A method and system for attendance management based on multi-dimensional data verification

Abstract

This invention provides a method and system for attendance management based on multi-dimensional data verification, relating to the field of attendance data processing technology. The method includes: acquiring multi-source data of employees within the attendance period; cleaning and parsing the data to generate a set of attendance events to be verified, each with a timestamp and source identifier; determining the basic attendance type based on attendance rules and scheduling information; calling a multi-dimensional verification model to perform weighted fusion analysis on mobile terminal positioning, access control direction, network connection characteristics, and frequency of operational behaviors at the same timestamp, calculating an on-duty confidence score, and marking any score below a threshold as a questionable attendance event; triggering the verification process by pushing a verification request to the employee's terminal, receiving feedback data including real-time positioning and on-site images, correcting questionable events, and generating a calibrated attendance record. This invention solves the problems of existing attendance management data having a single dimension, difficulty in accurately verifying on-duty status, and a lack of real-time and accurate verification.

Description

Technical Field

[0001] This invention relates to the field of attendance data processing technology, and more specifically, to an attendance management method and system based on multi-dimensional data verification. Background Technology

[0002] As companies increasingly demand more refined human resource management, attendance management, as a fundamental aspect of payroll calculation and performance evaluation, directly impacts management efficiency and employee satisfaction. Currently, some companies are combining mobile check-in with access control systems for attendance management. For example, patent application CN119887135A proposes an attendance data processing method and apparatus. This method acquires employee mobile check-in data and access control data, generates a preliminary attendance record based on their matching, and then uses travel data and leave / return data for secondary verification to adjust abnormal attendance statuses. This method achieves cross-validation of multi-source data to a certain extent, reducing manual verification workload and improving the accuracy of attendance data.

[0003] However, existing technologies still have the following shortcomings: First, the data collection dimensions for attendance are still relatively singular, mainly relying on clock-in time and access records, failing to effectively verify the actual on-duty status of employees after clocking in, leaving room for technical circumvention of behaviors such as "clocking in and leaving the post" and "proxy clocking in"; Second, the attendance judgment logic lacks weighted fusion analysis of multiple dimensions such as location, device, and behavior, making it difficult to achieve accurate judgment in complex scenarios (such as field work, working from home, and flexible work arrangements); Third, the identification of abnormal attendance is still mainly based on post-event processing, lacking real-time monitoring and hierarchical early warning mechanisms, resulting in delayed response to anomalies and affecting management efficiency. Therefore, existing attendance systems still need further optimization in terms of authenticity verification, multi-scenario adaptation, and real-time early warning. Summary of the Invention

[0004] The purpose of this invention is to provide an attendance management method and system based on multi-dimensional data verification, which aims to solve the problems of single data collection dimensions, difficulty in effectively verifying the actual attendance status of employees, and lack of real-time and accurate verification in existing attendance management methods.

[0005] This invention is achieved through the following technical solution:

[0006] An attendance management method based on multi-dimensional data verification includes the following steps:

[0007] Acquire multi-source data of employees within a preset attendance period, and clean and parse the multi-source data to generate a set of attendance events to be verified, including timestamps and data source identifiers; wherein, the multi-source data includes mobile terminal clock-in data, access control data, office area network connection data, and terminal operation behavior data;

[0008] Based on the preset attendance rules, the set of attendance events to be verified is matched with the pre-stored shift information of employees to determine the basic attendance type corresponding to each attendance event to be verified. The basic attendance types include normal attendance, lateness, early departure, field work and absence.

[0009] The preset multi-dimensional verification model is invoked to perform multi-dimensional weighted fusion analysis on mobile terminal location data, access control direction data, office area network connection characteristic data, and terminal operation behavior frequency data in attendance events to be verified belonging to the same timestamp. The employee on-duty confidence score corresponding to the target timestamp is calculated. If the employee on-duty confidence score is lower than the preset threshold, the attendance event to be verified corresponding to the target timestamp is marked as a questionable attendance event.

[0010] In response to the marking of questionable attendance events, a verification process is triggered, pushing an on-duty status verification request to the employee's terminal, and receiving verification feedback data from the employee's terminal that includes real-time location information and on-site images. Based on the verification feedback data, the questionable attendance events are corrected, and calibrated attendance records are generated.

[0011] Optionally, the specific process of obtaining multi-source data of employees within a preset attendance period is as follows:

[0012] By calling the positioning module through the mobile terminal application and combining base station and wireless network signals, the triangulation algorithm is used to obtain the location coordinates of employees when they clock in in real time, and generate mobile terminal clock-in data containing timestamps and latitude and longitude.

[0013] The access control system obtains employee entry and exit records of the office area in real time through the application programming interface, which are used as access control data. The access records include timestamps, access control point identifiers and passage directions.

[0014] The system collects the service set identifier, connection time, disconnection time, and media access control address of employees connecting to the office area's wireless network in real time through the operating system logs of mobile terminals or office computers, and uses this data as office area network connection data.

[0015] The system collects real-time data on employee actions on office terminals through the log interface in the office system backend, generating timestamped action logs. The action data includes login time, logout time, screen activity status, application startup records, and operation frequency.

