Method and system for accurate user linking across instant messaging platforms to combat location errors

Abstract

The present invention relates to the field of big data processing technology, and more particularly to a method and system for accurately linking users across instant messaging platforms for combating location errors. The method comprises performing grid transformation processing on user data across instant messaging platforms to obtain user grid trajectories and motion state trajectories. The method further comprises obtaining the motion state distance between the target platform user trajectory and the to-be-linked platform user trajectory under time window control, and obtaining a candidate trajectory corresponding to the target platform user trajectory based on the motion state distance. The method further comprises clustering the location points on the user trajectory to obtain the cluster center distance between the target platform user trajectory and the candidate trajectory. The method further comprises weighting the cluster center distances between the trajectories according to a trajectory similarity weight model and calculating the similarity between the target platform user trajectory and the candidate trajectory, thereby obtaining a matching trajectory for the target platform user trajectory and achieving user linking across instant messaging platforms. The method can accurately extract user features even when there are errors in the location data, thereby achieving higher accuracy in user linking.

Description

Technical Field

[0001] The present invention relates to the field of big data processing technology, and in particular to a method and system for accurately linking users across instant messaging platforms for combating location errors. Background Art

[0002] Instant messaging, a popular location-based social network (LBSN), provides location-based social discovery (LBSD) services, such as the "People Nearby" feature in WeChat and Momo. This allows users to obtain the identities and location information of other users near their current location, generating massive amounts of mobile location data. Cross-platform user linking refers to linking accounts belonging to the same user across different IM platforms. Since registered accounts on different IM platforms are independent, linking a person's user information and behavior across different platforms helps identify the user's multiple social identities and more accurately portrays the user through data fusion. Linking cross-platform IM users through user movement trajectories has important application value in real-time trajectory supplementation, collaborative digital tracking, and trajectory prediction.

[0003] Existing cross-platform user linking methods are primarily categorized into three types: those based on user attributes, network relationships, and location. Methods based on user attributes link users by measuring the similarity of attributes such as names, birthdays, and hobbies. For example, cross-platform user association is achieved based on mobile phone numbers. However, due to the uncertainty surrounding the accuracy of user-registered attribute data and the increasing importance of privacy protection, most social networks have restricted access to user attribute data, limiting methods based on user attributes. Methods based on network relationships primarily link users based on friendships and network structure. For example, methods learn the latent semantics of network structure end-to-end and encode the structural information of anchor nodes by predicting the proximity structure of nodes in the network to align the network, or methods align users across platforms based on the network's social structure. These methods often require latent knowledge learned from large amounts of prior data, which is difficult to obtain and often noisy. Location-based methods generally link users across platforms based on the similarity of user location factors. Examples include the k-BCT method, which measures trajectory similarity based on the shortest distance, and methods that measure user similarity based on stimulus signals. To accurately describe user mobility characteristics in sparse data, the space can be divided into a grid, and entropy-based kernel density estimation methods can be used to assess user similarity. However, these methods often focus on locations such as the shortest distance or co-occurrence, ignoring the role of other locations. The grid distance method also ignores the actual distribution of user locations.

[0004] In summary, existing research methods can effectively solve the problem of cross-platform user connection under certain conditions, but they still have shortcomings when dealing with inaccurate data information. Existing location-based methods mainly rely on the comparison and statistical results of locations. However, unlike trajectories composed of accurate locations, the location data in instant messaging user trajectories often contains random errors. Moreover, the number of location points cannot fully reflect the user's actual movement status. As a result, the comparison and statistical results of location data cannot accurately reflect the relationship between actual locations. As a result, the above methods are prone to misjudgment when solving user connections across instant messaging platforms, especially in situations where there is a high degree of overlap in activity spaces, which reduces the accuracy of user connections and further leads to problems such as low accuracy when solving user connections across instant messaging platforms. Summary of the Invention

[0005] To this end, the present invention provides a method and system for accurately linking users across instant messaging platforms to combat location errors, thereby solving the problems of errors in existing location data based on instant messaging user trajectories and low accuracy of cross-platform user links.

[0006] According to the design scheme provided by the present invention, on the one hand, a method for accurately linking users across instant messaging platforms for combating location errors is provided, comprising:

[0007] Obtaining cross-platform user data in a specified area, the cross-platform user data including user sets and user trajectory sets in both the target platform and the platform to be linked; and performing grid transformation on the cross-platform user data to obtain grid trajectories and motion state trajectories of users in each user set;

[0008] Under the time window control, the motion state distance between the target platform user trajectory and the to-be-linked platform user trajectory is obtained, so as to obtain the candidate trajectory corresponding to the target platform user trajectory according to the motion state distance;

[0009] Cluster the location points on the user trajectory according to the time window, and obtain the cluster center distance between the target platform user trajectory and the candidate trajectory. The time window adopts a rolling time window with adjustable step size.

[0010] The cluster center distances between trajectories are weighted according to a trajectory similarity weight model to obtain the similarity between the target platform user trajectory and other candidate trajectories. The trajectory similarity weight model is constructed using grid contribution and motion sequence similarity. The grid contribution is used to characterize the amount of user trajectory feature information possessed by each grid in the grid trajectory and is obtained using grid trajectory entropy. The motion sequence similarity is obtained based on the motion state trajectory sequence and the spatiotemporal co-occurrence between trajectories.

[0011] According to the similarity between the target platform user trajectory and other candidate trajectories, the candidate trajectory with the maximum similarity between trajectories is selected as the matching trajectory of the target platform user trajectory, and the cross-instant messaging platform users corresponding to the trajectory are linked.

[0012] As the cross-platform precise user linking method for combating location errors of the present invention, further, grid transformation processing is performed on the cross-platform user data, including:

[0013] Setting the spatiotemporal granularity for describing the size of the spatiotemporal grid, the spatiotemporal granularity includes the side length of the grid area and the width of the time window;

[0014] Divide the specified area into grids and obtain the grid numbers corresponding to each location point on the user's trajectory;

[0015] Sort the grids according to the time sequence to generate grid tracks;

[0016] Slide the time window and traverse the grid trajectory, record the number of grid changes of the user trajectory in each time window, and combine the motion state in the previous time window to determine the motion state of the trajectory in the current time window to generate the motion state trajectory.

[0017] As a cross-platform precise user linking method for combating position errors of the present invention, further, obtaining the motion state distance between the trajectory of the user on the target platform and the trajectory of the user on the to-be-linked platform under time window control includes:

[0018] Compare the trajectory of each target user in the target platform with the trajectory of the user in the platform to be linked, and obtain the trajectory motion state distance between the trajectory of each target user in the target platform and the trajectory of the user in the platform to be linked;

[0019] Obtain a trajectory filtering score based on the trajectory motion state distance and the set and number of position co-occurrence time windows in the trajectory. The trajectory filtering score is used to describe the mean difference in motion state of the matching trajectories to be compared during the spatiotemporal co-occurrence of the position points.

[0020] According to the trajectory filtering score, the user trajectories in the platform to be linked that do not match the target user trajectory are deleted to obtain the candidate trajectories corresponding to the target user trajectories on the target platform.

