Multi-dimensional graduate behavior-driven pre-officer default early warning intervention method

Abstract

The application discloses a multidimensional graduate behavior-driven pre-warning intervention method for pre-offending, relates to the field of graduate employment management, and discloses the following technical scheme. Multidimensional behavior data generated by a target graduate in an employment compliance period is acquired, and behavior characteristic sequences are formed by collation; a compliance stage is determined according to a current employment state and a time node; a stage-based behavior baseline is constructed in combination with a professional category, a post category, a signing unit type, a signing time node and historical compliance data of the same type of graduates; current behavior data is compared with the stage-based behavior baseline, single-point abnormal behavior is identified, and an abnormal behavior chain is constructed; a compliance risk state is determined according to the abnormal behavior chain, intervention tasks are generated for the graduate end, the school end and the enterprise end, the risk state is updated based on feedback data, and a pre-warning result or a de-warning result is output. The application can improve the timeliness and accuracy of identification of the pre-offending risk of the graduates.

Description

Technical Field

[0001] This invention relates to the field of graduate employment management, and in particular to a pre-emptive early warning and intervention method for resignation breach driven by multi-dimensional graduate behavior. Background Technology

[0002] As the informatization of graduate employment management, university-enterprise collaborative training, and employer recruitment management continues to improve, relevant data on graduates during job seeking, signing contracts, registration, and onboarding processes are increasingly recorded through employment management systems, corporate recruitment systems, online communication platforms, and document submission platforms. For both schools and employers, even after graduates sign employment agreements, confirm offers, or establish employment destinations, they may still face performance risks due to factors such as changes in employment intentions, decreased job matching, failure to submit registration materials on time, abnormal responses from employers, modifications to employment destinations, and contract termination or rescheduling. This can lead to issues such as abandoning registration, temporary breach of contract, failed onboarding, or short-term resignation. Failure to identify these risks before formal breaches or resignations occur will affect the accuracy of school employment management and increase employers' costs in recruiting replacements, addressing vacancies, and facilitating university-enterprise communication.

[0003] Current graduate employment management methods primarily rely on manual follow-up, milestone reminders, or individual information verification. For example, counselors, career advisors, or company personnel manually review graduates' contract status, registration materials, and communication feedback, contacting them for confirmation upon discovering obvious anomalies. While this approach achieves some degree of employment process management, it typically depends on the experience of management personnel and struggles to promptly detect early anomalies hidden within multi-source behavioral data. Furthermore, different graduates are at different stages of employment fulfillment, exhibiting significantly different behavioral patterns. For instance, the contract confirmation stage focuses more on employment intentions and agreement status, the material submission stage on the progress of material completion, and the registration confirmation stage on company notification responses and registration time confirmation. Using uniform thresholds or rules for judgment easily leads to false positives or false negatives.

[0004] Furthermore, existing early warning methods often rely on a single abnormal behavior as the basis for judgment. For example, failing to submit materials, not responding to company notifications, or changing employment destinations triggers an alert. However, a single abnormal behavior does not necessarily indicate a risk of breach of contract by the graduate. In reality, higher risk usually manifests as multiple abnormal behaviors occurring consecutively within a certain time window, forming a temporally related chain of abnormal behaviors across different stages of contract fulfillment. For example, a graduate frequently browsing other job postings after signing a contract, subsequently delaying the submission of registration materials, and then exhibiting a decline in responsiveness to company notifications or inquiries about contract termination or rescheduling are often a stronger indicator of contract fulfillment risk than a single abnormal behavior. Existing methods lack a comprehensive identification of the temporal proximity, behavioral dimension correlation, and continuity of contract fulfillment stages among abnormal behaviors, making it difficult to generate more accurate early warnings.

[0005] Furthermore, existing early warning systems often end their processing after issuing a risk alert. Information such as whether subsequent communication and confirmation have been completed, whether supplementary materials have been submitted, whether companies have provided feedback, whether schools have followed up, and whether graduates' employment destinations are stable is typically not updated in a timely manner, making it difficult to dynamically correct the warning results. If graduates have already confirmed or submitted supplementary materials after the warning, but the system still maintains a high-risk status, it will lead to repeated interventions; if new abnormal behaviors appear after several prognoses, the system may not be able to upgrade and handle them in a timely manner.

[0006] To address this, a multi-dimensional, graduate behavior-driven early warning and intervention method for breach of contract upon resignation is proposed. Summary of the Invention

[0007] The main objective of this invention is to provide a multi-dimensional graduate behavior-driven early warning and intervention method for resignation breaches, which can effectively solve the problems in the background technology.

[0008] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0009] A multi-dimensional, graduate behavior-driven approach to early warning and intervention for breach of contract during resignation includes the following steps:

[0010] S1. Obtain multi-dimensional behavioral data generated by the target graduate during the employment contract fulfillment period, and organize the multi-dimensional behavioral data according to the time window to obtain a behavioral feature sequence; wherein, the employment contract fulfillment period is from the time when the target graduate signs the employment agreement, confirms the offer of employment or forms the employment destination confirmation information, until the target graduate reports to the workplace and reaches the preset stable period.

[0011] S2. Determine the stage of contract fulfillment for the target graduates based on their current employment status and current timeframe.

[0012] S3. Based on the target graduates' major category, job category, type of contracting unit, contract signing time, and historical performance data of similar graduates, construct a phased behavioral baseline corresponding to the current performance stage;

[0013] S4. Compare the multidimensional behavioral data of the target graduate in the current performance stage with the staged behavioral baseline to obtain the behavioral offset features corresponding to each behavioral dimension.