[0016] Optionally, the specific process of cleaning and parsing multi-source data to generate a set of attendance events to be verified, including timestamps and data source identifiers, is as follows:

[0017] All acquired data is timestamped using the Network Time Protocol.

[0018] For mobile terminal check-in data, the Kalman filter algorithm is used to filter out drift noise in the positioning coordinates, and the cleaned coordinate data is matched with the preset device whitelist to remove check-in data generated by unbound devices;

[0019] For access control data, check the validity of the passage direction field and remove duplicate or logically incorrect passage records;

[0020] For network connection data in the office area, extract the service set identifier and media access control address of the connection, compare them with the pre-stored company network whitelist, and retain the data that successfully matches.

[0021] For terminal operation behavior data, a log parsing algorithm is used to transform unstructured operation logs into structured event sequences, and to extract login time, logout time, screen active periods, and operation frequency information.

[0022] The cleaned data is packaged according to a unified event format, and a calibrated timestamp and a data source identifier are added to each data item. The data source identifier is used to distinguish the type of data.

[0023] Arrange all encapsulated events in ascending order of timestamp to generate a set of attendance events to be verified.

[0024] Optionally, after cleaning and parsing the multi-source data to generate a set of attendance events to be verified, the process further includes the steps of constructing an electronic fence and performing dual location verification.

[0025] Preset the latitude and longitude boundaries of the office area or field location, and generate polygonal or circular electronic fences;

[0026] Obtain the location coordinates of the mobile terminal in the attendance event to be verified, and use a distance calculation algorithm to calculate the shortest distance between the location coordinates of the mobile terminal and the boundary of the electronic fence.

[0027] If the shortest distance is less than or equal to the preset threshold, the location is determined to be valid; otherwise, the wireless network service set identifier connected at the time of check-in is obtained. If the corresponding wireless network service set identifier matches the pre-stored company intranet service set identifier, the location is determined to be valid.

[0028] The valid location marker is stored as a feature dimension in the attendance event to be verified, and used for the subsequent calculation of the on-duty confidence score.

[0029] Optionally, the specific construction process of the multi-dimensional verification model is as follows:

[0030] Obtain multi-source data of employees within historical attendance periods as a training sample set;

[0031] Feature extraction is performed on various types of data in the training sample set to generate time matching features, location matching features, device trust features, and behavioral activity features. The time matching features reflect the degree of deviation between the clock-in time and the scheduled shift time; the location matching features reflect the spatial relationship between the clock-in coordinates and the electronic fence of the office area; the device trust features reflect whether the clock-in device belongs to a trusted device bound to the employee; and the behavioral activity features reflect the continuity and frequency of employee operations on the office terminal.

[0032] Based on the extracted features, the mapping relationship between each feature and the actual on-duty status is fitted by the logistic regression algorithm to generate the weight coefficients corresponding to each feature. The weight coefficients are used to characterize the importance of different features in the determination of attendance authenticity.

[0033] Based on the weight coefficients obtained after training, a weighted scoring function is constructed to score the on-duty confidence of real-time collected attendance events to be verified.

[0034] Based on the false positive rate and false negative rate in historical attendance data, a dynamically adjusted confidence threshold is set. The confidence threshold is used to determine whether the attendance event to be verified belongs to the questionable attendance event.

[0035] Optionally, the specific process of responding to the flag of a questionable attendance event, triggering a verification process, pushing an on-duty status verification request to the employee's terminal, receiving verification feedback data containing real-time location information and on-site images from the employee's terminal, and correcting the questionable attendance event based on the verification feedback data to generate a calibrated attendance record is as follows:

[0036] The timestamp corresponding to the questionable attendance event is obtained as the target verification time point. A request message containing verification instructions is sent to the employee terminal through the message push service. The request message requires the employee to start the terminal camera to take a live image with the real-time background within a preset response time, and simultaneously call the positioning module of the employee terminal to obtain the current real-time location coordinates.

[0037] Receive verification feedback data packets uploaded by employee terminals. The verification feedback data packets include at least the on-site image with a time watermark, real-time location coordinates, and the terminal device identifier at the time of shooting.

[0038] The time watermark information in the on-site image is extracted by optical character recognition technology and compared with the target verification time point to determine whether the shooting time is within the preset response time. If it exceeds the time limit, the verification time is determined to be overdue and the original doubtful mark is maintained.

[0039] If the shooting time is within the preset response time, the background features of the on-site image are analyzed by the image recognition algorithm, the features of office environment markers or field locations are extracted, and they are matched with the pre-stored office area image feature library or the field location image features submitted by employees to calculate the image environment matching degree.

[0040] The received real-time location coordinates are compared with the mobile terminal location data at the time of the suspected attendance incident, the location offset is analyzed, and the confidence level of the verification result is generated by combining the image environment matching degree.

[0041] If the confidence level of the verification result exceeds the preset correction threshold, the questionable attendance event will be corrected to the verification passed status, and the verification feedback data and correction results will be stored in the attendance record; otherwise, the questionable attendance event will be marked as pending manual review and pushed to the management personnel terminal.