[0021] As a cross-platform precise user linking method for combating location errors of the present invention, further, the location points on the user trajectory are clustered according to the time window, including:

[0022] Constructing a perception time window with a dynamically adjustable step size, and determining the time window width based on the distribution of the perception location data. The time window width includes the spatiotemporal grid time granularity and the rolling step size;

[0023] The rolling step of the time window is dynamically adjusted according to the time interval between the trajectory segments of the target platform user trajectory, and the cluster center distance between the location points between the target platform user trajectory and the to-be-linked platform user trajectory in the corresponding time window is obtained.

[0024] As a cross-platform precise user linking method for combating location errors in the present invention, the cluster center distances between trajectories are further weighted according to a trajectory similarity weight model to obtain the similarity between the target platform user trajectory and other candidate trajectories, including:

[0025] Obtain the basic similarity between the target platform user trajectory and the to-be-linked platform user trajectory based on the cluster center distance between the trajectories;

[0026] The trajectory entropy is obtained based on the number of grid cells contained in the grid trajectory and the number of location points where the user trajectory falls into each grid. The contribution of the grid in the grid trajectory is obtained based on the trajectory entropy.

[0027] Based on the motion state trajectory and the spatiotemporal co-occurrence of the trajectories between the target platform and the platform to be linked, the motion sequence similarity between the trajectory of the user on the target platform and the trajectory of the user on the platform to be linked is obtained;

[0028] The similarity between the user trajectory of the target platform and the user trajectory of the platform to be linked is obtained based on basic similarity, grid contribution and motion sequence similarity.

[0029] As a precise user linking method across instant messaging platforms for combating location errors of the present invention, further, a similarity calculation formula between trajectories is expressed as: Among them, T A 、T B are the target platform user trajectory and the to-be-linked platform user trajectory, μ is the adjustment parameter used to reduce the distance value, D dist (T A ,T B ) represents the distance between trajectories, W A ∩W B is the set of time windows where positions co-occur in two trajectories, f num (·) is a function that controls the influence of the spatiotemporal co-occurrence times between two trajectories on the similarity calculation, M sim (T A ,T B ) is the function for calculating the similarity of motion sequences between trajectories.

[0030] As a cross-platform precise user linking method for combating location errors of the present invention, further, selecting a candidate trajectory with the maximum similarity between trajectories as a matching trajectory of the target platform user trajectory includes:

[0031] The similarities between the trajectories of each user on the target platform and the candidate trajectories are sorted, and several user trajectories with the highest similarity rankings are selected from the candidate trajectories as the winning trajectories that match the corresponding user trajectories on the target platform.

[0032] On the other hand, the present invention also provides a cross-platform precise user linking system for combating location errors, comprising: a cross-platform user data acquisition module, a candidate user trajectory acquisition module, a trajectory location point clustering module, a trajectory similarity calculation module and a cross-platform user linking module, wherein:

[0033] A cross-platform user data acquisition module is used to acquire cross-instant messaging platform user data in a specified area, wherein the cross-instant messaging platform user data includes user sets and user trajectory sets from both the target platform and the platform to be linked; and to perform grid transformation processing on the cross-instant messaging platform user data to obtain the grid trajectory and motion state trajectory of the users in each user set;

[0034] The candidate user trajectory acquisition module is used to obtain the motion state distance between the target platform user trajectory and the to-be-linked platform user trajectory under time window control, so as to obtain the candidate trajectory corresponding to the target platform user trajectory based on the motion state distance;

[0035] The trajectory location point clustering module is used to cluster the location points on the user trajectory according to the time window, and obtain the cluster center distance between the target platform user trajectory and the candidate trajectory. The time window adopts a rolling time window with adjustable step size;

[0036] The trajectory similarity calculation module is used to weight the cluster center distances between trajectories according to a trajectory similarity weight model to obtain the similarity between the target platform user trajectory and other candidate trajectories. The trajectory similarity weight model is constructed using grid contribution and motion sequence similarity. The grid contribution is used to characterize the amount of user trajectory feature information possessed by each grid in the grid trajectory and is obtained using grid trajectory entropy. The motion sequence similarity is obtained based on the motion state trajectory sequence and the temporal and spatial co-occurrence between trajectories.

[0037] The cross-platform user linking module is used to select the candidate trajectory with the maximum similarity between the target platform user trajectory and other candidate trajectories as the matching trajectory of the target platform user trajectory, and link the cross-instant messaging platform users corresponding to the trajectory.

[0038] Beneficial effects of the present invention:

[0039] 1. This method uses a multi-granularity spatiotemporal grid to divide the target area. By constructing a spatiotemporal weighting model that incorporates grid contribution and motion sequence similarity, it weights the grid cells passed by a trajectory and the temporal movement pattern similarity between different trajectories. In the trajectory similarity calculation, weights are used to distinguish the importance of different feature vectors, allowing accurate extraction of user features even in the presence of errors in location data, resulting in higher accuracy in user linking.

[0040] 2. This invention extracts the coarse-grained motion state of trajectories and uses this to filter the candidate trajectory set. To comprehensively evaluate the filtering effect, we improve the existing hit rate metric and propose a filtering efficiency metric. Further experiments demonstrate that this filtering method effectively reduces the number of candidate trajectories and improves the efficiency of the user linking algorithm.

[0041] 3. Considering that the number of location points appearing on an instant messaging user's trajectory is subject to the user's subjective control and cannot fully reflect the user's actual movement status, the present invention uses a spatiotemporal co-occurrence window control and weights the cluster center distance of the trajectory within the time window according to a weight model. This makes the similarity calculation related to the spatiotemporal co-occurrence window between the two trajectories, reducing the dependence on the number of location points and helping to mine the implicit user location and movement characteristics in sparse data.

[0042] 4. This invention extracts position and temporal movement features from trajectories, calculates inter-trajectory similarity based on grid contribution and motion sequence similarity, and thus achieves user linking across instant messaging platforms. Experiments conducted on public and self-built datasets demonstrate that this solution has significant accuracy advantages. Compared with existing typical location-based methods (k-BCT, GS, and TF-IWF), the accuracy of this solution (Acc@1) increases by an average of 32.86%, 44.8%, and 14.4%, respectively, and the accuracy of this solution (Acc@3) increases by an average of 30.52%, 34.67%, and 13.19%, respectively. This makes it suitable for user linking across instant messaging platforms. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 Schematic diagram of a precise user linking process across instant messaging platforms for combating location errors in an embodiment;

[0044] Figure 2 This is a schematic diagram of the principle framework of the cross-instant messaging platform precise user linking algorithm in the embodiment;

[0045] Figure 3 Schematic diagram of a perception time window with dynamically adjustable step size in an embodiment;

[0046] Figure 4 This is a schematic diagram showing the effect of subset size on filtering effect in the embodiment;

[0047] Figure 5This is a diagram illustrating the effect of subset size on hit rate in the embodiment;

[0048] Figure 6 This is a diagram showing the effect of subset size on accuracy in the embodiment;

[0049] Figure 7 Schematic diagram of the comparison results of the user link experiment in the embodiment. DETAILED DESCRIPTION

[0050] In order to make the purpose, technical solutions and advantages of the present invention clearer and more understandable, the present invention is further described in detail below with reference to the accompanying drawings and technical solutions.

