Staying activity place analysis method based on user behaviors

Abstract

The invention relates to a staying activity place analysis method based on user behaviors. The method comprises the following steps: acquiring staying point data of a user in a physical space; associating external geographic information data on the basis of the stay point data, and obtaining marked site function attribute data; extracting a time transfer sequence between the places from the marked place function attribute data, and processing the transfer sequence by adopting a sequence modeling algorithm to obtain a predicted moving mode path; wherein the moving mode path reflects a site association rule; and judging and correcting the predicted moving mode path, and generating a user behavior dynamic report. By integrating staying and moving details, a user behavior rule is accurately described, the accuracy and integrity of trajectory analysis are remarkably improved, and the method is suitable for scenes such as intelligent navigation and behavior prediction.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of user behavior analysis, in particular to a stay activity location analysis method based on user behavior. BACKGROUND

[0002] The analysis of user behavior stay activity location is an indispensable field in today's digital age, and its importance lies in providing key basis for urban planning, business operation, personalized services, etc. by understanding the activity patterns of users in physical space. Whether it is to optimize traffic flow, accurately push business services, or improve public safety management, it is crucial to understand where users stay and how they move. This field reveals the living habits and behavior patterns of users by analyzing their location data, providing a solid foundation for intelligent decision-making. However, current research and application still faces significant challenges and needs to be broken through to meet the increasingly complex needs. Existing methods are often limited by the lack of capturing the dynamics of user behavior when analyzing user stay activity location. Many solutions rely on simple aggregation of positioning data, making it difficult to accurately distinguish the behavior intention of users at different locations. For example, the system may only record that the user appears at a certain location, but cannot determine whether the location is a brief passing or an important activity place such as a workplace or a frequently visited leisure place. In addition, existing methods often fail to fully present the whole picture of user behavior due to the lack of in-depth mining of the correlation between locations when processing user movement trajectories. These limitations make the analysis results difficult to meet the fine-grained needs in practical applications. In terms of technology, the core difficulty lies in accurately positioning the key stay locations of users and reconstructing their complete activity trajectories. First, identifying key stay locations requires the system not only to detect the time and frequency of user stays, but also to understand the functional attributes of the locations, such as distinguishing between home addresses and workplaces. This requires the system to extract location labels with semantic meaning from massive location data, rather than just geographical coordinates. Further, trajectory reconstruction based on these key locations faces greater challenges, as user movement patterns between different locations are often influenced by multiple factors such as time and context. Existing technologies fail to effectively integrate these dynamic factors, resulting in broken or biased trajectory reconstruction. For example, a user may frequently commute between home and office during weekdays, but visit multiple business places on weekends. If the system cannot accurately associate the functions of these locations with the movement patterns, it cannot restore the complete activity path. Therefore, how to accurately identify the functional attributes of key stay locations of users in complex and variable location data, and seamlessly reconstruct dynamic activity trajectories based on these locations, becomes a key problem in this research. SUMMARY

[0003] The purpose of the present application is to provide a stay activity location analysis method based on user behavior to solve the problems existing in the prior art.

[0004] To achieve the above objectives, the present invention provides the following solution: Methods for analyzing user behavior-based activity locations include: Obtain data on user dwell times in the physical space; Based on the aforementioned stop point data, external geographic information data is correlated to obtain the functional attribute data of the marked locations; Temporal transition sequences between locations are extracted from the labeled location functional attribute data. Sequence modeling algorithms are used to process the transition sequences to obtain predicted movement pattern paths. The movement pattern paths reflect the location association patterns. The system analyzes and corrects predicted mobile patterns and paths, generating dynamic reports on user behavior.

[0005] Optionally, obtaining user dwell time data in physical space includes: Collect user location data in physical space, and use clustering algorithms to group the coordinate sequences to obtain a preliminary set of dwell points; Based on the timestamp information in the initial set of stops, the duration and frequency of each stop are calculated to determine the list of key stops, which includes locations with high frequency.

[0006] Optionally, collecting user location data in physical space includes: The system acquires user location data from physical space using sensors or positioning devices, and generates a coordinate sequence that includes timestamps.