[0042] Optionally, the specific process of comparing the received real-time location coordinates with the mobile terminal location data at the time of the questionable attendance event, analyzing the location offset, and combining it with the image environment matching degree to comprehensively generate the confidence level of the verification result is as follows:

[0043] Extract the initial location coordinates recorded in the questionable attendance events, combine them with the verification location coordinates uploaded in real time by the employee terminal, calculate the spatial straight-line distance between the two, and obtain the location offset value.

[0044] The position offset value is input into the preset offset scoring function. When the offset is less than the first threshold, a high confidence weight is added. When the offset is between the first threshold and the second threshold, a medium confidence weight is added. When the offset exceeds the second threshold, a low confidence weight is added.

[0045] The on-site images are feature extracted by a preset image recognition model. The extracted image features are then compared with the pre-stored office environment image feature library or the standard image features of field locations to generate an image environment matching score.

[0046] If the image environment matching score is higher than the preset image confidence threshold, the dynamic weighted fusion logic is triggered, and the confidence weight corresponding to the position offset is added to the image environment matching score according to the preset fusion coefficient to generate the initial value of the verification result confidence.

[0047] If the image environment matching score is lower than or equal to the preset image credibility threshold, the on-site image is determined to be invalid, the verification result confidence is set to zero, and the verification feedback data is marked as unreliable data.

[0048] The initial confidence level of the verification result is compared with the preset pass threshold. If it is higher than the pass threshold, the verification is deemed to have passed. If it is lower than the pass threshold but higher than the doubt threshold, a second dynamic verification instruction is triggered. If it is lower than the doubt threshold, it is marked as pending manual review.

[0049] Based on the same inventive concept, this invention also provides an attendance management system based on multi-dimensional data verification, used to implement the aforementioned attendance management method based on multi-dimensional data verification, comprising:

[0050] The multi-source data acquisition module is used to acquire multi-source data of employees within a preset attendance period;

[0051] The data preprocessing module, connected to the multi-source data acquisition module, is used to perform timestamp calibration, noise filtering, and legality checks on the multi-source data, and generate a set of attendance events to be verified that includes timestamps and data source identifiers.

[0052] The attendance rule matching module, connected to the data preprocessing module, is used to match the set of attendance events to be verified with the pre-stored shift information of employees to determine the basic attendance type corresponding to each attendance event to be verified.

[0053] The multi-dimensional verification engine module is connected to the data preprocessing module and the attendance rule matching module respectively. It is used to perform multi-dimensional weighted fusion analysis on multiple types of data in attendance events to be verified belonging to the same timestamp, calculate the employee on-duty confidence score corresponding to the target timestamp, and mark the attendance event to be verified corresponding to the target timestamp as a questionable attendance event when the on-duty confidence score is lower than a preset threshold.

[0054] The dynamic interactive verification module, connected to the multi-dimensional verification engine module, is used to push an on-duty status verification request to the employee terminal in response to the marking of a questionable attendance event, and to receive verification feedback data returned by the employee terminal containing real-time location information and on-site images. Based on the verification feedback data, the questionable attendance event is corrected and a calibrated attendance record is generated.

[0055] The attendance record generation module is connected to the dynamic interactive verification module. It is used to store the verified and corrected attendance events, form the final attendance record, and synchronize the attendance record to the payroll system.

[0056] The technical solution of the present invention has at least the following advantages and beneficial effects:

[0057] It not only collects traditional clock-in and access control data, but also integrates multi-source data such as network connection characteristics of the office area and frequency of terminal operation behavior. It can effectively verify the actual on-duty status of employees after clocking in, reduce the technical circumvention space for behaviors such as "clocking in and leaving the post" and "clocking in on behalf of others", and enhance the anti-cheating ability of attendance management.

[0058] By pre-setting a multi-dimensional verification model, a weighted fusion analysis is performed on mobile terminal positioning, access control direction, network connection characteristics, and frequency of operation behavior to generate an on-duty confidence score. This model can adapt to complex scenarios such as field work, working from home, and flexible work arrangements, avoiding misjudgments caused by a single data source and improving the accuracy and robustness of attendance determination.

[0059] By calculating the on-duty confidence score in real time and setting preset thresholds to mark suspicious events, the verification process is immediately triggered once an anomaly is detected, and a verification request is pushed to the employee, which effectively shortens the lag time of anomaly response and improves management efficiency.

[0060] By automating data collection, integrating multi-source data, and conducting real-time verification, the workload of manual verification has been significantly reduced. At the same time, the verification method triggered by questionable events ensures the seriousness of the attendance system while avoiding excessive interference with employees who are on duty, thereby improving employee satisfaction and the level of management refinement. Attached Figure Description

[0061] Figure 1 This is a flowchart illustrating the attendance management method based on multi-dimensional data verification according to an embodiment of the present invention.

[0062] Figure 2 This is a schematic diagram of the attendance management system based on multi-dimensional data verification according to an embodiment of the present invention. Detailed Implementation

[0063] The following is a detailed description of the embodiments, in conjunction with the accompanying drawings.

[0064] Reference Figure 1 An attendance management method based on multi-dimensional data verification includes the following steps:

[0065] Step 1: Obtain multi-source data of employees within the preset attendance period, and clean and parse the multi-source data to generate a set of attendance events to be verified, including timestamps and data source identifiers; wherein, the multi-source data includes mobile terminal check-in data, access control data, office area network connection data, and terminal operation behavior data.