[0051] Cross-platform user linking can link the same user's accounts across different instant messaging systems, which is crucial for identifying multiple social identities and analyzing cross-platform associations. To facilitate understanding of this solution, we first introduce the data characteristics of instant messaging user trajectories.

[0052] Unlike trajectories consisting of precise locations, IM user trajectories consist of inferred locations with errors, making cross-platform user linking more difficult. IM user trajectories are obtained by using positioning methods to locate and track target users. Because IM platforms scramble user locations, the positioning result is an inferred location, which differs from the actual location, and this error is random. For example, in a study of WeChat user positioning, approximately 56% of positioning errors were within 60 meters, with the maximum error being approximately 112 meters. Because different IM platforms use different location scrambling strategies, even using the same positioning method, the positioning error range for users on different platforms can vary. For example, the average errors for WeChat and Momo user positioning were 56.6 meters and 15.3 meters, respectively. Therefore, IM user trajectories are only a rough reflection of user movement, and the trajectories obtained for the same user on different IM platforms can vary significantly.

[0053] Instant messaging user trajectories also suffer from data sparsity. Users determine when and how long they use instant messaging LBSD services. Furthermore, instant messaging platforms don't always display all nearby users in the "Nearby People" query list and generally limit the number of nearby users a user can query. This results in location data on trajectories often spanning large temporal and spatial spans and exhibiting uneven sparsity, leading to data sparsity. Because the number of locations on a trajectory is subjectively controlled by the user, the number of locations sometimes doesn't directly reflect their importance to the user, necessitating the identification of hidden user location features.

[0054] Cross-platform user linking based on location information is a typical method currently under study. This method often calculates user similarity based on the check-in location, but the location data of instant messaging user trajectories has errors, resulting in a low accuracy rate of user linking in this method. Figure 1 As shown, a method for accurately linking users across instant messaging platforms to combat location errors is provided, comprising:

[0055] S101. Obtain cross-instant messaging platform user data in a specified area, the cross-instant messaging platform user data including user sets and user trajectory sets in both the target platform and the platform to be linked; and perform grid transformation processing on the cross-instant messaging platform user data to obtain grid trajectories and motion state trajectories of users in each user set;

[0056] S102: Obtaining the motion state distance between the target platform user trajectory and the to-be-linked platform user trajectory under time window control, so as to obtain a candidate trajectory corresponding to the target platform user trajectory based on the motion state distance;

[0057] S103: Clustering the location points on the user trajectory according to a time window, and obtaining the cluster center distance between the target platform user trajectory and the candidate trajectory. The time window adopts a rolling time window with an adjustable step size.

[0058] S104. Weighting the cluster center distances between trajectories according to a trajectory similarity weight model to obtain the similarity between the target platform user trajectory and other candidate trajectories. The trajectory similarity weight model is constructed using grid contribution and motion sequence similarity. The grid contribution is used to characterize the amount of user trajectory feature information possessed by each grid in the grid trajectory and is obtained using grid trajectory entropy. The motion sequence similarity is obtained based on the motion state trajectory sequence and combined with the spatiotemporal co-occurrence between trajectories.

[0059] S105 , based on the similarity between the target platform user trajectory and other candidate trajectories, select the candidate trajectory with the maximum similarity as the matching trajectory of the target platform user trajectory, and link the cross-instant messaging platform users corresponding to the trajectory.

[0060] To address the problem of low accuracy of existing location-based methods in processing cross-platform instant messaging user links, the present invention proposes an accurate user linkage method across instant messaging platforms (UeLP) to combat location errors. Figure 2As shown in the figure, under the control of the perception time window, based on the distance between the location point cluster centers of the trajectories, the grid contribution and motion sequence similarity are calculated to extract the location area and temporal movement features of the trajectory, so that the spatial and temporal features of the trajectory are combined to calculate the trajectory similarity and finally complete the user link.

[0061] To facilitate reading and understanding of the above algorithm, some symbols used and the problems to be solved are described below.

[0062] Table 1 Symbols

[0063]

[0064] Instant messaging user trajectory T u :For each user u∈U, record the instant messaging user trajectory as Position points on the track The grid cell g into which it falls m Instead, the trajectory T u Transformed into user grid trajectory, denoted as A grid track is a collection of spatiotemporal grids arranged in time sequence, representing the time at time stamp t i u , user u is located in grid cell g m Inside.

[0065] Trajectory time window set W u :The trajectory time window set represents the time window set of location data on the trajectory of user u. For each user u∈U, the trajectory time window set is recorded as Among them, f time (·) is to and Δt are converted into a function of time period h and time window number k. Δt is the time window width, which is determined by the sparsity of the trajectory data and is set to 1, 5, 10 minutes in the experiment.

[0066] Location data of instant messaging user trajectories In order to accurately extract the movement pattern of the trajectory in the presence of errors, the embodiment of this case proposes a "coarse-grained motion state".

[0067] Coarse-grained motion state s: Coarse-grained motion states are categorized as stationary, moving, and ambiguous, represented by 0, 1, and 2, respectively. The "stationary" state does not mean the user remains motionless. Rather, it means that within a certain time interval, the user's movement is less than a range E, such as a grid area. In other words, when a user is "stationary," they may be moving, but their current movement pattern prevents them from exceeding range E. Correspondingly, the "moving" state refers to the user's movement exceeding range E within a time interval Δt. When there is no position data on the trajectory within the time window Δt, the motion state is ambiguous.

[0068] Time window motion state For each user u∈U, the time window motion state is expressed as It means that the coarse-grained motion state of the trajectory of user u in the kth time window of the hth period is s.

[0069] Trajectory time window set W u :The trajectory time window set represents the time window set of location data on the trajectory of user u. For each user u∈U, the trajectory time window set is recorded as Among them, f time (·) is to and Δt are converted into a function of time period h and time window number k. Δt is the time window width, which is determined by the sparsity of the trajectory data and is set to 1, 5, 10 minutes in the experiment.

[0070] Motion state trajectory Ψ u : For each user u∈U, the motion state trajectory is expressed as It is a collection of coarse-grained motion states of trajectories arranged in time sequence.

[0071] The problem to be solved can be described as: Given two user sets U of different instant messaging platforms A and U B , and the instant messaging user trajectory set D generated by users on these two platforms A and D B , in the cross-platform track collection

[0072] {(T A,i ,T B,j )|T A,i ∈D A ,T B,j ∈D B}, find all trace pairs belonging to the same user, and thus link the cross-platform user accounts belonging to the same user.

[0073] Figure 2 The algorithm steps shown can be summarized as follows:

[0074] Step 1: Data preprocessing. Preprocessing involves replacing the position points in the original trajectory with the grid they fall into to generate a grid trajectory. Focusing on grid changes, the motion state of the trajectory within the time window is extracted to generate a motion state trajectory.

[0075] Step 2: Trajectory filtering. Based on the motion state trajectories, under the control of the time window, the coarse-grained motion state distance between the target trajectory and the candidate trajectories is calculated. The candidate trajectory set is filtered according to the score, and trajectories with obvious mismatches are filtered out to generate the corresponding candidate trajectory subset.