[0014] S5. Identify single-point abnormal behavior based on the behavioral deviation characteristics, and construct an abnormal behavior chain according to the occurrence time, behavioral dimension, and performance stage of the single-point abnormal behavior.

[0015] S6. Determine the performance risk status of the target graduate based on the duration, stage span, abnormal intensity, and behavior sequence of the abnormal behavior chain;

[0016] S7. Match the corresponding intervention strategy based on the performance risk status, and generate an intervention task for at least one of the graduate end, school end, and enterprise end;

[0017] S8. Collect feedback data after the execution of the intervention task, update the performance risk status of the target graduate based on the feedback data, and output the pre-departure breach warning result or the warning cancellation result based on the updated performance risk status.

[0018] Furthermore, the performance stage includes at least two of the following: employment intention stage, contract confirmation stage, material submission stage, registration confirmation stage, onboarding transition stage, and stable performance stage.

[0019] Each performance stage corresponds to a set of preset behavioral dimensions and at least one key performance task, and different performance stages correspond to different anomaly judgment conditions; the key performance tasks include at least one of the following: confirmation of employment intention, confirmation of employment agreement, submission of reporting materials, confirmation of company notification, confirmation of reporting time, and confirmation of onboarding information.

[0020] Furthermore, the multidimensional behavioral data includes at least two of the following: employment system login data, employment agreement status data, job browsing data, resume submission data, interview invitation data, registration material submission data, company notification reading data, company message reply data, employment destination modification data, contract termination application data, contract rescheduling application data, and questionnaire feedback data.

[0021] The multidimensional behavioral data is organized according to time windows, including: performing identity desensitization processing on the multidimensional behavioral data to obtain desensitized behavioral data; performing time alignment processing on the desensitized behavioral data according to daily, weekly, or performance stage windows; and converting the time-aligned desensitized behavioral data into behavioral feature sequences according to preset behavioral dimension encoding rules.

[0022] Furthermore, a phased behavioral baseline corresponding to the current performance stage is constructed, including: selecting historical graduate samples that match the target graduates from historical graduate performance data based on the target graduates' major category, job category, type of contracting unit, and contract signing time, forming a similar sample set; calculating the similarity of the historical graduate samples with the target graduates in terms of major category, job category, type of contracting unit, and contract signing time based on the similarity of the historical graduate samples with the target graduates in terms of major category, job category, type of contracting unit, and contract signing time. Sample matching degree between a sample of historical graduates and the target graduates:

[0023]

[0024] in, Indicates the first Sample matching degree between a sample of historical graduates and the target graduates Indicates the similarity of professional categories. Indicates the similarity of job categories. Indicates the similarity of the types of contracting entities. Indicates the similarity of the signing time points. , , , Let represent the corresponding weight coefficients, and satisfy . ;

[0025] The similarity of professional categories, job categories, and types of contracting units are determined according to preset classification codes or preset classification hierarchical relationships, and the similarity of contract signing time nodes is determined according to the time interval between the contract signing of historical graduate samples and target graduates. The value range of each similarity is 0 to 1.

[0026] Based on the sample matching degree And the historical behavior values ​​of the historical graduate sample at the current performance stage, to calculate the first The first phase of the performance Phased behavioral baselines for each behavioral dimension:

[0027]

[0028] in, Indicates the first The first phase of the performance Staged behavioral baselines for each behavioral dimension Indicates the first A sample of historical graduates in the first The first phase of the performance Historical behavioral values ​​for each behavioral dimension This represents the number of historical graduate samples in the same sample set.

[0029] Furthermore, the multidimensional behavioral data of the target graduate during the current performance phase is compared with the staged behavioral baseline to obtain the behavioral offset features corresponding to each behavioral dimension, including: obtaining the target graduate's behavior data during the current performance phase. The first phase of the performance Actual behavioral values ​​of each behavioral dimension Based on the actual behavior value Phased behavioral baseline and the standard deviation of similar sample sets under the corresponding behavioral dimension. Calculate the behavioral offset intensity:

[0030]

[0031] in, Indicates the first The first phase of the performance Behavioral offset strength of each behavioral dimension For the first sample in the same type of sample set The first phase of the performance Standard deviation of historical behavioral values ​​for each behavioral dimension Indicates the smoothing coefficient;

[0032] Calculate the comprehensive behavioral offset index based on the behavioral offset intensity across multiple behavioral dimensions:

[0033]

[0034] in, Indicates the target graduates in the A comprehensive behavioral deviation index for each stage of performance. Indicates the first The weight coefficients of each behavioral dimension, Indicates the number of behavioral dimensions;

[0035] When the behavior offset intensity Exceeding the single-dimensional offset threshold of the corresponding behavioral dimension, or the comprehensive behavioral offset index. When the overall offset threshold for the corresponding performance stage is exceeded, the corresponding behavior will be identified as a single point of abnormality.

[0036] Furthermore, an abnormal behavior chain is constructed based on the occurrence time, behavior dimension, and performance stage of a single abnormal behavior, including: representing each single abnormal behavior as an abnormal behavior node, wherein the abnormal behavior node includes at least the occurrence time, the performance stage to which it belongs, the behavior dimension to which it belongs, and the abnormal intensity; sorting multiple abnormal behavior nodes according to their occurrence time, and selecting two abnormal behavior nodes whose time interval is within a preset association time window as abnormal behavior nodes to be associated; calculating the chain association strength based on the time proximity, behavior dimension association, and performance stage continuity between the two abnormal behavior nodes to be associated.