[0066] In some embodiments, the specific process of obtaining multi-source data of employees within a preset attendance period is as follows:

[0067] By calling the positioning module through the mobile terminal application and combining base station and wireless network signals, the triangulation algorithm is used to obtain the location coordinates of employees when they clock in in real time, and generate mobile terminal clock-in data containing timestamps and latitude and longitude.

[0068] The access control system obtains employee entry and exit records in real time through its application programming interface (API), which serve as access control data. The access records include timestamps, access point identifiers, and directions of passage.

[0069] The system collects the service set identifier, connection time, disconnection time, and media access control address of employees connecting to the office area's wireless network in real time through the operating system logs of mobile terminals or office computers, and uses this data as office area network connection data.

[0070] The system collects real-time data on employee actions on office terminals through the log interface in the office system backend, generating timestamped action logs. The action data includes login time, logout time, screen activity status, application startup records, and operation frequency.

[0071] In some embodiments, the specific process of cleaning and parsing multi-source data to generate a set of attendance events to be verified, including timestamps and data source identifiers, is as follows:

[0072] All acquired data is timestamped using the Network Time Protocol.

[0073] For mobile terminal check-in data, the Kalman filter algorithm is used to filter out drift noise in the positioning coordinates, and the cleaned coordinate data is matched with the preset device whitelist to remove check-in data generated by unbound devices;

[0074] For access control data, check the validity of the passage direction field and remove duplicate or logically incorrect passage records;

[0075] For network connection data in the office area, extract the service set identifier and media access control address of the connection, compare them with the pre-stored company network whitelist, and retain the data that successfully matches.

[0076] For terminal operation behavior data, a log parsing algorithm is used to transform unstructured operation logs into structured event sequences, and to extract login time, logout time, screen active periods, and operation frequency information.

[0077] The cleaned data is packaged according to a unified event format, and a calibrated timestamp and a data source identifier are added to each data item. The data source identifier is used to distinguish the type of data.

[0078] Arrange all encapsulated events in ascending order of timestamp to generate a set of attendance events to be verified.

[0079] In some embodiments, after cleaning and parsing the multi-source data to generate a set of attendance events to be verified, the method further includes the steps of constructing an electronic fence and performing dual location verification.

[0080] Preset the latitude and longitude boundaries of the office area or field location, and generate polygonal or circular electronic fences;

[0081] Obtain the location coordinates of the mobile terminal in the attendance event to be verified, and use a distance calculation algorithm to calculate the shortest distance between the location coordinates of the mobile terminal and the boundary of the electronic fence.

[0082] If the shortest distance is less than or equal to the preset threshold, the location is determined to be valid; otherwise, the wireless network service set identifier connected at the time of check-in is obtained. If the corresponding wireless network service set identifier matches the pre-stored company intranet service set identifier, the location is determined to be valid.

[0083] The valid location marker is stored as a feature dimension in the attendance event to be verified, and used for the subsequent calculation of the on-duty confidence score.

[0084] Step 2: Based on the preset attendance rules, match the set of attendance events to be verified with the pre-stored shift information of employees to determine the basic attendance type corresponding to each attendance event to be verified. The basic attendance types include normal attendance, lateness, early departure, field work, and absence.

[0085] In some embodiments, a visual rule configuration tool can be provided through the backend management interface, allowing administrators to set differentiated attendance rules for different departments, positions, or individual employees, including:

[0086] Shift definitions include fixed shifts (09:00-18:00), flexible shifts (core working hours 10:00-16:00, flexible start and end times ±1 hour), and scheduling (different shifts per day within a cycle).

[0087] Lateness or early departure thresholds, such as the number of allowable grace minutes (e.g., no lateness within 5 minutes), are accumulated by the minute after exceeding the threshold;

[0088] Absence determination rules: If there is no clock-in record and no valid approval, it will be marked as absence.

[0089] Rules for field work or working from home, related approval form types, allow location verification to be ignored within a specific time period.

[0090] The configured rules are transformed into executable decision logic (such as Drools and Easy Rules) by the rule engine and stored in the rule table of the MySQL database. At the same time, they are cached in Redis (Remote Dictionary Server) to improve access efficiency.

[0091] Employee work schedules, including daily arrival and departure times and rest days, are synchronized in real-time from the Human Resources System (HRIS) or scheduling module via API and stored as a structured schedule. Approved leave, business trip, fieldwork, and overtime applications are retrieved in real-time via the workflow engine interface. Key fields such as employee ID, approval type, start time, end time, and approval status are extracted and converted into time interval rules (e.g., "fieldwork from [year] [time] to [year] [time]"), which are then cached in Redis for fast retrieval.

[0092] After obtaining employee scheduling or shift information, the system iterates through the set of attendance events to be verified, extracts events strongly correlated with clock-in time (mainly entry and exit records from mobile terminal clock-in data and access control data), and compares them with the prescribed clock-in and clock-out times. After determining the basic attendance type, the system uses some feature dimensions (such as device reliability) generated in step one for preliminary verification to reduce false positives.