[0007] Optionally, the list of key stops includes: By parsing the initial set of stop points, timestamp data is obtained; If the timestamp data is complete, group by stop point, calculate the entry and exit time difference for each stop point, and obtain the duration; Based on the duration, a time window is used to divide the visit into segments, and the number of visits at each stop point is counted to determine the visit frequency; If the access frequency exceeds the preset threshold, the corresponding stop point will be marked as a high-frequency point; By using cluster analysis and the K-means algorithm to group high-frequency points, key location data is obtained. Based on key location data, combined with access frequency and duration, a weighted sorting algorithm is used to determine the location ranking; By sorting the locations, an ordered list of key stops is generated.

[0008] Optionally, obtaining the location functional attribute data for the labeled location includes: For each location of stay, location data is obtained from an external geographic information database, and geocoding technology is used to match latitude and longitude with location descriptions to generate a set of location geographic information. The dwell time is extracted from the location geographic information set. If the dwell time exceeds a preset threshold, it is marked as a potential workplace and a list of potential workplaces is generated. For the list of potential workplaces, analyze whether the dwell time falls within the working hours. If the dwell time matches the working hours, the location is confirmed as a workplace, and a workplace set is generated. Dwell frequency is extracted from the set of workplaces, and the number of dwell times per unit time is calculated using frequency statistics methods to obtain the dwell frequency distribution; Based on the frequency distribution of stay and combined with the location geographic information, the decision tree algorithm is used to classify the functional attributes of the locations and generate a set of location functional attributes. Extract locations labeled as workplaces from the set of location functional attributes, and generate the final labeled location functional attribute data.

[0009] Optionally, obtaining the predicted movement pattern path includes: A long short-term memory network algorithm is used to model the time transition sequence and generate a location transition relationship model; The path prediction result is obtained by predicting the location transfer probability in subsequent time steps using a location transfer relationship model. High-probability location shift sequences are extracted from the path prediction results to generate movement pattern paths; Based on the analysis of the correlation patterns between locations using mobile path analysis, the correlation patterns between locations are determined. If the confidence level of the location association pattern is lower than the preset threshold, the sequence modeling algorithm is iteratively optimized to regenerate the movement pattern path.

[0010] Optionally, judging and correcting the predicted movement pattern path includes: Based on the predicted movement pattern path, analyze the discontinuities in the path. If the time difference of the discontinuities is less than a preset threshold, fill the discontinuities using an interpolation method to determine a continuous sequence of trajectory segments. For a continuous sequence of trajectory segments, semantic tags from location functional attributes are integrated to determine whether the sequence of trajectory segments conforms to behavioral intent logic. If it does not conform, deviation points in the sequence are adjusted to obtain a corrected complete trajectory. A dynamic report on user behavior is generated using the corrected complete trajectory. This report integrates details of dwell times and movements within the trajectory, determines whether deviations in the report have been eliminated, and yields the final trajectory reconstruction result.

[0011] Optionally, determining a continuous sequence of trajectory segments includes: Based on the predicted movement pattern path, extract the timestamps and location information of the path points to determine the path data; By analyzing path data and comparing the timestamps of adjacent path points, time gaps are identified, and a set of time gaps is obtained. For a set of time discontinuities, calculate the time difference of each discontinuity. If the time difference is less than a preset threshold, mark it as a segment to be filled and determine the list of segments to be filled. A linear interpolation method is used to fill the discontinuous segments in the list of segments to be filled, generate the filled path points, extract continuous trajectory segments, merge adjacent continuous path points, and obtain a continuous trajectory segment sequence.

[0012] The beneficial effects of this invention are as follows: This invention addresses the problem of analyzing user movement behavior in physical space. It collects coordinate sequences and uses clustering algorithms to generate a preliminary set of dwell points. By combining timestamps to calculate dwell time and access frequency, it filters high-frequency key dwell locations and integrates external geographic information databases to determine functional attributes such as workplaces, forming a set of location functional attributes. This invention analyzes the time transition sequences between locations using sequence modeling algorithms to predict movement pattern paths. It then interpolates and fills in path gaps to generate continuous trajectory segment sequences, integrates semantic tags to adjust deviation points, and finally generates a dynamic user behavior report, eliminating deviations and achieving complete trajectory reconstruction. By integrating dwell and movement details, this invention accurately characterizes user behavior patterns, significantly improving the accuracy and completeness of trajectory analysis, and is applicable to scenarios such as intelligent navigation and behavior prediction. Attached Figure Description