[0076] Step 3: Distance calculation. In a time window with a dynamically adjustable step size, cluster the position points on the trajectory and calculate the cluster center distance between the target trajectory and the candidate trajectory.

[0077] Step 4: Calculate spatiotemporal weights. Construct spatiotemporal weight models for grid contribution and motion sequence similarity, assigning weights to the grid cells a trajectory passes through and the temporal similarity of the movement patterns of two different trajectories. Based on the grid trajectories, calculate the contribution weights of the grids passed by the trajectory; based on the motion state trajectories, calculate the motion sequence similarity weights between the trajectories to be compared.

[0078] Step 5: User Linking. Based on the weighted model, the cluster center distances between trajectories are weighted to obtain trajectory similarity. The trajectory with the maximum similarity is selected as the matching trajectory, and the cross-platform user accounts corresponding to the two trajectories are linked.

[0079] Specifically, the grid transformation processing of user data across instant messaging platforms can be designed to include:

[0080] Setting the spatiotemporal granularity for describing the size of the spatiotemporal grid, the spatiotemporal granularity includes the side length of the grid area and the width of the time window;

[0081] Divide the specified area into grids and obtain the grid numbers corresponding to each location point on the user's trajectory;

[0082] Sort the grids according to the time sequence to generate grid tracks;

[0083] Slide the time window and traverse the grid trajectory, record the number of grid changes of the user trajectory in each time window, and combine the motion state in the previous time window to determine the motion state of the trajectory in the current time window to generate the motion state trajectory.

[0084] The target area can be divided according to a multi-granularity spatiotemporal grid scheme. Then, the location points on the original instant messaging user trajectory are transformed into a grid to generate a grid trajectory and a motion state trajectory. The preprocessing pseudocode is shown in Algorithm 1.

[0085]

[0086] First, a multi-granularity, adaptable spatiotemporal grid scheme is established. Because the location points on instant messaging user trajectories are subject to error, to accurately describe the importance of the location area to the trajectory and the trajectory's movement patterns, it is necessary to select an appropriate spatiotemporal granularity based on the trajectory's geographic span, data collection frequency, and the algorithm's processing steps. This involves determining the size of the spatiotemporal grid, including the grid area's side length d and the rolling time window width Δt. If the grid granularity is too small, insufficient information will be gathered within the location area; if the granularity is too large, user characteristics per unit area will be unclear.

[0087] Then, the map area is divided into grids (see line 1 of the algorithm). Then, for each position point on the trajectory, the grid number g in which it falls is calculated. m (5th row), arrange the grids in time sequence to generate grid trajectory Φ u (Lines 4-10). Next, the time window is traversed through the grid trajectory, the number of grid changes in the trajectory in the current time window is recorded, and the motion state of the trajectory in the previous time window is combined to determine the motion state of the trajectory in the current time window, and the motion state trajectory Ψ is generated. u (lines 11-20).

[0088] Among them, obtaining the motion state distance between the target platform user trajectory and the to-be-linked platform user trajectory under the time window control may include:

[0089] Compare the trajectory of each target user in the target platform with the trajectory of the user in the platform to be linked, and obtain the trajectory motion state distance between the trajectory of each target user in the target platform and the trajectory of the user in the platform to be linked;

[0090] Obtain a trajectory filtering score based on the trajectory motion state distance and the set and number of position co-occurrence time windows in the trajectory. The trajectory filtering score is used to describe the mean difference in motion state of the matching trajectories to be compared during the spatiotemporal co-occurrence of the position points.

[0091] According to the trajectory filtering score, the user trajectories in the platform to be linked that do not match the target user trajectory are deleted to obtain the candidate trajectories corresponding to the target user trajectories on the target platform.

[0092] In the trajectory set D A and D B When you link a user in |D A |×|D B | comparison. To improve efficiency, D A Each trajectory T in A , filter out the candidate trajectory set D B The trajectory that is obviously mismatched in the , generates a candidate trajectory subset D ′ BThe main steps of filtering are summarized as follows:

[0093] Assume T A ∈D A is the target trajectory, and compare it with each T B ∈D B The motion state trajectory is compared with the motion state trajectory, and the distance of the motion state of the trajectory is calculated using formula (1):

[0094]

[0095] Here, the state distance calculation is allowed only when the state is clear, that is, 0 or 1. According to the motion state distance, the filtering score will be calculated using formula (2) to achieve a more accurate filtering effect.

[0096]

[0097] Among them, W A ∩W B It is the time window set where the location point data exists simultaneously in the two trajectories, which is called the location co-occurrence time window set. A ∩W B | represents the trajectory T A and T B The number of time windows of position co-occurrence between two trajectories reflects the number of spatiotemporal co-occurrences between the two trajectories. The filtering score describes the mean difference in the motion state of the two trajectories during the period of spatiotemporal co-occurrence of the positions. For example, trajectory T A 、T B,1 and T B,2 After the preprocessing of Algorithm 1, the corresponding motion state trajectory Ψ is obtained A ,Ψ B,1 and Ψ B,2 , the motion state trajectory is shown in Table 2. Table 3 describes the trajectory T A The situation and calculation results of the position co-occurrence time window between the other two trajectories.

[0098] Table 2 Trajectory examples

[0099]

[0100] Table 3 Example of location co-occurrence time window

[0101]

[0102] According to the sample data in Table 2 and Table 3, the filtering scores of the two groups of trajectory pairs are calculated according to formula (2): Track T A and T B,1 The filter score is higher.

[0103] Among them, clustering the location points on the user trajectory according to the time window can be designed to include:

[0104] Constructing a perception time window with a dynamically adjustable step size, and determining the time window width based on the distribution of the perception location data. The time window width includes the spatiotemporal grid time granularity and the rolling step size;

[0105] The rolling step of the time window is dynamically adjusted according to the time interval between the trajectory segments of the target platform user trajectory, and the cluster center distance between the location points between the target platform user trajectory and the to-be-linked platform user trajectory in the corresponding time window is obtained.

[0106] Taking the distance between trajectories as the basic element for similarity judgment, the distance calculation can be divided into two stages.

[0107] Phase 1: Constructing a perception time window with dynamically adjustable step size. Due to data missing issues, the temporal and spatial spans of adjacent location points in instant messaging user trajectories are often large. To improve the efficiency of trajectory comparison, a perception time window with dynamically adjustable step size is constructed. By sensing the distribution of location data, the width of the time window, i.e., the time granularity of the spatiotemporal grid, and the rolling step size are determined, such as Figure 3 When calculating the distance, the target trajectory T A As a benchmark, according to T A The time interval between the track segments is used to dynamically adjust the scrolling step of the window.

[0108] The second stage: Calculate the distance between the cluster centers of the position points between the trajectories. Under the control of the rolling time window, the basic similarity of the two trajectories is calculated based on the distance between the cluster centers of the position points between the trajectories, as shown in formula (3):

[0109]

[0110] in,

[0111]

[0112]

[0113] Where (h,k)∈W A ∩W B , K T (h,k,Δt)K T The (h,k,Δt) function is used to calculate the center coordinates of all position clusters of trajectory T in the kth time window of period h, N h is the number of time periods involved in distance calculation, N kis the number of time windows involved in the distance calculation within time period h, r is the radius of the Earth, and the distance is in meters. To more accurately characterize the differences in distances in the similarity function Sim(·), the parameter μ can be adjusted to reduce the distance value, achieving a more refined similarity calculation.