[0037]

[0038] in, Indicates a single point of abnormal behavior Single point of abnormal behavior The strength of the chain association between them Indicates temporal proximity. Indicates the correlation between behavioral dimensions. Indicates the continuity of the performance phase. , , Let represent the corresponding weight coefficients, and satisfy . ;

[0039] The temporal proximity is determined based on the time interval between the occurrence of two single-point abnormal behaviors, and decreases as the time interval increases; the behavioral dimension correlation is determined based on a preset behavioral dimension correlation matrix, which is an m×m matrix with matrix elements ranging from 0 to 1, used to represent the degree of correlation between any two behavioral dimensions, and the preset behavioral dimension correlation matrix is ​​determined based on the co-occurrence relationship of historical abnormal behaviors or preset correlation rules.

[0040] The continuity of the performance stage is determined based on the stage interval between the performance stages of the two single-point abnormal behaviors, and decreases as the stage interval increases; when the chain association strength is greater than the preset association threshold, the corresponding two single-point abnormal behaviors are connected as adjacent nodes in the same abnormal behavior chain.

[0041] Furthermore, based on the duration, stage span, abnormal intensity, and behavioral sequence of the abnormal behavior chain, the performance risk status of the target graduate is determined, including: determining the chain abnormal intensity of the abnormal behavior chain by weighting the number of abnormal behavior nodes in the abnormal behavior chain, the abnormal intensity of each abnormal behavior node, the chain association strength between adjacent abnormal behavior nodes, the number of performance stages spanned by the abnormal behavior chain, and the duration of the abnormal behavior chain; and determining the performance risk index of the target graduate by weighting the chain abnormal intensity, the comprehensive behavioral deviation index under the current performance stage, the proportion of uncompleted key performance tasks under the current performance stage, and historical ineffective intervention factors. The proportion of uncompleted key performance tasks is the ratio of the number of uncompleted key performance tasks in the current performance stage to the total number of key performance tasks that should be completed in the current performance stage; the historical intervention ineffective factor is determined based on the number of times the risk status of the target graduate did not decrease after historical intervention and the total number of historical interventions in the current employment performance cycle; the performance risk status is divided into normal performance status, attention status, early warning status, high-risk status and intervention escalation status, and assigned status levels of 0, 1, 2, 3 and 4 respectively; based on the comparison result of the performance risk index and the preset status threshold, the performance risk status of the target graduate is determined as the performance risk status of the corresponding status level.

[0042] Furthermore, feedback data is collected after the intervention task is executed, and the performance risk status of the target graduates is updated based on the feedback data. This includes obtaining at least one type of feedback data: graduate confirmation feedback, material supplementation feedback, company communication feedback, school follow-up feedback, stable employment destination feedback, and contract termination / rescheduling processing feedback. The feedback data is converted into feedback evaluation values ​​ranging from 0 to 1, and the intervention effectiveness index is calculated based on multiple feedback evaluation values.

[0043]

[0044] in, Indicating the effectiveness index of the intervention, This indicates the graduate's confirmed feedback value. This represents the enterprise communication feedback value. This indicates that the school is following up on the feedback values. This indicates feedback value for supplementary material submission or completion of key performance tasks. , , , Let represent the corresponding weight coefficients, and satisfy . ;

[0045] When the intervention effectiveness index meets the preset effective conditions and no new single-point abnormal behavior is added within the preset observation window after the intervention task is executed, the performance risk status level of the target graduate is reduced according to the preset adjustment range; when the intervention effectiveness index does not meet the preset effective conditions, or a new single-point abnormal behavior is added within the preset observation window after the intervention task is executed, the performance risk status level of the target graduate is maintained or increased; when the updated performance risk status level is less than the status level corresponding to the warning status, the warning is lifted; when the updated performance risk status level is greater than or equal to the status level corresponding to the warning status, the corresponding intervention task continues to be generated.

[0046] A multi-dimensional graduate behavior-driven early warning and intervention system for pre-exit breach of contract includes:

[0047] The data acquisition module is used to acquire multidimensional behavioral data generated by the target graduates during their employment contract fulfillment period, and to organize the multidimensional behavioral data according to time windows to obtain a behavioral feature sequence.

[0048] The phase determination module is used to determine the performance phase of the target graduate based on their current employment status and current time point.

[0049] The behavioral baseline construction module is used to construct a phased behavioral baseline corresponding to the current performance stage based on the target graduate's major category, job category, type of contracting unit, contract signing time, and historical performance data of similar graduates.

[0050] The behavior offset analysis module is used to compare the multidimensional behavior data of the target graduate in the current performance stage with the staged behavior baseline to obtain the behavior offset features corresponding to each behavior dimension.

[0051] An abnormal behavior identification module is used to identify single-point abnormal behavior based on the behavior offset features;

[0052] The abnormal behavior chain construction module is used to construct abnormal behavior chains according to the occurrence time, behavior dimension, and performance stage of a single abnormal behavior.

[0053] The risk status determination module is used to determine the performance risk status of the target graduate based on the duration, stage span, abnormal intensity, and behavior sequence of the abnormal behavior chain.