[0013] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0014] Figure 1 This is a schematic flowchart of the user behavior-based location analysis method according to an embodiment of the present invention. Detailed Implementation

[0015] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0016] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0017] like Figure 1 As shown, this embodiment proposes a method for analyzing user activity locations based on user behavior, including: Obtain data on user dwell times in the physical space; Based on the aforementioned stop point data, external geographic information data is correlated to obtain the functional attribute data of the marked locations; Temporal transition sequences between locations are extracted from the labeled location functional attribute data. Sequence modeling algorithms are used to process the transition sequences to obtain predicted movement pattern paths. The movement pattern paths reflect the location association patterns. The system analyzes and corrects predicted mobile patterns and paths, generating dynamic reports on user behavior.

[0018] Furthermore, obtaining data on user dwell times in physical spaces includes: Collect user location data in physical space, and use clustering algorithms to group the coordinate sequences to obtain a preliminary set of dwell points; Based on the timestamp information in the initial set of stops, the duration and frequency of each stop are calculated to determine the list of key stops, which includes locations with high frequency.

[0019] Furthermore, collecting user location data in physical space includes: The system acquires user location data from physical space using sensors or positioning devices, and generates a coordinate sequence that includes timestamps.

[0020] Specifically, in this embodiment, the coordinate sequence of the user in physical space is collected by location data. First, the GPS module records the user's position every second to generate sequence data containing timestamps, longitude and latitude. For example, the user generates 600 sets of coordinate data in 10 minutes, and each set of data is in the form of (time t, longitude 116.305, latitude 39.977).

[0021] Furthermore, the list of key stops includes: By parsing the initial set of stop points, timestamp data is obtained; If the timestamp data is complete, group by stop point, calculate the entry and exit time difference for each stop point, and obtain the duration; Based on the duration, a time window is used to divide the visit into segments, and the number of visits at each stop point is counted to determine the visit frequency; If the access frequency exceeds the preset threshold, the corresponding stop point will be marked as a high-frequency point; By using cluster analysis and the K-means algorithm to group high-frequency points, key location data is obtained. Based on key location data, combined with access frequency and duration, a weighted sorting algorithm is used to determine the location ranking; By sorting the locations, an ordered list of key stops is generated.

[0022] Specifically, in this embodiment, each stop point in the initial set of stops contains timestamp information. For example, user A's trajectory data records multiple points: point 1 (latitude and longitude, start time, end time), point 2 (latitude and longitude, start time, end time), and point 3 (latitude and longitude, start time, end time). This data is automatically collected by a GPS device and stored in a database. First, the duration of each stop point is processed using a time difference calculation algorithm. For example, for point 1, the Python datetime module is used to calculate the duration by subtracting the start time from the end time: duration 1 = (09:30:00 - 08:00:00) = 1.5 hours; duration 2 = (11:00:00 - 10:15:00) = 0.75 hours; duration 3 = (15:45:00 - 14:00:00) = 1.75 hours. This algorithm ensures accuracy to the second and converts the data to hourly floating-point numbers for easy subsequent aggregation and analysis. Secondly, based on the statistical access frequency of the entire dataset, the algorithm traverses all locations and clusters similar locations according to a latitude and longitude coordinate threshold of 0.01 degrees (approximately 1 kilometer). For example, locations 1 and 3 have the same coordinates, so frequency 1 = 2 times and frequency 2 = 1 time. This clustering uses the DBSCAN algorithm with parameters eps = 0.01 and min_samples = 1 for automatic grouping, avoiding manual intervention and handling of noise points. Next, a list of key stay locations is determined by threshold filtering. A composite condition is set: frequency threshold > 1.5 times and duration threshold > 1 hour. The analysis process calculates that location 1 has a total duration (1.5 + 1.75 = 3.25 hours), an average duration of 1.625 hours, and a frequency of 2 times, meeting the condition and being selected; location 2 has a total duration of 0.75 hours and a frequency of 1 time, not meeting the condition and being excluded; location 3 is merged with location 1, so the key list only includes location 1 (high-frequency residential area).