[0093] Step 3: Call the preset multi-dimensional verification model to perform multi-dimensional weighted fusion analysis on the mobile terminal location data, access control direction data, office area network connection characteristic data, and terminal operation behavior frequency data of the attendance events to be verified belonging to the same timestamp, and calculate the employee on-duty confidence score corresponding to the target timestamp. If the employee on-duty confidence score is lower than the preset threshold, the attendance event to be verified corresponding to the target timestamp is marked as a questionable attendance event.

[0094] In some embodiments, the specific construction process of the multi-dimensional verification model is as follows:

[0095] Multi-source data of employees within historical attendance periods is acquired as a training sample set. Features are extracted from various types of data in the training sample set to generate time-matching features, location-matching features, device credibility features, and behavioral activity features. The time-matching feature reflects the deviation between clock-in time and shift schedule; the location-matching feature reflects the spatial relationship between clock-in coordinates and the electronic fence of the office area; the device credibility feature reflects whether the clock-in device belongs to a trusted device bound to the employee; and the behavioral activity feature reflects the continuity and frequency of employee operations on office terminals. Based on the extracted features, a logistic regression algorithm is used to fit the mapping relationship between each feature and the actual on-duty status, generating weight coefficients for each feature. These weight coefficients characterize the importance of different features in determining the authenticity of attendance. Based on the weight coefficients obtained after training, a weighted scoring function is constructed to score the on-duty confidence of real-time collected attendance events to be verified. A dynamically adjusted confidence threshold is set based on the false positive and false negative rates in historical attendance data. This confidence threshold is used to determine whether an attendance event to be verified is a questionable attendance event.

[0096] It is possible to set a total of [number] attendance periods within the historical attendance cycle. The nth sample, the th The feature vector of each sample is ,in, For time-matching features; For location matching features; For trusted device characteristics; This represents a behavioral activity characteristic. The corresponding actual on-duty status label is: ,in, This indicates that the individual is indeed currently employed. This indicates that the individual is not at their post. The mapping relationship between each feature and the actual post status is fitted using a logistic regression algorithm, as shown in the following formula:

[0097]

[0098] in, Indicates that in a given number Feature vector of each sample Under the condition that the employee is actually on duty (i.e.) The probability of ). This is the weight vector; This is the transpose of the weight vector; Weights for time features; The weights for location features; Weights for the trusted features of the device; The weights for behavioral activity features.

[0099] Weight vector It can be estimated by maximizing the log-likelihood function, as shown in the following equation:

[0100]

[0101] Based on the weight coefficients obtained during training, a weighted scoring function is constructed as shown in the following formula:

[0102]

[0103] It can also be normalized to a percentage score according to actual needs, as shown in the following formula:

[0104]

[0105] in, This is a weighted scoring function used to measure an employee's confidence level in being on duty at a given point in time. The normalized percentage score maps the original score to the range of 0 to 100 for easier intuitive understanding; This represents the maximum weighted score across all samples (or all possible events).

[0106] The false positive rate in historical attendance data can be set. The false negative rate is Set dynamically adjustable confidence thresholds based on business needs. Used to determine whether there is any doubt. If it is a questionable attendance event, it will be marked as such; among them, The adjustment can be dynamically made based on historical ROC curves (recipient operating characteristic curves) or business tolerance, as shown in the following formula:

[0107]

[0108] in, The optimal threshold; This is a weighting factor used to balance the importance of false positive rate and false negative rate; Indicates at a given threshold The false positive rate refers to the proportion of employees who are not actually on duty but are incorrectly classified as normal (not marked as questionable) because their confidence score is higher than or equal to the threshold. Indicates at a given threshold The false negative rate refers to the proportion of employees who are actually on duty but are incorrectly marked as suspicious because their confidence scores are below the threshold.

[0109] Step 4: In response to the flagging of questionable attendance events, trigger the verification process, push the on-duty status verification request to the employee terminal, and receive the verification feedback data returned by the employee terminal, which includes real-time location information and on-site images. Based on the verification feedback data, correct the questionable attendance events and generate calibrated attendance records.

[0110] In some embodiments, the specific process of triggering a verification process in response to the flagging of a questionable attendance event, pushing an on-duty status verification request to the employee terminal, receiving verification feedback data containing real-time location information and on-site images returned by the employee terminal, correcting the questionable attendance event based on the verification feedback data, and generating a calibrated attendance record is as follows:

[0111] The timestamp corresponding to the questionable attendance event is obtained as the target verification time point. A request message containing verification instructions is sent to the employee terminal through the message push service. The request message requires the employee to start the terminal camera to take a live image with the real-time background within a preset response time, and simultaneously call the positioning module of the employee terminal to obtain the current real-time location coordinates.

[0112] Receive verification feedback data packets uploaded by employee terminals. The verification feedback data packets include at least the on-site image with a time watermark, real-time location coordinates, and the terminal device identifier at the time of shooting.

[0113] The time watermark information in the on-site image is extracted by optical character recognition technology and compared with the target verification time point to determine whether the shooting time is within the preset response time. If it exceeds the time limit, the verification time is determined to be overdue and the original doubtful mark is maintained.

