time and higher efficiency compared with existing works.

[0119] Using two datasets obtained from Gowalla and Brightkite software, four variants of the model of this invention were trained against models of existing methods. The training time was recorded and compared in hours (h), and the training time comparison results are shown in Table 3 below:

[0120] Table 3

[0121]

[0122] As shown in Table 3 above, the model of the present invention requires the least training time. Therefore, the model of the present invention not only achieves a high trajectory recognition accuracy, but also achieves higher training efficiency.

[0123] The existing method is as follows:

[0124] TULER: refers to the above reference [1], where TULER-LSTM method, TULER-GRU method and BiTULER method are variants of TULER;

[0125] TULVAE method: as mentioned in the above reference [2];

[0126] TULAR method: T. Sun, Y. Xu, F. Wang, L. Wu, T. Qian, Z. Shao, Trajectory-user link with attention recurrent networks, in: ICPR, 2020, pp.4589–4596.

[0127] STULIG method: see the above reference [3];

[0128] GNNTUL method: F. Zhou, S. Chen, J. Wu, C. Cao, S. Zhang, Trajectory-userlinking via graph neural network, in: ICC 2021, 2021, pp. 1–6.

[0129] TULRN method: Y. Sang, Z. Xie, W. Chen, L. Zhao, Tulrn: Trajectory userlinking on road networks, World Wide Web (2022) 1–17.

[0130] It should be noted that although the steps are described in a specific order above, it does not mean that the steps must be executed in the above specific order. In fact, some of these steps can be executed concurrently, or even in a different order, as long as the required function can be achieved.

[0131] This invention can be a system, method, and / or computer program product. A computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for causing a processor to implement various aspects of the invention.

[0132] Computer-readable storage media can be tangible devices that hold and store instructions for use by an instruction execution device. Computer-readable storage media can include, for example, but not limited to, electrical storage devices, magnetic storage devices, optical storage devices, electromagnetic storage devices, semiconductor storage devices, or any suitable combination thereof. More specific examples (a non-exhaustive list) of computer-readable storage media include: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static random access memory (SRAM), portable compact disc read-only memory (CD-ROM), digital multifunction disc (DVD), memory sticks, floppy disks, mechanical encoding devices, such as punch cards or recessed protrusions storing instructions thereon, and any suitable combination thereof.

[0133] The various embodiments of the present invention have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles, practical application, or technical improvements to the embodiments in the market, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. A model for anonymous spatiotemporal trajectory recognition, characterized in that, The model includes: The original trajectory graph construction module is used to construct an original trajectory graph based on the trajectory data generated by anonymous users over a period of time. It includes multiple nodes representing multiple check-in records of trajectory data and multiple edges representing the relationships between nodes. The module also embeds multiple nodes of the original trajectory graph to obtain the spatial information of multiple nodes. The multi-scale trajectory map construction module is used to sample nodes from the original trajectory map at multiple preset sampling rates to construct multiple trajectory maps at different scales. A trajectory spatial information extraction module is used to extract spatial features of each node in multiple trajectory graphs based on the spatial information of multiple nodes, and to aggregate the spatial features of each node in each trajectory graph to obtain multi-scale trajectory spatial features; wherein, the trajectory spatial information extraction module includes: Multiple graph attention networks are used, each graph attention network is used to determine the attention value between each node in each trajectory graph and any other node in all other trajectories based on the spatial information of the nodes in each trajectory graph, and update the initial features of each node according to the attention value; The multi-head self-attention mechanism layer is used to determine the relation weight between each node and any other node in each trajectory graph based on the initial features of each node in each trajectory graph using the multi-head self-attention mechanism, and obtain the spatial features of each node in each trajectory graph based on the relation weight. The global average pooling unit is used to perform global average pooling on the spatial features of each node of each trajectory graph in multiple trajectory graphs to obtain the graph space features of each trajectory graph. The graph space features of multiple trajectory graphs have the same dimension. The aggregation unit is used to stitch together the graph space features of multiple trajectory maps to obtain multi-scale trajectory space features. The trajectory recognition module is used to identify the probability value of the trajectory data belonging to each known user based on the multi-scale trajectory spatial features.

2. The model according to claim 1, characterized in that, Each of the multiple check-in records includes a timestamp and a geographical location. The original trajectory map construction module constructs the original trajectory map in the following ways: Obtain multiple grid areas obtained by dividing the land surface space into average sizes according to a preset geographic space, and use the grid area where the geographical location of each check-in record in the trajectory data is located as the node used to represent that check-in record; Based on the chronological order of the timestamps of all check-in records in the trajectory data, the grid areas corresponding to each check-in record are connected sequentially to form multiple edges, thus obtaining the original trajectory map.

3. The model according to claim 2, characterized in that, The multi-scale trajectory map construction module includes: The node sampling unit is used to sample nodes from the original trajectory map multiple times according to multiple preset sampling rates and a preset sampling method for each sampling rate, to obtain multiple node sets. Among them, the multiple sampling rates include sampling at 100%. The edge construction unit is used to determine whether there is an edge relationship between any two nodes in each node set according to the preset edge construction rules, and to establish an edge between the two nodes with an edge relationship, so as to obtain a trajectory map of one scale under each sampling rate.

4. The model according to claim 3, characterized in that, The edge construction unit includes the following preset edge construction rules: When any two nodes meet either a preset time condition or a preset spatial distance condition, it is determined that there is an edge relationship between the two nodes. The preset time condition includes that the timestamps of the check-in records of the two nodes are adjacent, and the preset spatial distance condition includes that the distance between the geographical locations of the two nodes is less than a preset threshold.

5. The model according to claim 1, characterized in that, The trajectory recognition module includes: The trajectory embedding layer is used to process multi-scale trajectory spatial features using recurrent neural networks to obtain fused trajectory spatial features. The linking unit is used to represent the fused trajectory space features using a multilayer perceptron to obtain the final trajectory space features, and then uses the Softmax function to obtain the probability value of the trajectory data belonging to each known user based on the final trajectory space features.

6. A training method for a model used for anonymous spatiotemporal trajectory recognition, characterized in that, Training methods include: Obtain a training set, which includes multiple samples. Each sample includes trajectory data generated by a known user over a period of time and a corresponding user label sequence. The user label sequence includes multiple user labels, and each user label is used to indicate whether the corresponding trajectory data belongs to the known user. The model according to any one of claims 1-5 is trained using the training set. Based on the input trajectory data, the probability value of the trajectory data belonging to each known user is obtained. The parameters of the model are updated with the optimization objective of minimizing the difference between the probability values ​​of each known user and the user label sequence, thus obtaining the trained model.

7. An anonymous spatiotemporal trajectory recognition method, characterized in that, The methods include: Obtain trajectory data generated by anonymous users over a period of time; Using the trained model obtained by the method described in claim 6, the trajectory data is processed to obtain the probability values ​​of the trajectory data belonging to each known user; Identify the known user to whom the trajectory data belongs based on the probability values ​​of each known user.

8. A computer-readable storage medium, characterized in that, It stores a computer program that can be executed by a processor to implement the steps of the method of claim 6 or 7.

9. An electronic device, characterized in that, include: One or more processors; as well as Memory, wherein the memory is used to store executable instructions; The one or more processors are configured to implement the steps of the method of claim 6 or 7 by executing the executable instructions.

Images