[0114] Specifically, the cluster center distances between trajectories are weighted according to the trajectory similarity weight model to obtain the similarity between the target platform user trajectory and other candidate trajectories. The model can be designed to include:

[0115] Obtain the basic similarity between the target platform user trajectory and the to-be-linked platform user trajectory based on the cluster center distance between the trajectories;

[0116] The trajectory entropy is obtained based on the number of grid cells contained in the grid trajectory and the number of location points where the user trajectory falls into each grid. The contribution of the grid in the grid trajectory is obtained based on the trajectory entropy.

[0117] Based on the motion state trajectory and the spatiotemporal co-occurrence of the trajectories between the target platform and the platform to be linked, the motion sequence similarity between the trajectory of the user on the target platform and the trajectory of the user on the platform to be linked is obtained;

[0118] The similarity between the user trajectory of the target platform and the user trajectory of the platform to be linked is obtained based on basic similarity, grid contribution and motion sequence similarity.

[0119] User behavior has certain regularities and uniqueness, and trajectories are the product of individual user behavior. Therefore, identifying the important locations and movement pattern characteristics of the trajectory helps to characterize user behavior. Trajectory features are extracted using grid contribution and motion sequence similarity.

[0120] Due to different personal living habits and movement patterns, the frequency of occurrence of each location in the trajectory is different. It is generally believed that the more frequently a location appears in a trajectory, the greater the importance it carries. However, sometimes a high frequency only means that the location is important, but the location may not be able to depict more user characteristics. For example, "home" is an important location, but when the trajectory is composed entirely of the location of home, the amount of information about the user's characteristics contained in the trajectory is very small, and it is difficult to distinguish it from people living nearby. Inspired by information entropy and grid Shannon entropy, the calculation formula for grid-based trajectory entropy can be expressed as:

[0121]

[0122] Where |T| is the number of position points contained in the trajectory T, that is, the number of grid cells contained in the corresponding grid trajectory, N T (g m ) is the trajectory T that falls into the grid g mThe larger the trajectory entropy value, the more information related to user features the trajectory contains. Based on the trajectory entropy, the calculation formula of the grid contribution can be expressed as follows:

[0123]

[0124] The grid contribution is a product term of formula (6), and the grid contribution C T (g m ) is larger, the grid g m The more information that contributes to the trajectory entropy of trajectory T, the more information the grid has about user features. Based on the grid contribution, the trajectory similarity can be redefined as:

[0125]

[0126] g i Representative trajectory T A The cluster center coordinates K TA The grid number that (h, k, Δt) falls into; accordingly, g j is the trajectory T B The cluster center position K TB The grid number where (h,k,Δt) is located.

[0127] In the similarity calculation, since the grid contribution emphasizes the position characteristics of the trajectory again, in order to prevent the target trajectory from being misjudged with other trajectories that highly overlap with its activity area, formula (8) introduces the function f num (·), which is used to control the impact of the number of spatiotemporal co-occurrences between two trajectories on the similarity calculation.

[0128] When the activity areas between trajectories have a high degree of overlap, such as for students in a school or employees in a company, their trajectories are often difficult to distinguish. However, everyone has their own unique life patterns and habits, which makes the temporal movement patterns of user trajectories different. In the case of data errors, the coarse-grained motion state is used to generate the motion state sequence of the trajectory. Combined with the spatiotemporal co-occurrence between trajectories, the user's private mobility pattern is extracted. The motion sequence similarity of two trajectories is calculated using formula (11).

[0129]

[0130] As mentioned above, since the number of location points on the instant messaging user trajectory does not fully represent the importance of the location to the user, the parameter θ is used to adjust the number of co-occurrence time windows |W A ∩W B | Similarity of motion sequence M simBased on the similarity of motion sequences, the trajectory similarity calculation formula can be redefined as follows:

[0131]

[0132] Grid contribution and motion sequence similarity are fine-grained extractions of the location and mobility pattern features of trajectory users. When calculating these two weights, a smaller grid system can be used. For example, in the experiment, the grid size is set to 220 meters. Unlike formulas (3) and (8), formula (12) introduces grid contribution when calculating trajectory similarity, which more effectively distinguishes the contribution of different location areas on the trajectory to the trajectory identification of user features. At the same time, formula (12) also combines motion sequence similarity and considers the similarity of mobility patterns between different user trajectories. This enables the trajectory similarity calculation to more comprehensively characterize user features in terms of both geography and time sequence, thereby improving the accuracy of trajectory matching.

[0133] After constructing the weights based on grid contribution and motion sequence similarity, for the trajectory set D A Each trajectory T in A , and the candidate trajectory subset D′ B For each trajectory in , calculate the trajectory similarity according to formula (12). In the similarity calculation results, select the top n trajectories as the winning trajectories and regard them as the target trajectory T A The matching trajectory of A The account of the winning track is associated with the account information of the winning track, thereby completing the link between cross-platform users.

[0134] In summary, after Figure 2 The IMUL algorithm in this case completes cross-platform user linking through the five steps in [1]. The pseudo code of the method can be seen in Algorithm 2. It includes data preprocessing (line 2), filtering the trajectory library according to the target trajectory (line 4), calculating the weights including grid contribution (lines 6-7) and motion sequence similarity (lines 8-9), calculating trajectory similarity (line 10), and finally selecting the top n users with the highest similarity values ​​to link with the target user (lines 12-13).

[0135]

[0136] Furthermore, based on the above method, an embodiment of the present invention also provides a cross-platform precise user linking system for combating location errors, comprising: a cross-platform user data acquisition module, a candidate user trajectory acquisition module, a trajectory location point clustering module, a trajectory similarity calculation module, and a cross-platform user linking module, wherein:

[0137] A cross-platform user data acquisition module is used to acquire cross-instant messaging platform user data in a specified area, wherein the cross-instant messaging platform user data includes user sets and user trajectory sets from both the target platform and the platform to be linked; and to perform grid transformation processing on the cross-instant messaging platform user data to obtain the grid trajectory and motion state trajectory of the users in each user set;

[0138] The candidate user trajectory acquisition module is used to obtain the motion state distance between the target platform user trajectory and the to-be-linked platform user trajectory under time window control, so as to obtain the candidate trajectory corresponding to the target platform user trajectory based on the motion state distance;

[0139] The trajectory location point clustering module is used to cluster the location points on the user trajectory according to the time window, and obtain the cluster center distance between the target platform user trajectory and the candidate trajectory. The time window adopts a rolling time window with adjustable step size;

[0140] The trajectory similarity calculation module is used to weight the cluster center distances between trajectories according to a trajectory similarity weight model to obtain the similarity between the target platform user trajectory and other candidate trajectories. The trajectory similarity weight model is constructed using grid contribution and motion sequence similarity. The grid contribution is used to characterize the amount of user trajectory feature information possessed by each grid in the grid trajectory and is obtained using grid trajectory entropy. The motion sequence similarity is obtained based on the motion state trajectory sequence and the temporal and spatial co-occurrence between trajectories.