[0023] Furthermore, obtaining the location functional attribute data for the labeled locations includes: For each location of stay, location data is obtained from an external geographic information database, and geocoding technology is used to match latitude and longitude with location descriptions to generate a set of location geographic information. The dwell time is extracted from the location geographic information set. If the dwell time exceeds a preset threshold, it is marked as a potential workplace and a list of potential workplaces is generated. For the list of potential workplaces, analyze whether the dwell time falls within the working hours. If the dwell time matches the working hours, the location is confirmed as a workplace, and a workplace set is generated. Dwell frequency is extracted from the set of workplaces, and the number of dwell times per unit time is calculated using frequency statistics methods to obtain the dwell frequency distribution; Based on the frequency distribution of stay and combined with the location geographic information, the decision tree algorithm is used to classify the functional attributes of the locations and generate a set of location functional attributes. Extract locations labeled as workplaces from the set of location functional attributes, and generate the final labeled location functional attribute data.

[0024] Specifically, in this embodiment, when processing user trajectory data, the system first calls an external geographic information database, such as the Gaode Map API, to perform a related query on the list of key locations, such as latitude and longitude coordinates. The system inputs coordinate parameters to obtain the returned JSON format data, including the location name "a certain office building in a certain district of a certain city", the surrounding POI type "commercial office area", and detailed information about the building area of ​​approximately 5,000 square meters. The system extracts street and district attributes by parsing the "address_component" field in the Analyzing user request-API response, ensuring that the data accuracy reaches over 95%.

[0025] Next, the duration of each stay record is calculated. For example, if the time stamp sequence of the trajectory point is t1=08:00:00 and t2=18:00:00, the algorithm diff=t2-t1 is used to obtain a stay duration of 10 hours, which is compared with the preset threshold of 8 hours. If diff>8, it is marked as a long stay. At the same time, the frequency is calculated. If the location appears 15 times in the past 30 days, combined with the working hours defined as Monday to Friday 08:00-18:00, the frequency value is obtained by SQL query, which is 15 times / day. If the frequency>10 times / day, further analysis is performed.

[0026] Based on the above conditions, if the duration exceeds the threshold and the frequency meets the requirements within the working hours, the decision tree algorithm is used to determine it as a workplace attribute. The input feature vector [duration=10, frequency=15, time period matching degree=0.9] is fed into the pre-trained model. The model calculates the Gini impurity, and if it reaches below 0.2 after splitting from the root node, the category is confirmed as "workplace". Thus, the location functional attributes such as "office-financial services" are labeled. Finally, all locations that meet the conditions are aggregated to form a set {"Location 1: Workplace-financial services, duration 10h, frequency 15 times", "Location 2: Workplace-technology research and development, duration 9h, frequency 12 times"}. This set can be used for user behavior profile analysis and supports subsequent recommendation service optimization. The entire process is executed on a cloud server through automated scripts, ensuring a processing efficiency of <1 second per batch of data.

[0027] Furthermore, obtaining the predicted movement pattern path includes: A long short-term memory network algorithm is used to model the time transition sequence and generate a location transition relationship model; The path prediction result is obtained by predicting the location transfer probability in subsequent time steps using a location transfer relationship model. High-probability location shift sequences are extracted from the path prediction results to generate movement pattern paths; Based on the analysis of the correlation patterns between locations using mobile path analysis, the correlation patterns between locations are determined. If the confidence level of the location association pattern is lower than the preset threshold, the sequence modeling algorithm is iteratively optimized to regenerate the movement pattern path.

[0028] Specifically, in this embodiment, when processing the labeled set of location function attributes, the time transition sequence between locations is first extracted from the set using a spatiotemporal data mining algorithm. For example, given a user trajectory dataset containing 1000 records, each record is labeled with location functions such as residential area, office building, and shopping mall, as well as a timestamp. The DBSCAN clustering algorithm is used with parameters ε=0.5 km and minPts=5 to cluster the trajectory points, identifying continuous trajectory segments such as the sequence of moving from residential area A to office building B at 8:00 am, generating a transition sequence. This sequence captures the time dependency of the user's daily commuting and shopping patterns.