[0054] The intervention task generation module is used to match the corresponding intervention strategy based on the performance risk status and generate intervention tasks for at least one of the graduate end, school end and enterprise end.

[0055] The feedback update module is used to collect feedback data after the intervention task is executed, and update the performance risk status of the target graduates based on the feedback data.

[0056] The early warning output module is used to output early warning results for resignation default or to lift the early warning result based on the updated performance risk status.

[0057] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described method.

[0058] Compared with the prior art, the present invention has the following beneficial effects:

[0059] 1. This solution obtains multidimensional behavioral data generated by target graduates during their employment contract fulfillment period, and determines their contract fulfillment stage by combining their current employment status and current time point. Then, it constructs a staged behavioral baseline based on professional category, job category, type of contracting unit, contract signing time point, and historical contract fulfillment data of similar graduates. This allows risk identification to move beyond a single fixed threshold and instead make dynamic judgments based on the graduate's stage and the behavioral characteristics of similar samples, thereby improving the accuracy of identifying abnormal behaviors at different contract fulfillment stages.

[0060] 2. This solution compares the multi-dimensional behavioral data of the target graduates in the current performance stage with the staged behavioral baseline, calculates the behavioral deviation features corresponding to each behavioral dimension, and identifies single-point abnormal behaviors accordingly. It can transform scattered data such as employment system login, agreement status changes, material submission, company notification response, employment destination modification, and contract termination and rescheduling into quantifiable behavioral abnormal features, reducing the lag and subjectivity caused by relying solely on human experience judgment.

[0061] 3. This solution further constructs an abnormal behavior chain according to the occurrence time, behavior dimension, and performance stage of a single abnormal behavior, and calculates the chain correlation strength based on time proximity, behavior dimension correlation, and performance stage continuity. This enables the system to identify the temporal and stage correlation relationships between multiple abnormal behaviors, avoids false alarms caused by triggering warnings based on a single abnormal behavior, and can detect performance risks that gradually accumulate from signing the contract to reporting for duty earlier.

[0062] 4. This invention determines the performance risk status of target graduates based on the duration, stage span, abnormal intensity, and sequence of abnormal behavior chains, and divides the risk status into normal performance status, attention status, early warning status, high-risk status, and intervention escalation status. This allows the early warning results to be output in a hierarchical status format, which facilitates schools, enterprises, and graduates to adopt differentiated processing strategies according to different risk levels, thereby improving the pertinence of pre-intervention.

[0063] 5. After generating intervention tasks, this invention continues to collect data such as graduate confirmation feedback, material supplementation feedback, enterprise communication feedback, school follow-up feedback, employment stability feedback, and contract termination / rescheduling feedback. Based on this feedback data, the project updates the target graduate's performance risk status, achieving a closed-loop process of "early warning—intervention—feedback—reassessment." When the intervention is effective and no new abnormal behavior occurs, the risk status level can be lowered or the early warning lifted. When the intervention is ineffective or new abnormal behavior occurs, the risk status level can be maintained or raised, and intervention tasks can continue to be generated, thereby improving the dynamic correction capability and continuous management effectiveness of the early warning and intervention process. Attached Figure Description

[0064] Figure 1 This is a schematic diagram of the method flow of the present invention;

[0065] Figure 2 This is a schematic diagram of the system modules of the present invention. Detailed Implementation

[0066] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0067] Example 1

[0068] like Figure 1-2 As shown, a multi-dimensional graduate behavior-driven early warning and intervention method for pre-existing breaches of employment contracts is presented. This method can be deployed in university employment management platforms, university-enterprise collaborative employment management systems, enterprise recruitment management systems, or third-party employment service platforms. The system collects multi-dimensional behavioral data of graduates during their employment contract fulfillment period to pre-identify contract fulfillment risks and generates corresponding intervention tasks when the risk reaches a preset level.

[0069] The employment contract fulfillment period begins from the date the target graduate signs the employment agreement, confirms the offer of employment, or establishes employment destination confirmation information, until the target graduate reports for duty and reaches the preset stability period. The preset stability period can be set according to the management needs of the school or employer, for example, it can be set to 1 month, 3 months, or 6 months after reporting for duty.

[0070] Specifically, the method includes the following steps.

[0071] S1. Obtain multidimensional behavioral data and form a behavioral feature sequence:

[0072] The system acquires multi-dimensional behavioral data generated by target graduates during their employment contract fulfillment period. This multi-dimensional behavioral data can come from employment management systems, recruitment systems, material submission systems, enterprise notification platforms, questionnaire systems, school employment management terminals, and school-enterprise communication terminals.

[0073] In one implementation, the multidimensional behavioral data includes at least two of the following: employment system login data, employment agreement status data, job browsing data, resume submission data, interview invitation data, registration material submission data, company notification reading data, company message reply data, employment destination modification data, contract termination application data, contract rescheduling application data, and questionnaire feedback data.

[0074] For example, the system can collect the following data: employment system login data, to reflect whether graduates continue to pay attention to the employment system; employment agreement status data, to reflect whether graduates have completed signing the contract and whether there have been any changes to the contract; job browsing data and resume submission data, to reflect whether graduates continue to frequently pay attention to other jobs after signing the contract; registration material submission data, to reflect whether graduates have completed the submission of registration materials as required; company notification reading data and company message reply data, to reflect the communication and response between graduates and employers; and employment destination modification data, contract termination application data, and contract re-signing application data, to reflect whether the graduate's employment destination is stable.