[0114] If the shooting time is within the preset response time, the background features of the on-site image are analyzed by the image recognition algorithm, the features of office environment markers or field locations are extracted, and they are matched with the pre-stored office area image feature library or the field location image features submitted by employees to calculate the image environment matching degree.

[0115] The received real-time location coordinates are compared with the mobile terminal location data at the time of the suspected attendance incident, the location offset is analyzed, and the confidence level of the verification result is generated by combining the image environment matching degree.

[0116] If the confidence level of the verification result exceeds the preset correction threshold, the questionable attendance event will be corrected to the verification passed status, and the verification feedback data and correction results will be stored in the attendance record; otherwise, the questionable attendance event will be marked as pending manual review and pushed to the management personnel terminal.

[0117] In some embodiments, the specific process of comparing the received real-time location coordinates with the mobile terminal location data at the time of the questionable attendance event, analyzing the location offset, and combining the image environment matching degree to comprehensively generate the confidence level of the verification result is as follows:

[0118] Extract the initial location coordinates recorded in questionable attendance events, combine them with the verification location coordinates uploaded in real time by the employee's terminal, calculate the spatial straight-line distance between the two, and obtain the location offset value; input the location offset value into a preset offset scoring function, assign a high confidence weight when the offset is less than a first threshold, a medium confidence weight when the offset is between the first and second thresholds, and a low confidence weight when the offset exceeds the second threshold; extract features from the on-site image using a preset image recognition model, compare the extracted image features with a pre-stored office environment image feature library or standard image features of field locations for similarity comparison, and generate an image environment matching score; if If the image environment matching score is higher than the preset image credibility threshold, the dynamic weighted fusion logic is triggered. The confidence weight corresponding to the position offset is added to the image environment matching score and weighted and summed according to the preset fusion coefficient to generate the initial value of the verification result confidence. If the image environment matching score is lower than or equal to the preset image credibility threshold, the on-site image is determined to be invalid, the verification result confidence is set to zero, and the verification feedback data is marked as unreliable data. The initial value of the verification result confidence is compared with the preset pass threshold. If it is higher than the pass threshold, the verification is determined to be passed. If it is lower than the pass threshold but higher than the doubt threshold, a second dynamic verification instruction is triggered. If it is lower than the doubt threshold, it is marked as pending manual review.

[0119] The initial location coordinates in a questionable attendance event can be set as follows: The verification location coordinates uploaded in real time by the employee terminal are The spatial straight-line distance is shown in the following formula:

[0120]

[0121] in, This represents Euclidean distance (latitude and longitude coordinates can be projected onto a plane for calculation).

[0122] Preset offset first threshold With the second threshold ( ), and the corresponding confidence weighting value. , and ( Then the offset scoring function The definition is as follows:

[0123]

[0124] Let the image environment matching score be 1. (Output from image recognition model, normalized to) The image confidence threshold is . and These represent the weights of location and image, respectively. ,and The initial confidence level of the verification result is then... As shown in the following formula:

[0125]

[0126] when At the same time, the verification feedback data is marked as unreliable.

[0127] The preset threshold is The threshold for suspicion is ( The verification result will then be marked. As shown in the following formula:

[0128]

[0129] in, This indicates that the verification has passed. This indicates that a second dynamic verification has been triggered. This indicates that the document is marked as awaiting manual review.

[0130] Based on the same inventive concept, and corresponding to any of the above embodiments, refer to... Figure 2 This invention provides an attendance management system based on multi-dimensional data verification, used to implement the aforementioned attendance management method based on multi-dimensional data verification, including:

[0131] The multi-source data acquisition module is used to acquire multi-source data of employees within a preset attendance period;

[0132] The data preprocessing module, connected to the multi-source data acquisition module, is used to perform timestamp calibration, noise filtering, and legality checks on the multi-source data, and generate a set of attendance events to be verified that includes timestamps and data source identifiers.

[0133] The attendance rule matching module, connected to the data preprocessing module, is used to match the set of attendance events to be verified with the pre-stored shift information of employees to determine the basic attendance type corresponding to each attendance event to be verified.

[0134] The multi-dimensional verification engine module is connected to the data preprocessing module and the attendance rule matching module respectively. It is used to perform multi-dimensional weighted fusion analysis on multiple types of data in attendance events to be verified belonging to the same timestamp, calculate the employee on-duty confidence score corresponding to the target timestamp, and mark the attendance event to be verified corresponding to the target timestamp as a questionable attendance event when the on-duty confidence score is lower than a preset threshold.

[0135] The dynamic interactive verification module, connected to the multi-dimensional verification engine module, is used to push an on-duty status verification request to the employee terminal in response to the marking of a questionable attendance event, and to receive verification feedback data returned by the employee terminal containing real-time location information and on-site images. Based on the verification feedback data, the questionable attendance event is corrected and a calibrated attendance record is generated.

[0136] The attendance record generation module is connected to the dynamic interactive verification module. It is used to store the verified and corrected attendance events, form the final attendance record, and synchronize the attendance record to the payroll system.