[0029] Subsequently, the extracted transfer sequences were input into a sequence modeling algorithm, which used a Long Short-Term Memory (LSTM) network model for processing. The model architecture included an input layer, two hidden layers with 128 neurons each, a dropout rate of 0.2 to prevent overfitting, and a softmax output layer. The training dataset consisted of 800 sequences and the validation set consisted of 200 sequences. The optimizer used Adam with a learning rate of 0.001. After training for 50 epochs, the loss function decreased from the initial 1.2 to 0.15. Analysis showed that the model achieved a sequence prediction accuracy of 85.7%. By adjusting the weights through backpropagation, it was found that the probability of transfer from residence to office during weekday peak hours (0.92) was higher than that during weekends (0.65).

[0030] Finally, based on the trained LSTM model, forward prediction is performed on the transfer sequence S. After inputting the current sequence (A, B, 8:00), the next transfer probability distribution P={(B, shopping mall C, 12:00):0.78,(B, gym D, 13:00):0.15} is output. The path with the highest probability is selected to generate the predicted movement pattern path M={A→B→C→A}. This path reflects location association patterns such as a cyclical pattern of residential-office-shopping mall. The association strength is calculated to be 0.81 using Jaccard similarity, indicating a regular behavioral chain formed by high-frequency repetitive transfers. Furthermore, the determination and correction of the predicted movement pattern path includes: Based on the predicted movement pattern path, analyze the discontinuities in the path. If the time difference of the discontinuities is less than a preset threshold, fill the discontinuities using an interpolation method to determine a continuous sequence of trajectory segments. For a continuous sequence of trajectory segments, semantic tags from location functional attributes are integrated to determine whether the sequence of trajectory segments conforms to behavioral intent logic. If it does not conform, deviation points in the sequence are adjusted to obtain a corrected complete trajectory. A dynamic report on user behavior is generated using the corrected complete trajectory. This report integrates details of dwell times and movements within the trajectory, determines whether deviations in the report have been eliminated, and yields the final trajectory reconstruction result.

[0031] Furthermore, determining a continuous sequence of trajectory segments includes: Based on the predicted movement pattern path, extract the timestamps and location information of the path points to determine the path data; By analyzing path data and comparing the timestamps of adjacent path points, time gaps are identified, and a set of time gaps is obtained. For a set of time discontinuities, calculate the time difference of each discontinuity. If the time difference is less than a preset threshold, mark it as a segment to be filled and determine the list of segments to be filled. A linear interpolation method is used to fill the discontinuous segments in the list of segments to be filled, generate the filled path points, extract the continuous trajectory segments, merge adjacent continuous path points, and obtain a continuous trajectory segment sequence.

[0032] Specifically, in this embodiment, the discontinuities in the predicted path are analyzed. By traversing the coordinate sequence, the Euclidean distance and timestamp difference between adjacent points are calculated. For example, it is found that the time difference between the segment from (101.5, 201.8) to (102.9, 203.4) is 45 seconds and the spatial distance is 2.1 km. The difference between the previous segment (100.1, 200.2) and (101.5, 201.8) is 15 seconds and 1.2 km. This 45-second difference is identified as exceeding the normal continuity threshold of 30 seconds and is thus marked as a discontinuity. For this discontinuity, if the time difference is less than a preset threshold such as 60 seconds, a linear interpolation method is used to fill it. For example, if the time t1 of the starting point (101.5, 201.8) of the discontinuity is 300 seconds and the time t2 of the ending point (102.9, 203.4) is 345 seconds, and the speed v = 2.1km / 45s = 0.0467km / s, then at t=315 seconds, the interpolation points x = 101.5 + (102.9 - 101.5) * (15 / 45) = 102.2 and y = 201.8 + (203.4 - 201.8) * (15 / 45) = 202.6 are formed, creating the filling point (102.2, 202.6). This is further repeated at t=330 seconds to obtain (102.55, 203.0), thus generating three new points to bridge the discontinuity. If the time difference exceeds 60 seconds, the discontinuity marker is retained to trigger an anomaly alarm. Finally, by splicing the original continuous segments, the filled segments, and the subsequent segments, the complete trajectory segment sequence [(100.1,200.2), (101.5,201.8), (102.2,202.6), (102.55,203.0), (102.9,203.4)] is determined.