[0075] After obtaining the above data, the system first performs identity desensitization processing on the data, such as replacing the graduate's real name, ID number, mobile phone number and other identity information with the target graduate number; then it performs time alignment processing according to daily, weekly or performance stage windows so that data from different sources can fall into a unified time window; and then, according to the preset behavioral dimension coding rules, it converts the time-aligned data into a behavioral feature sequence.

[0076] For example, for a specific target graduate, their behavioral characteristics within a certain time window can be represented as:

[0077]

[0078] in, Indicates the target graduates in the behavioral feature vectors within a time window Indicates the first The behavioral feature values ​​corresponding to each behavioral dimension Indicates the number of behavioral dimensions.

[0079] Behavioral characteristics can be the number of behaviors, completion rate, response time, number of status changes, or normalized behavioral scores. For example, data on registration material submission can be converted into a material completion rate, data on company message replies can be converted into an average response time, and data on job postings can be converted into the number of job postings per unit of time.

[0080] S2. Determine the stage of contract fulfillment for the target graduates:

[0081] The system determines the stage of contract fulfillment for each target graduate based on their current employment status and the current timeframe.

[0082] In one implementation, the performance phase includes the employment intention phase, the contract confirmation phase, the document submission phase, the registration confirmation phase, the onboarding transition phase, and the stable performance phase. Different performance phases correspond to different behavioral dimensions and key performance tasks.

[0083] For example, during the contract confirmation stage, the system focuses on behavioral dimensions such as the status of the employment agreement, the confirmation time, and the number of times the employment destination has been modified; during the material submission stage, the system focuses on behavioral dimensions such as the progress of the reporting materials submission, the number of times the materials have been returned, and the duration of supplementary material submission; during the reporting confirmation stage, the system focuses on behavioral dimensions such as reading of company notices, confirmation of reporting time, and the duration of message replies; and during the onboarding transition stage, the system focuses on behavioral dimensions such as confirmation of onboarding information, company communication feedback, and changes in employment destination before onboarding.

[0084] By dividing the system into stages, the system can avoid using the same judgment criteria for different employment stages, thereby improving the adaptability of risk identification.

[0085] S3. Establish a phased behavioral baseline:

[0086] The system constructs a phased behavioral baseline corresponding to the current stage of contract performance based on the target graduate's major category, job category, type of employer, contract signing time, and historical performance data of similar graduates.

[0087] First, the system filters historical graduate samples that match the target graduates from historical graduate performance data, forming a set of similar samples. This set of similar samples can be filtered based on at least two of the following: major category, job category, type of employing unit, and contract signing time.

[0088] Then, the system calculates the first... Sample matching degree between a sample of historical graduates and the target graduates:

[0089]

[0090] in, Indicates the first Sample matching degree between a sample of historical graduates and the target graduates Indicates the similarity of professional categories. Indicates the similarity of job categories. Indicates the similarity of the types of contracting entities. Indicates the similarity of the signing time points. , , , Let represent the corresponding weight coefficients, and satisfy . ;

[0091] In one implementation, the similarity of professional categories, job categories, and employer types can be determined based on preset classification codes or hierarchical relationships. For example, the similarity for the same professional category is set to 1, the similarity for different specific majors within the same major category is set to 0.6 to 0.9, and the similarity for different major categories is set to 0 to 0.5. The similarity for job categories and employer types can be determined in a similar manner. The similarity of signing time points can be determined based on the time interval between the signing of historical graduate samples and the target graduate; the smaller the time interval, the higher the similarity.

[0092] When the sample matching degree meets the preset conditions, the corresponding historical graduate sample is added to the same type of sample set. The system calculates the [number]th [item] based on the historical behavior values ​​of the historical graduate samples in the same type of sample set. The first phase of the performance Phased behavioral baselines for each behavioral dimension:

[0093]

[0094] in, Indicates the first The first phase of the performance Staged behavioral baselines for each behavioral dimension Indicates the first A sample of historical graduates in the first The first phase of the performance Historical behavioral values ​​for each behavioral dimension This represents the number of historical graduate samples in the same sample set.

[0095] For example, during the material submission stage, if the first If the behavioral dimension is "material submission completion rate", then It can represent the average completion rate of materials for similar history graduates within the same stage; if the first The behavioral dimension is "average response time for enterprise messages". It can represent the average response time of graduates with similar historical backgrounds within the same stage.

[0096] S4. Calculate behavioral offset features:

[0097] The system compares the multidimensional behavioral data of the target graduates in the current performance stage with the staged behavioral baseline to obtain the behavioral offset features corresponding to each behavioral dimension.

[0098] Specifically, the system obtains the target graduates in the first... The first phase of the performance Actual behavioral values ​​of each behavioral dimension And based on the actual behavior value Phased behavioral baseline and the standard deviation of similar sample sets under the corresponding behavioral dimension. Calculate the behavioral offset intensity:

[0099]

[0100] in, Indicates the first The first phase of the performance Behavioral offset strength of each behavioral dimension For the first sample in the same type of sample set The first phase of the performance Standard deviation of historical behavioral values ​​for each behavioral dimension This represents the smoothing coefficient, used to avoid the denominator being zero.

[0101] Then, the system calculates a comprehensive behavioral offset index based on the behavioral offset intensity across multiple behavioral dimensions:

[0102]

[0103] in, Indicates the target graduates in the A comprehensive behavioral deviation index for each stage of performance. Indicates the first The weight coefficients of each behavioral dimension, Indicates the number of behavioral dimensions.