[0141] The cross-platform user linking module is used to select the candidate trajectory with the maximum similarity between the target platform user trajectory and other candidate trajectories as the matching trajectory of the target platform user trajectory, and link the cross-instant messaging platform users corresponding to the trajectory.

[0142] To verify the effectiveness of this solution, the following is a further explanation based on experimental data:

[0143] We conduct an experiment on instant messaging user linking on 7 datasets, including 6 public datasets and 1 self-built dataset, and compare it with three existing typical cross-platform user linking methods. R Conduct experimental verification.

[0144] Due to the difficulty in obtaining real-world cross-platform instant messaging user trajectories, the experiment first used the public Geolife dataset from Microsoft Research Asia for validation. This dataset consists of 17,621 trajectories, each consisting of a series of timestamped location points, recording users' daily, real-life movements across various transportation modes. Most trajectories are composed of multiple segments of varying lengths, with the intervals between segments varying from a dozen minutes to several hours. To fully validate the algorithm's performance, Geolife was sequentially split into two datasets: Sub-Geolife and Mix-Sub-Geolife.

[0145] (1) Sub-Geolife dataset

[0146] Randomly select a track T from Geolife i , put T i The position points on the dataset are arranged in time sequence and numbered. Then, according to the operations shown in Table 4, the trajectory is split into two parts, which are regarded as two cross-platform trajectory sets, and datasets G1, G2 and G3 are generated in sequence.

[0147] Table 4 Dataset generation

[0148]

[0149] In the process of splitting the data set, the temporal and spatial spans of adjacent position points on the trajectory are gradually enlarged, and the sparsity of the position data is gradually increased, so as to compare and analyze the performance of the algorithm under different data sparsity levels.

[0150] (2) Mix-Sub-Geolife dataset

[0151] To further validate the algorithm's performance in the presence of data errors, we continued to reduce the data sampling rate while adding random errors to each location data point. This resulted in the Mix-Sub-Geolife dataset, which includes G4, G5, and G6. The process is shown in Table 4. The average positioning errors for WeChat and Momo users were 56.6 meters and 15.3 meters, respectively. The experiment selected 80 meters and 20 meters as the positioning error ranges for the two platforms, respectively, to generate dataset G4.

[0152] During the experiment, 10,500 trajectories were randomly selected from the Geolife dataset, and 10,500 cross-platform trajectory pairs (T A ,T B ). Then, we randomly select 10,000 trajectory pairs as positive samples, i.e., trajectory pairs with matching trajectories. In the remaining 500 trajectory pairs, we remove the trajectory T from platform A.A , retain the trajectory T of platform B B As negative samples, they are used to interfere with trajectory matching and test the performance of the user linking method. The data set is shown in Table 5.

[0153] Table 5 Description of the dataset

[0154]

[0155] In Table 5, in order to highlight the sparsity of the position data, the average number of position points in the trajectory is used to describe the sparsity of the data on each dataset and its changes. In order to prevent the data from being too sparse and causing the trajectory to remain unchanged before and after splitting, the original trajectory T i They all consist of more than 10 position points. However, datasets G1 to G5 do not delete any sparse trajectories, that is, trajectories containing only 1 to 2 position data.

[0156] We recruited 30 volunteers from the self-built dataset and completed the collection of the real dataset. We selected WeChat and Momo as two different instant messaging platforms, developed the TrigLBSD.APP application, and installed it on the experimental mobile phones. The function of the TrigLBSD application is to randomly refresh the current location and start a round of LBSD service on WeChat or Momo. In the experiment, volunteers carried the experimental mobile phones in the urban area of ​​Zhengzhou (about 1200km 2 ) activities, daily life activities, such as walking, running, riding, staying, etc. At the same time, an Android simulator was run on the computer, and virtual probes were deployed using virtual location software to locate and track the volunteers. From September to November 2021, a total of 2679 sets of cross-platform trajectory pairs (T W ,T M ). Among them, T W and T M They represent the WeChat and Momo user trajectories of the same target user on the same day. In the collected data, 2400 sets of trajectory pairs were randomly selected as positive samples, and the WeChat user trajectories in the trajectory pairs were extracted to form a trajectory library. Momo user trajectory composition Then, randomly select 120 pairs of the remaining trajectory pairs and put the Momo user trajectories into the trajectory library. As negative samples. S In the dataset, each trajectory contains 86.84 location points on average.

[0157] Experiments were conducted on the above-mentioned public datasets and self-built datasets, and the relevant experimental parameter settings are shown in Table 6.

[0158] Table 6 Experimental parameter setting description

[0159]

[0160] In Table 6, the grid side length d determines the grid trajectory, the coarse-grained motion state of the trajectory, and the generated motion state trajectory. As mentioned above, in the filtering stage, d is taken as 330 meters, while when calculating the weight, d is taken as 220 to extract user features at different granularities. In formula (3), μ is used to refine the impact of distance value on similarity. β and γ appear in formula (10), and θ appears in formula (11). Because the number of occurrences in a certain location area on the instant messaging user trajectory cannot fully reflect the user's actual activity status, these three parameters are used to adjust the impact of the number of spatiotemporal co-occurrence windows between two trajectories on the similarity calculation. In the experiment, the parameters were set based on experience.

[0161] The accuracy rate is used as the evaluation index of the effectiveness of the solution in this case, and the effectiveness of the algorithm filtering stage is evaluated by filtering efficiency.

[0162] (1) Accuracy: Since the thresholds of different methods are different, similar to the user link article, the first n values ​​of the similarity calculation results (n≤|D|) are taken as the algorithm result, and the user link accuracy is recorded as Acc@n:

[0163]

[0164] Where D represents the target trajectory library, Indicates whether the matching trajectory exists in the top-n (n≤|D|) list.

[0165] (2) Filtering efficiency: Some existing works usually use hit rate as an indicator to evaluate the effectiveness of filtering algorithms. However, the hit rate only considers whether the candidate subset of the filtered trajectory contains matching trajectories, ignoring the impact of the candidate subset size on the subsequent similarity calculation overhead. Although a larger trajectory candidate subset improves the hit rate, it also causes more trajectories to participate in the subsequent similarity calculation, thereby increasing the computational and time costs. Therefore, in order to balance the hit rate and the candidate subset size, in order to achieve a balance between the accuracy of user links and computational efficiency, the hit rate is improved and the filtering efficiency is proposed, as shown in Equation (15).

[0166]

[0167] Among them, D A represents the target trajectory library, H represents D A The number of trajectories hit, H / |D A | is the hit rate. The two proportional coefficients need to satisfy: λ1>0, λ2>0, and λ1+λ2=1.

[0168] The experimental results and analysis are as follows:

[0169] 1. Selection of subscale parameters

[0170] After filtering, the original candidate trajectory library D B will retain α|D B | trajectories, generate candidate subset D′ B The candidate subset size α determines the number of trajectories involved in trajectory similarity calculation, which directly affects the performance of the user linking method. In order to select a suitable subset size parameter α, experiments were conducted on 7 datasets, and the results are shown in the following table. Figure 4 and 5 shown.