[0033] The corrected complete trajectory includes: Using preset behavioral intent logic rules, sequence analysis is performed on the trajectory segment sequence. If the semantic label combination of the trajectory segment sequence does not conform to the behavioral intent logic, then the set of deviation points is determined. For the set of deviation points, obtain the timestamp and spatial location of each deviation point, combine the location functional attributes, calculate the semantic tag correction value that matches the behavioral intent logic, and obtain the correction tag set; By correcting the label set and replacing the semantic labels corresponding to the deviation points, an adjusted trajectory segment sequence is generated, resulting in a semantically corrected trajectory.

[0034] Specifically, in this embodiment, generating a dynamic user behavior report using the corrected complete trajectory includes: Extract stop point and movement path information from complete trajectory data to determine initial behavior patterns; If the initial behavior pattern contains outliers, then the stop points and movement paths are grouped using a clustering algorithm to obtain classified trajectory data; Based on the classified trajectory data, time series analysis methods are used to extract dynamic information and determine the trend of behavioral changes; If the trend of behavior change does not match the preset threshold, the trajectory data is corrected by interpolation to obtain the corrected trajectory data. Based on the corrected trajectory data, a decision tree algorithm is used to determine whether the deviation has been eliminated, and the deviation judgment result is obtained. By integrating the deviation assessment results with information on dwell points and movement paths, a dynamic analysis report is generated. Based on the dynamic analysis report, the final trajectory reconstruction results are extracted.

[0035] First, assume the input trajectory data is a sequence of GPS coordinates for the user throughout the day, including timestamps, latitude and longitude, and speed, such as {(t1, 116.305, 39.965, 5 m / s), (t2, 116.306, 39.966, 0 m / s)}. The DBSCAN algorithm (radius ε = 50 meters, minimum number of points MinPts = 5) is used to cluster stop points and calculate the dwell time at each point. For example, point t2 has a speed of 0, and all 10 surrounding points are within 50 meters, so it is clustered as a stop point with a dwell time of t3 - t2 = 300 seconds. Motion details are calculated using the Euclidean distance between trajectory points, with the formula d = √((x2 - x1)). 2 +(y2-y1) 2 For example, the distance from t1 to t2 is √((116.306-116.305) 2 +(39.966-39.965) 2 =141.4 meters, speed 5 m / s, time 28.28 seconds. The integrated report describes the user moving 141.4 meters after staying in a certain location for 300 seconds. Deviation is determined by comparing the trajectory points before and after correction. Assuming the original trajectory contains a noisy point (t2, 116.400, 39.900), this point is removed after correction, and the deviation distance is √((116.400 - 116.306)). 2 +(39.900-39.966) 2The distance is 10360 meters, indicating that noise has been eliminated. The final reconstructed trajectory is smoothed using a Kalman filter with parameters Q=0.01 (process noise) and R=0.1 (measurement noise), generating a smooth trajectory {(t1, 116.305, 39.965), (t2, 116.306, 39.966)}. The analysis process ensures that the stop point and movement distance are logically consistent, and after the deviation is eliminated, the trajectory points match the actual behavior. If the business needs to associate with mall user behavior, the stop point can be mapped to the mall POI, inferring that the user spent 300 seconds in a certain store, and the movement trajectory reflects their shopping path, enhancing the business insights of the report.

[0036] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made to the technical solutions of the present invention by those skilled in the art without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A method for analyzing user behavior-based activity location analysis, characterized in that, include: Obtain data on user dwell times in the physical space; Based on the aforementioned stop point data, external geographic information data is correlated to obtain the functional attribute data of the marked locations; Temporal transition sequences between locations are extracted from the labeled location functional attribute data. Sequence modeling algorithms are used to process the transition sequences to obtain predicted movement pattern paths. The movement pattern paths reflect the location association patterns. The system analyzes and corrects predicted mobile patterns and paths, generating dynamic reports on user behavior.

2. The method for analyzing user behavior-based activity locations according to claim 1, characterized in that, Obtaining user dwell time data in physical spaces includes: Collect user location data in physical space, and use clustering algorithms to group the coordinate sequences to obtain a preliminary set of dwell points; Based on the timestamp information in the initial set of stops, the duration and frequency of each stop are calculated to determine the list of key stops, which includes locations with high frequency.