Claims

1. An attendance management method based on multi-dimensional data verification, characterized in that, Includes the following steps: Acquire multi-source data of employees within a preset attendance period, and clean and parse the multi-source data to generate a set of attendance events to be verified, including timestamps and data source identifiers; wherein, the multi-source data includes mobile terminal clock-in data, access control data, office area network connection data, and terminal operation behavior data; Based on the preset attendance rules, the set of attendance events to be verified is matched with the pre-stored shift information of employees to determine the basic attendance type corresponding to each attendance event to be verified. The basic attendance types include normal attendance, lateness, early departure, field work and absence. The preset multi-dimensional verification model is invoked to perform multi-dimensional weighted fusion analysis on mobile terminal location data, access control direction data, office area network connection characteristic data, and terminal operation behavior frequency data in attendance events to be verified belonging to the same timestamp. The employee on-duty confidence score corresponding to the target timestamp is calculated. If the employee on-duty confidence score is lower than the preset threshold, the attendance event to be verified corresponding to the target timestamp is marked as a questionable attendance event. In response to the marking of questionable attendance events, a verification process is triggered, pushing an on-duty status verification request to the employee's terminal, and receiving verification feedback data from the employee's terminal that includes real-time location information and on-site images. Based on the verification feedback data, the questionable attendance events are corrected, and calibrated attendance records are generated.

2. The attendance management method based on multi-dimensional data verification as described in claim 1, characterized in that... The specific process for obtaining multi-source data of employees within a preset attendance period is as follows: By calling the positioning module through the mobile terminal application and combining base station and wireless network signals, the triangulation algorithm is used to obtain the location coordinates of employees when they clock in in real time, and generate mobile terminal clock-in data containing timestamps and latitude and longitude. The access control system obtains employee entry and exit records of the office area in real time through the application programming interface, which are used as access control data. The access records include timestamps, access control point identifiers and passage directions. The system collects the service set identifier, connection time, disconnection time, and media access control address of employees connecting to the office area's wireless network in real time through the operating system logs of mobile terminals or office computers, and uses this data as office area network connection data. The system collects real-time data on employee actions on office terminals through the log interface in the office system backend, generating timestamped action logs. The action data includes login time, logout time, screen activity status, application startup records, and operation frequency.

3. The attendance management method based on multi-dimensional data verification as described in claim 1, characterized in that, The specific process of cleaning and parsing multi-source data to generate a set of attendance events to be verified, including timestamps and data source identifiers, is as follows: All acquired data is timestamped using the Network Time Protocol. For mobile terminal check-in data, the Kalman filter algorithm is used to filter out drift noise in the positioning coordinates, and the cleaned coordinate data is matched with the preset device whitelist to remove check-in data generated by unbound devices; For access control data, check the validity of the passage direction field and remove duplicate or logically incorrect passage records; For network connection data in the office area, extract the service set identifier and media access control address of the connection, compare them with the pre-stored company network whitelist, and retain the data that successfully matches. For terminal operation behavior data, a log parsing algorithm is used to transform unstructured operation logs into structured event sequences, and to extract login time, logout time, screen active periods, and operation frequency information. The cleaned data is packaged according to a unified event format, and a calibrated timestamp and a data source identifier are added to each data item. The data source identifier is used to distinguish the type of data. Arrange all encapsulated events in ascending order of timestamp to generate a set of attendance events to be verified.

4. The attendance management method based on multi-dimensional data verification as described in claim 1, characterized in that, After cleaning and parsing the multi-source data to generate a set of attendance events to be verified, the process also includes the steps of constructing an electronic fence and performing dual location verification. Preset the latitude and longitude boundaries of the office area or field location, and generate polygonal or circular electronic fences; Obtain the location coordinates of the mobile terminal in the attendance event to be verified, and use a distance calculation algorithm to calculate the shortest distance between the location coordinates of the mobile terminal and the boundary of the electronic fence. If the shortest distance is less than or equal to the preset threshold, the location is determined to be valid; otherwise, the wireless network service set identifier connected at the time of check-in is obtained. If the corresponding wireless network service set identifier matches the pre-stored company intranet service set identifier, the location is determined to be valid. The valid location marker is stored as a feature dimension in the attendance event to be verified, and used for the subsequent calculation of the on-duty confidence score.

5. The attendance management method based on multi-dimensional data verification as described in claim 1, characterized in that, The specific construction process of the multi-dimensional verification model is as follows: Obtain multi-source data of employees within historical attendance periods as a training sample set; Feature extraction is performed on various types of data in the training sample set to generate time matching features, location matching features, device trust features, and behavioral activity features. The time matching features reflect the degree of deviation between the clock-in time and the scheduled shift time; the location matching features reflect the spatial relationship between the clock-in coordinates and the electronic fence of the office area; the device trust features reflect whether the clock-in device belongs to a trusted device bound to the employee; and the behavioral activity features reflect the continuity and frequency of employee operations on the office terminal. Based on the extracted features, the mapping relationship between each feature and the actual on-duty status is fitted by the logistic regression algorithm to generate the weight coefficients corresponding to each feature. The weight coefficients are used to characterize the importance of different features in the determination of attendance authenticity. Based on the weight coefficients obtained after training, a weighted scoring function is constructed to score the on-duty confidence of real-time collected attendance events to be verified. Based on the false positive rate and false negative rate in historical attendance data, a dynamically adjusted confidence threshold is set. The confidence threshold is used to determine whether the attendance event to be verified belongs to the questionable attendance event.