[0171] Figure 4 and Figure 5 The changes in filtering efficiency and hit rate are shown respectively when the subset size α changes from 0 to 1 with a span of 0.1. Figure 4 It shows that as the subset size α increases, the filtering efficiency decreases on most data sets. Figure 5 It can be seen that as α increases, the hit rate gradually increases, but the candidate subset D′ B The number of trajectories in also increases, which increases the cost of subsequent calculations. Figure 4 and Figure 5 It can be seen that on most data sets, the filtering efficiency is higher when α is 0.1 than when it is other values, and the hit rate is not much lower when α is 0.1 and 0.2 than when it is 1.0. Therefore, it is recommended to set α to 0.1-0.2.

[0172] The experiment also analyzes the impact of different subset sizes α on the user link results of the proposed method. The experimental results are as follows: Figure 6 As shown in Figure 2, when the subset size α changes from 0 to 1 in steps of 0.1, the accuracy of user links first decreases and then stabilizes as α increases. On most datasets, α of 0.1 achieves the highest accuracy, so we set α to 0.1 in our experiments.

[0173] 2. Comparison results of user link experiments

[0174] In the datasets G1 to G6 and the self-built G S We conducted cross-platform user link experiments on seven datasets. The experimental results of this algorithm were compared with three typical methods, including k-BCT, GS, and TF-IWF. Figure 7 , as shown in Tables 7 and 8.

[0175] Table 7 Acc@1 comparison results of this algorithm and three baseline methods (%)

[0176]

[0177] Table 8 Acc@3 comparison results of this algorithm and three baseline methods (%)

[0178]

[0179] The bold part in the table is the highest value in the comparison results. Figure 7 As shown in Tables 7 and 6, our algorithm outperforms the three existing baseline methods on all seven datasets. Across the seven datasets, the Acc@1 accuracy is 32.86%, 44.8%, and 14.4% higher than the three baseline methods, respectively. The Acc@3 accuracy is 30.52%, 34.67%, and 13.19% higher than the three baseline methods, respectively.

[0180] The improvement in user linking accuracy is primarily due to the fact that this algorithm simultaneously selects both spatial and temporal features of trajectories, making it more user-specific than the baseline method. The baseline method extracts trajectory features through factors such as the shortest distance between locations and the number of co-occurrences of locations. It focuses solely on the role of certain specific location areas on the trajectory, ignoring other locations in the trajectory and the connections between them. It also fails to utilize the temporal features of the trajectory, thus affecting the accuracy of user links across instant messaging platforms. This algorithm, on the other hand, extracts trajectory location area features based on grid contribution and also extracts temporal motion state sequence features of the trajectory through changes between locations. This results in more user-specific location and movement pattern features obtained even when there are errors in the location data.

[0181] 3. Analysis of the impact of data sparsity and error on experimental results

[0182] Further inspection of Tables 7 and 8 reveals that the accuracy of each method on datasets G4, G5, and G6, which contain erroneous location data, is significantly lower than that of datasets G1, G2, and G3, which contain accurate location data. Furthermore, the performance of each method drops sharply on dataset G5. For example, in G5, the accuracy Acc@1 of the k-BCT, GS, TF-IWF, and our algorithm is 61.64%, 41.58%, 53.7%, and 32.6% lower than in G1, respectively. Acc@3 is 55.84%, 46.68%, 43.36%, and 18.34% lower, respectively. Analysis reveals that the significant decline in user linking accuracy is primarily due to the inherent erroneous location data in dataset G5, which also has the highest data sparsity. This demonstrates that data erroneousness and sparsity significantly impact user linking performance.

[0183] In order to analyze the impact of data sparsity on the performance of the method, the sparsity of the trajectory data in the dataset is calculated according to the following formula:

[0184]

[0185] Among them, N t It represents the number of trajectories with a specific number of position points in the dataset D. The sparsity of the dataset is shown in Table 9.

[0186] Table 9 Trajectory sparsity of the dataset (%)

[0187]

[0188]

[0189] Table 9 shows that the sparsity of trajectory data increases from datasets G1 to G5, reaching its highest level in G5. For example, in G5, trajectories containing 5 or fewer points account for 34.56%, 5.86 times that of G4; trajectories containing fewer than 10 points account for 54.8%, 3.83 times that of G4. Combining Tables 6 and 5, we can see that as dataset sparsity increases, the performance of each method decreases, reaching its lowest point in dataset G5. The performance of the k-BCT method on dataset G1 is worse than that on G2 and G3. This is because the activity areas of the trajectories in dataset G1 are highly overlapping and the position data on the trajectories is dense, making misjudgments more likely for methods like k-BCT, which rely solely on distance. Since our algorithm significantly outperforms the other three baseline methods and achieves significantly higher accuracy on G5, this demonstrates our approach's robustness against data sparsity and interference.

[0190] The above data shows that this solution uses multi-granularity spatiotemporal grids to organize data, dynamically adjusts the time window sliding step, and extracts the characteristics of trajectory users' location areas and temporal mobility patterns based on the distance between cluster centers under the time window through grid contribution and motion sequence similarity, thereby improving the accuracy of cross-platform user links. The filtering algorithm based on the coarse-grained motion state of the trajectory not only improves the accuracy of user links, but also reduces the computational and time overhead of comparing the similarity with candidate trajectories, and has good application prospects in the field of cross-platform user links.

[0191] Unless otherwise specifically stated, the relative steps, numerical expressions and values ​​of the components and steps set forth in these embodiments do not limit the scope of the present invention.

[0192] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Reference can be made to the common and similar parts between the various embodiments. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the method description.

[0193] The units and method steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the components and steps of each example have been generally described in terms of function in the above description. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. A person of ordinary skill in the art may use different methods to implement the described functions for each specific application, but such implementation is not considered to be beyond the scope of the present invention.

[0194] Those skilled in the art will appreciate that all or part of the steps in the above method can be performed by a program instructing related hardware. The program can be stored in a computer-readable storage medium, such as a read-only memory, a magnetic disk, or an optical disk. Alternatively, all or part of the steps in the above embodiment can be implemented using one or more integrated circuits. Accordingly, each module / unit in the above embodiment can be implemented in the form of hardware or software functional modules. The present invention is not limited to any specific combination of hardware and software.

[0195] Finally, it should be noted that the above-described embodiments are only specific implementation methods of the present invention, which are used to illustrate the technical solutions of the present invention, rather than to limit them. The scope of protection of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the above-described embodiments, those skilled in the art should understand that any person skilled in the art can modify or easily conceive of changes to the technical solutions described in the above-described embodiments within the technical scope disclosed by the present invention, or replace some of the technical features therein with equivalents. Such modifications, changes, or replacements do not deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should be included in the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be subject to the scope of protection of the claims.