3. The method for analyzing user behavior-based activity locations according to claim 2, characterized in that, The collection of user location data in physical space includes: The system acquires user location data from physical space using sensors or positioning devices, and generates a coordinate sequence that includes timestamps.

4. The method for analyzing user behavior-based activity locations according to claim 2, characterized in that, The list of key stops includes: By parsing the initial set of stop points, timestamp data is obtained; If the timestamp data is complete, group by stop point, calculate the entry and exit time difference for each stop point, and obtain the duration; Based on the duration, a time window is used to divide the visit into segments, and the number of visits at each stop point is counted to determine the visit frequency; If the access frequency exceeds the preset threshold, the corresponding stop point will be marked as a high-frequency point; By using cluster analysis and the K-means algorithm to group high-frequency points, key location data is obtained. Based on key location data, combined with access frequency and duration, a weighted sorting algorithm is used to determine the location ranking; By sorting the locations, an ordered list of key stops is generated.

5. The method for analyzing user behavior-based activity locations according to claim 1, characterized in that, The location attribute data to be retrieved includes: For each location of stay, location data is obtained from an external geographic information database, and geocoding technology is used to match latitude and longitude with location descriptions to generate a set of location geographic information. The dwell time is extracted from the location geographic information set. If the dwell time exceeds a preset threshold, it is marked as a potential workplace and a list of potential workplaces is generated. For the list of potential workplaces, analyze whether the dwell time falls within the working hours. If the dwell time matches the working hours, the location is confirmed as a workplace, and a workplace set is generated. Dwell frequency is extracted from the set of workplaces, and the number of dwell times per unit time is calculated using frequency statistics methods to obtain the dwell frequency distribution; Based on the frequency distribution of stay and combined with the location geographic information, the decision tree algorithm is used to classify the functional attributes of the locations and generate a set of location functional attributes. Extract locations labeled as workplaces from the set of location functional attributes, and generate the final labeled location functional attribute data.

6. The method for analyzing user behavior-based dwell time locations according to claim 1, characterized in that, The predicted movement pattern path includes: A long short-term memory network algorithm is used to model the time transition sequence and generate a location transition relationship model; The path prediction result is obtained by predicting the location transfer probability in subsequent time steps using a location transfer relationship model. High-probability location shift sequences are extracted from the path prediction results to generate movement pattern paths; Based on the analysis of the correlation patterns between locations using mobile path analysis, the correlation patterns between locations are determined. If the confidence level of the location association pattern is lower than the preset threshold, the sequence modeling algorithm is iteratively optimized to regenerate the movement pattern path.

7. The method for analyzing user behavior-based activity locations according to claim 1, characterized in that, Determining and correcting predicted movement patterns includes: Based on the predicted movement pattern path, analyze the discontinuities in the path. If the time difference of the discontinuities is less than a preset threshold, fill the discontinuities using an interpolation method to determine a continuous sequence of trajectory segments. For a continuous sequence of trajectory segments, semantic tags from location functional attributes are integrated to determine whether the sequence of trajectory segments conforms to behavioral intent logic. If it does not conform, deviation points in the sequence are adjusted to obtain a corrected complete trajectory. A dynamic report on user behavior is generated using the corrected complete trajectory. This report integrates details of dwell times and movements within the trajectory, determines whether deviations in the report have been eliminated, and yields the final trajectory reconstruction result.

8. The method for analyzing user behavior-based activity locations according to claim 7, characterized in that, Determining a continuous sequence of trajectory segments includes: Based on the predicted movement pattern path, extract the timestamps and location information of the path points to determine the path data; By analyzing path data and comparing the timestamps of adjacent path points, time gaps are identified, and a set of time gaps is obtained. For a set of time discontinuities, calculate the time difference of each discontinuity. If the time difference is less than a preset threshold, mark it as a segment to be filled and determine the list of segments to be filled. A linear interpolation method is used to fill the discontinuous segments in the list of segments to be filled, generate the filled path points, extract the continuous trajectory segments, merge adjacent continuous path points, and obtain a continuous trajectory segment sequence.

Images