[0104] In one implementation, different weighting coefficients can correspond to different performance stages. For example, in the material submission stage, the weight of data related to the submission of registration materials can be higher than that of data related to job browsing; in the registration confirmation stage, the weight of data related to reading company notifications and replying to company messages can be higher than that of data related to logging into the employment system.

[0105] When the behavior offset intensity Exceeding the single-dimensional offset threshold of the corresponding behavioral dimension, or the comprehensive behavioral offset index. When the overall offset threshold for the corresponding performance stage is exceeded, the system will identify the corresponding behavior as a single point of abnormality.

[0106] For example, if a target graduate fails to read the company's notification for an extended period during the registration confirmation phase, and this behavior deviates significantly from the baseline of phased behavior of similar samples, the system can identify "failure to read company notification in a timely manner" as a single point of abnormal behavior. If a target graduate simultaneously exhibits delayed material submission, increased response time to company messages, and an increased number of times their employment destination is modified, the system can determine that they have a comprehensive abnormality based on the comprehensive behavioral deviation index.

[0107] S5. Constructing an abnormal behavior chain:

[0108] After identifying single-point abnormal behavior based on behavioral deviation characteristics, the system constructs an abnormal behavior chain according to the occurrence time, behavioral dimension, and performance stage of the single-point abnormal behavior.

[0109] In one implementation, the system represents each single-point abnormal behavior as an abnormal behavior node, which includes at least the occurrence time, the corresponding performance stage, the corresponding behavior dimension, and the abnormality intensity. The abnormality intensity can be determined based on the behavior offset intensity of the corresponding single-point abnormal behavior.

[0110] For example, an abnormal behavior node can be represented as:

[0111]

[0112] in, Indicates the time of occurrence of a single point of abnormal behavior. This indicates the performance stage to which the single point of abnormal behavior belongs. This indicates the behavioral dimension to which a single point of abnormal behavior belongs. This indicates the intensity of the abnormal behavior at a single point.

[0113] The system sorts multiple abnormal behavior nodes according to their occurrence time and selects two abnormal behavior nodes whose time interval falls within a preset association time window as the abnormal behavior nodes to be associated. For two abnormal behavior nodes to be associated, the system calculates the chain association strength based on time proximity, behavioral dimension association, and continuity of performance stages:

[0114]

[0115] in, Indicates a single point of abnormal behavior Single point of abnormal behavior The strength of the chain association between them Indicates temporal proximity. Indicates the correlation between behavioral dimensions. Indicates the continuity of the performance phase. , , Let represent the corresponding weight coefficients, and satisfy . ;

[0116] Among them, the temporal proximity is determined based on the time interval between the occurrence of two single-point abnormal behaviors, and decreases as the time interval increases; the behavioral dimension correlation is determined based on the preset behavioral dimension correlation matrix; the performance stage continuity is determined based on the stage interval between the performance stages in which the two single-point abnormal behaviors are located, and decreases as the stage interval increases.

[0117] In one implementation, the preset behavioral dimension association matrix is ​​an m×m matrix, with matrix elements ranging from 0 to 1, used to represent the degree of association between any two behavioral dimensions. This matrix can be determined based on the co-occurrence relationship of historical abnormal behaviors or based on preset association rules. For example, the association degree between abnormal job browsing and abnormal resume submission can be high, as can the association degree between abnormal registration material submission and abnormal company notification response.

[0118] When the chain association strength When the number of occurrences exceeds a preset association threshold, the system connects the two corresponding single-point abnormal behaviors as adjacent nodes in the same abnormal behavior chain. In this way, the system can combine multiple discrete abnormal behaviors into an abnormal behavior chain with temporal sequence and stage relationship.

[0119] For example, if a target graduate shows an abnormally high frequency of browsing job postings during the contract signing and confirmation stage, delays in submitting registration materials during the material submission stage, or fails to read company notices in a timely manner during the registration confirmation stage, the system can construct an abnormal behavior chain based on the temporal proximity, behavioral dimension correlation, and stage continuity among these abnormal behaviors.

[0120] S6. Determine the status of performance risk:

[0121] The system determines the performance risk status of target graduates based on the duration, stage span, abnormal intensity, and sequence of abnormal behavior chains.

[0122] Specifically, the system determines the chain anomaly strength of the abnormal behavior chain by weighting the number of abnormal behavior nodes in the abnormal behavior chain, the anomaly strength of each abnormal behavior node, the chain association strength between adjacent abnormal behavior nodes, the number of performance stages traversed by the abnormal behavior chain, and the duration of the abnormal behavior chain.

[0123] In one implementation, the chain anomaly strength of the abnormal behavior chain can be determined based on the following factors: the more abnormal behavior nodes there are, the higher the chain anomaly strength; the higher the anomaly strength of each abnormal behavior node, the higher the chain anomaly strength; the higher the chain association strength between adjacent abnormal behavior nodes, the higher the chain anomaly strength; the more performance stages the abnormal behavior chain spans, the higher the chain anomaly strength; and the longer the duration of the abnormal behavior chain, the higher the chain anomaly strength.

[0124] Then, the system determines the performance risk index of the target graduates by weighting the chain anomaly intensity, the comprehensive behavioral deviation index under the current performance stage, the proportion of key performance tasks not completed under the current performance stage, and historical intervention ineffective factors.