Claims

1. A method for accurately linking users across instant messaging platforms to combat location errors, characterized in that: Include: Obtaining cross-platform user data in a specified area, the cross-platform user data including user sets and user trajectory sets in both the target platform and the platform to be linked; and performing grid transformation on the cross-platform user data to obtain grid trajectories and motion state trajectories of users in each user set; Under the time window control, the motion state distance between the target platform user trajectory and the to-be-linked platform user trajectory is obtained, so as to obtain the candidate trajectory corresponding to the target platform user trajectory according to the motion state distance; Cluster the location points on the user trajectory according to the time window, and obtain the cluster center distance between the target platform user trajectory and the candidate trajectory. The time window adopts a rolling time window with adjustable step size. The cluster center distances between trajectories are weighted according to a trajectory similarity weight model to obtain the similarity between the target platform user trajectory and other candidate trajectories. The trajectory similarity weight model is constructed using grid contribution and motion sequence similarity. The grid contribution is used to characterize the amount of user trajectory feature information possessed by each grid in the grid trajectory and is obtained using grid trajectory entropy. The motion sequence similarity is obtained based on the motion state trajectory sequence and the spatiotemporal co-occurrence between trajectories. According to the similarity between the target platform user trajectory and other candidate trajectories, the candidate trajectory with the maximum similarity between trajectories is selected as the matching trajectory of the target platform user trajectory, and the cross-instant messaging platform users corresponding to the trajectory are linked.

2. The method for accurately linking users across instant messaging platforms to combat location errors according to claim 1, characterized in that: Perform grid transformation on user data across instant messaging platforms, including: Setting the spatiotemporal granularity for describing the size of the spatiotemporal grid, the spatiotemporal granularity includes the side length of the grid area and the width of the time window; Divide the specified area into grids and obtain the grid numbers corresponding to each location point on the user's trajectory; Sort the grids according to the time sequence to generate grid tracks; Slide the time window and traverse the grid trajectory, record the number of grid changes of the user trajectory in each time window, and combine the motion state in the previous time window to determine the motion state of the trajectory in the current time window to generate the motion state trajectory.

3. The method for accurately linking users across instant messaging platforms to combat location errors according to claim 1, characterized in that: Obtain the motion state distance between the target platform user trajectory and the to-be-linked platform user trajectory under time window control, including: Compare the trajectory of each target user in the target platform with the trajectory of the user in the platform to be linked, and obtain the trajectory motion state distance between the trajectory of each target user in the target platform and the trajectory of the user in the platform to be linked; Obtain a trajectory filtering score based on the trajectory motion state distance and the set and number of position co-occurrence time windows in the trajectory. The trajectory filtering score is used to describe the mean difference in motion state of the matching trajectories to be compared during the spatiotemporal co-occurrence of the position points. According to the trajectory filtering score, the user trajectories in the platform to be linked that do not match the target user trajectory are deleted to obtain the candidate trajectories corresponding to the target user trajectories on the target platform.

4. The method for accurately linking users across instant messaging platforms to combat location errors according to claim 1, characterized in that: Clustering of location points on user trajectories based on time windows, including: Constructing a perception time window with a dynamically adjustable step size, and determining the time window width based on the distribution of the perception location data. The time window width includes the spatiotemporal grid time granularity and the rolling step size; The rolling step of the time window is dynamically adjusted according to the time interval between the trajectory segments of the target platform user trajectory, and the cluster center distance between the location points between the target platform user trajectory and the to-be-linked platform user trajectory in the corresponding time window is obtained.

5. The method for accurately linking users across instant messaging platforms to combat location errors according to claim 1, characterized in that: The cluster center distances between trajectories are weighted according to the trajectory similarity weight model to obtain the similarity between the target platform user trajectory and other candidate trajectories, including: Obtain the basic similarity between the target platform user trajectory and the to-be-linked platform user trajectory based on the cluster center distance between the trajectories; The trajectory entropy is obtained based on the number of grid cells contained in the grid trajectory and the number of location points where the user trajectory falls into each grid. The contribution of the grid in the grid trajectory is obtained based on the trajectory entropy. Based on the motion state trajectory and the spatiotemporal co-occurrence of the trajectories between the target platform and the platform to be linked, the motion sequence similarity between the trajectory of the user on the target platform and the trajectory of the user on the platform to be linked is obtained; The similarity between the user trajectory of the target platform and the user trajectory of the platform to be linked is obtained based on basic similarity, grid contribution and motion sequence similarity.

6. The method for accurately linking users across instant messaging platforms to combat location errors according to claim 5, characterized in that: The similarity calculation formula between trajectories is expressed as: Among them, T A 、T B are the target platform user trajectory and the to-be-linked platform user trajectory, μ is the adjustment parameter used to reduce the distance value, D dist (T A ,T B ) represents the distance between trajectories, W A ∩W B is the set of time windows where positions co-occur in two trajectories, f num (·) is a function that controls the influence of the spatiotemporal co-occurrence times between two trajectories on the similarity calculation, M sim (T A ,T B ) is the function for calculating the similarity of motion sequences between trajectories.

7. The method for accurately linking users across instant messaging platforms to combat location errors according to claim 1, characterized in that: The candidate trajectory with the maximum similarity between trajectories is selected as the matching trajectory of the target platform user trajectory, including: The similarities between the trajectories of each user on the target platform and the candidate trajectories are sorted, and several user trajectories with the highest similarity rankings are selected from the candidate trajectories as the winning trajectories that match the corresponding user trajectories on the target platform.

8. A precise user linking system across instant messaging platforms for combating location errors, characterized in that: It includes: cross-platform user data acquisition module, candidate user trajectory acquisition module, trajectory location point clustering module, trajectory similarity calculation module and cross-platform user link module, among which, A cross-platform user data acquisition module is used to acquire cross-instant messaging platform user data in a specified area, wherein the cross-instant messaging platform user data includes user sets and user trajectory sets from both the target platform and the platform to be linked; and to perform grid transformation processing on the cross-instant messaging platform user data to obtain the grid trajectory and motion state trajectory of the users in each user set; The candidate user trajectory acquisition module is used to obtain the motion state distance between the target platform user trajectory and the to-be-linked platform user trajectory under time window control, so as to obtain the candidate trajectory corresponding to the target platform user trajectory based on the motion state distance; The trajectory location point clustering module is used to cluster the location points on the user trajectory according to the time window, and obtain the cluster center distance between the target platform user trajectory and the candidate trajectory. The time window adopts a rolling time window with adjustable step size; The trajectory similarity calculation module is used to weight the cluster center distances between trajectories according to a trajectory similarity weight model to obtain the similarity between the target platform user trajectory and other candidate trajectories. The trajectory similarity weight model is constructed using grid contribution and motion sequence similarity. The grid contribution is used to characterize the amount of user trajectory feature information possessed by each grid in the grid trajectory and is obtained using grid trajectory entropy. The motion sequence similarity is obtained based on the motion state trajectory sequence and the temporal and spatial co-occurrence between trajectories. The cross-platform user linking module is used to select the candidate trajectory with the maximum similarity between the target platform user trajectory and other candidate trajectories as the matching trajectory of the target platform user trajectory, and link the cross-instant messaging platform users corresponding to the trajectory.

9. An electronic device, characterized in that: include: at least one processor, and a memory coupled to the at least one processor; The memory stores a computer program, and the computer program can be executed by the at least one processor to implement the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, and when the computer program is executed, the method according to any one of claims 1 to 7 can be implemented.

Images