6. The attendance management method based on multi-dimensional data verification as described in claim 1, characterized in that, The specific process of responding to the flag of a questionable attendance event, triggering a verification process, pushing an on-duty status verification request to the employee's terminal, receiving verification feedback data containing real-time location information and on-site images from the employee's terminal, and correcting the questionable attendance event based on the verification feedback data to generate a calibrated attendance record is as follows: The timestamp corresponding to the questionable attendance event is obtained as the target verification time point. A request message containing verification instructions is sent to the employee terminal through the message push service. The request message requires the employee to start the terminal camera to take a live image with the real-time background within a preset response time, and simultaneously call the positioning module of the employee terminal to obtain the current real-time location coordinates. Receive verification feedback data packets uploaded by employee terminals. The verification feedback data packets include at least the on-site image with a time watermark, real-time location coordinates, and the terminal device identifier at the time of shooting. The time watermark information in the on-site image is extracted by optical character recognition technology and compared with the target verification time point to determine whether the shooting time is within the preset response time. If it exceeds the time limit, the verification time is determined to be overdue and the original doubtful mark is maintained. If the shooting time is within the preset response time, the background features of the on-site image are analyzed by the image recognition algorithm, the features of office environment markers or field locations are extracted, and they are matched with the pre-stored office area image feature library or the field location image features submitted by employees to calculate the image environment matching degree. The received real-time location coordinates are compared with the mobile terminal location data at the time of the suspected attendance incident, the location offset is analyzed, and the confidence level of the verification result is generated by combining the image environment matching degree. If the confidence level of the verification result exceeds the preset correction threshold, the questionable attendance event will be corrected to the verification passed status, and the verification feedback data and correction results will be stored in the attendance record; otherwise, the questionable attendance event will be marked as pending manual review and pushed to the management personnel terminal.

7. The attendance management method based on multi-dimensional data verification as described in claim 6, characterized in that, The specific process of comparing the received real-time location coordinates with the mobile terminal location data at the time of the questionable attendance event, analyzing the location offset, and combining it with the image environment matching degree to comprehensively generate the confidence level of the verification result is as follows: Extract the initial location coordinates recorded in the questionable attendance events, combine them with the verification location coordinates uploaded in real time by the employee terminal, calculate the spatial straight-line distance between the two, and obtain the location offset value. The position offset value is input into the preset offset scoring function. When the offset is less than the first threshold, a high confidence weight is added. When the offset is between the first threshold and the second threshold, a medium confidence weight is added. When the offset exceeds the second threshold, a low confidence weight is added. The on-site images are feature extracted by a preset image recognition model. The extracted image features are then compared with the pre-stored office environment image feature library or the standard image features of field locations to generate an image environment matching score. If the image environment matching score is higher than the preset image confidence threshold, the dynamic weighted fusion logic is triggered, and the confidence weight corresponding to the position offset is added to the image environment matching score according to the preset fusion coefficient to generate the initial value of the verification result confidence. If the image environment matching score is lower than or equal to the preset image credibility threshold, the on-site image is determined to be invalid, the verification result confidence is set to zero, and the verification feedback data is marked as unreliable data. The initial confidence level of the verification result is compared with the preset pass threshold. If it is higher than the pass threshold, the verification is deemed to have passed. If it is lower than the pass threshold but higher than the doubt threshold, a second dynamic verification instruction is triggered. If it is lower than the doubt threshold, it is marked as pending manual review.

8. An attendance management system based on multi-dimensional data verification, used to implement the attendance management method based on multi-dimensional data verification as described in any one of claims 1-7, characterized in that, include: The multi-source data acquisition module is used to acquire multi-source data of employees within a preset attendance period; The data preprocessing module, connected to the multi-source data acquisition module, is used to perform timestamp calibration, noise filtering, and legality checks on the multi-source data, and generate a set of attendance events to be verified that includes timestamps and data source identifiers. The attendance rule matching module, connected to the data preprocessing module, is used to match the set of attendance events to be verified with the pre-stored shift information of employees to determine the basic attendance type corresponding to each attendance event to be verified. The multi-dimensional verification engine module is connected to the data preprocessing module and the attendance rule matching module respectively. It is used to perform multi-dimensional weighted fusion analysis on multiple types of data in attendance events to be verified belonging to the same timestamp, calculate the employee on-duty confidence score corresponding to the target timestamp, and mark the attendance event to be verified corresponding to the target timestamp as a questionable attendance event when the on-duty confidence score is lower than a preset threshold. The dynamic interactive verification module, connected to the multi-dimensional verification engine module, is used to push an on-duty status verification request to the employee terminal in response to the marking of a questionable attendance event, and to receive verification feedback data returned by the employee terminal containing real-time location information and on-site images. Based on the verification feedback data, the questionable attendance event is corrected and a calibrated attendance record is generated. The attendance record generation module is connected to the dynamic interactive verification module. It is used to store the verified and corrected attendance events, form the final attendance record, and synchronize the attendance record to the payroll system.

Images