[0125] The proportion of unfulfilled key performance tasks is the ratio of the number of unfulfilled key performance tasks in the current performance phase to the total number of key performance tasks that should be completed in the current performance phase. The historical intervention ineffectiveness factor is determined based on the number of times the target graduate's risk status did not decrease after historical interventions during the current employment performance cycle, and the total number of historical interventions. When the total number of historical interventions is zero, the historical intervention ineffectiveness factor can take a preset initial value, such as 0.

[0126] The system categorizes contract performance risk status into four levels: normal performance, attention, early warning, high risk, and escalation of intervention, and assigns them corresponding levels of 0, 1, 2, 3, and 4. Based on a comparison between the contract performance risk index and preset status thresholds, the system determines the contract performance risk status of the target graduate to the corresponding level.

[0127] For example, when a target graduate has only a slight delay in submitting materials and has not formed a chain of abnormal behavior, it can be identified as a state of concern; when a target graduate has consecutive delays in submitting materials, decreased response to company notifications, and abnormal changes in employment destination within multiple time windows, it can be identified as a state of warning or high risk; when the risk status has not decreased after multiple interventions and new abnormal behaviors continue to appear, it can be identified as a state of escalated intervention.

[0128] S7. Generate intervention tasks:

[0129] The system matches corresponding intervention strategies based on the performance risk status and generates intervention tasks for at least one of the graduate, school, and enterprise ends.

[0130] In one implementation, if the target graduate is in a state of interest, the system can generate reminder tasks, such as pushing employment information confirmation reminders, material submission reminders, or company notification reading reminders to the graduate's end.

[0131] If a target graduate is in an alert state, the system can simultaneously generate a confirmation task for the graduate and a follow-up task for the school. For example, it can push an employment destination confirmation questionnaire to the graduate and generate a follow-up task for the school, while also indicating the main abnormal nodes in the corresponding abnormal behavior chain.

[0132] If a target graduate is in a high-risk situation, the system can generate collaborative intervention tasks for both the school and the company. For example, it can send a communication and confirmation task to the company and a key follow-up task to the school, requiring feedback on the follow-up results within a preset time.

[0133] If the target graduate is in an escalation intervention state, the system can generate escalation processing tasks, such as prompting the school's employment management department to conduct manual review, or prompting the company to prepare a replacement recruitment plan in advance.

[0134] By implementing tiered interventions, the system can avoid applying the same approach to all abnormal behaviors, thereby improving the targeted nature of intervention tasks.

[0135] S8. Collect feedback data and update the performance risk status:

[0136] The system collects feedback data after the intervention task is executed and updates the performance risk status of the target graduates based on the feedback data.

[0137] In one implementation, the feedback data includes at least one of the following: graduate confirmation feedback, material supplementation feedback, enterprise communication feedback, school follow-up feedback, employment stability feedback, and contract termination / rescheduling processing feedback.

[0138] The system converts the feedback data into feedback evaluation values ​​ranging from 0 to 1, and calculates the intervention effectiveness index based on multiple feedback evaluation values:

[0139]

[0140] in, Indicating the effectiveness index of the intervention, This indicates the graduate's confirmed feedback value. This represents the enterprise communication feedback value. This indicates that the school is following up on the feedback values. This indicates feedback value for supplementary material submission or completion of key performance tasks. , , , Let represent the corresponding weight coefficients, and satisfy . ;

[0141] For example, when a graduate has confirmed their employment destination, the graduate confirmation feedback value can be taken as higher; when a company has completed communication and confirmed that the graduate still intends to join, the company communication feedback value can be taken as higher; when the school has completed follow-up and confirmed that the risk has been reduced, the school follow-up feedback value can be taken as higher; when a graduate submits supplementary registration materials or completes key performance tasks as required, the feedback value for supplementary materials or completion of key performance tasks can be taken as higher.

[0142] When the intervention effectiveness index meets the preset effectiveness conditions and no new single-point abnormal behavior is added within the preset observation window after the intervention task is executed, the system reduces the performance risk status level of the target graduate according to the preset adjustment range. When the intervention effectiveness index does not meet the preset effectiveness conditions, or a new single-point abnormal behavior is added within the preset observation window after the intervention task is executed, the system maintains or increases the performance risk status level of the target graduate.

[0143] When the updated performance risk status level is lower than the status level corresponding to the warning status, the system outputs a result to lift the warning; when the updated performance risk status level is greater than or equal to the status level corresponding to the warning status, the system continues to generate the corresponding intervention task.

[0144] Thus, the system can form a closed-loop processing flow of "behavior collection, stage judgment, baseline construction, anomaly identification, risk warning, intervention tasks, and feedback updates".

[0145] This embodiment also provides a multi-dimensional graduate behavior-driven pre-exit breach warning and intervention system. The system includes a data acquisition module, a stage determination module, a behavior baseline construction module, a behavior deviation analysis module, an abnormal behavior identification module, an abnormal behavior chain construction module, a risk status determination module, an intervention task generation module, a feedback update module, and an early warning output module.

[0146] The data acquisition module is used to acquire multidimensional behavioral data generated by target graduates during their employment contract fulfillment period, and to organize the multidimensional behavioral data according to time windows to obtain behavioral feature sequences.

[0147] The stage determination module is used to determine the performance stage of the target graduate based on the target graduate's current employment status and current time point.

[0148] The behavioral baseline construction module is used to construct a phased behavioral baseline corresponding to the current performance stage based on the target graduate's major category, job category, type of contracting unit, contract signing time, and historical performance data of similar graduates.

[0149] The behavioral deviation analysis module is used to compare the multidimensional behavioral data of the target graduates in the current performance stage with the staged behavioral baseline to obtain the behavioral deviation features corresponding to each behavioral dimension.

[0150] The abnormal behavior recognition module is used to identify single-point abnormal behavior based on behavior offset characteristics.

[0151] The abnormal behavior chain construction module is used to construct abnormal behavior chains according to the occurrence time, behavior dimension, and performance stage of a single abnormal behavior.

[0152] The risk status determination module is used to determine the performance risk status of target graduates based on the duration, stage span, abnormal intensity, and sequence of abnormal behavior chains.

[0153] The intervention task generation module is used to match corresponding intervention strategies based on the performance risk status and generate intervention tasks for at least one of the graduate, school, and enterprise ends.

[0154] The feedback update module is used to collect feedback data after the intervention task is executed, and update the performance risk status of the target graduates based on the feedback data.

[0155] The early warning output module is used to output early warning results for resignation default or to lift early warning results based on the updated performance risk status.

[0156] The modules described above can be deployed on the same server, or they can be distributed across school servers, enterprise servers, and cloud servers. The modules can interact with each other via databases, message queues, or API services.

[0157] This embodiment also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, it implements the aforementioned multi-dimensional graduate behavior-driven pre-emptive warning and intervention method for resignation breaches.

[0158] The computer-readable storage medium can be a read-only memory, random access memory, disk, optical disk, solid-state drive, flash memory, or other media capable of storing program code. When the processor executes the program code, it can complete steps such as acquiring multi-dimensional behavioral data, determining the performance stage, constructing a staged behavioral baseline, analyzing behavioral deviations, constructing abnormal behavior chains, determining the performance risk status, generating intervention tasks, and updating feedback.

[0159] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of this invention is defined by the appended claims and their equivalents.

Claims

1. A multi-dimensional graduate behavior-driven pre-quit pre-warning intervention method, characterized in that, Includes the following steps: S1. Obtain multi-dimensional behavioral data generated by the target graduate during the employment contract fulfillment period, and organize the multi-dimensional behavioral data according to the time window to obtain a behavioral feature sequence; wherein, the employment contract fulfillment period is from the time when the target graduate signs the employment agreement, confirms the offer of employment or forms the employment destination confirmation information, until the target graduate reports to the workplace and reaches the preset stable period. S2. Determine the stage of contract fulfillment for the target graduates based on their current employment status and current timeframe. S3. Based on the target graduates' major category, job category, type of contracting unit, contract signing time, and historical performance data of similar graduates, construct a phased behavioral baseline corresponding to the current performance stage; S4. Compare the multidimensional behavioral data of the target graduate in the current performance stage with the staged behavioral baseline to obtain the behavioral offset features corresponding to each behavioral dimension. S5. Identify single-point abnormal behavior based on the behavioral deviation characteristics, and construct an abnormal behavior chain according to the occurrence time, behavioral dimension, and performance stage of the single-point abnormal behavior. S6. Determine the performance risk status of the target graduate based on the duration, stage span, abnormal intensity, and behavior sequence of the abnormal behavior chain; S7. Match the corresponding intervention strategy based on the performance risk status, and generate an intervention task for at least one of the graduate end, school end, and enterprise end; S8. Collect feedback data after the execution of the intervention task, update the performance risk status of the target graduate based on the feedback data, and output the pre-departure breach warning result or the warning cancellation result based on the updated performance risk status.

2. The multi-dimensional graduate behavior-driven pre-resignation early warning intervention method according to claim 1, characterized in that, The contract performance stage includes at least two of the following: employment intention stage, contract confirmation stage, document submission stage, registration confirmation stage, onboarding transition stage, and stable contract performance stage. Each performance stage corresponds to a set of preset behavioral dimensions and at least one key performance task, and different performance stages correspond to different anomaly judgment conditions; the key performance tasks include at least one of the following: confirmation of employment intention, confirmation of employment agreement, submission of reporting materials, confirmation of company notification, confirmation of reporting time, and confirmation of onboarding information.

3. The multi-dimensional graduate behavior-driven pre-resignation lead warning intervention method according to claim 1, characterized in that, The multidimensional behavioral data includes at least two of the following: employment system login data, employment agreement status data, job browsing data, resume submission data, interview invitation data, registration material submission data, company notification reading data, company message reply data, employment destination modification data, contract termination application data, contract rescheduling application data, and questionnaire feedback data. The multidimensional behavioral data is organized according to time windows, including: performing identity desensitization processing on the multidimensional behavioral data to obtain desensitized behavioral data; performing time alignment processing on the desensitized behavioral data according to daily, weekly, or performance stage windows; and converting the time-aligned desensitized behavioral data into behavioral feature sequences according to preset behavioral dimension encoding rules.

4. The multi-dimensional graduate behavior-driven early warning and intervention method for resignation breach of contract as described in claim 1, characterized in that, Constructing a staged behavior baseline corresponding to the current compliance stage, including: according to the professional category, post category, signing unit type and signing time node of the target graduate, screening the historical graduate samples matched with the target graduate from the historical graduate compliance data to form the same sample set; according to the professional category similarity, post category similarity, signing unit type similarity and signing time node similarity between the historical graduate samples and the target graduate, calculating the sample matching degree between the target graduate and the historical graduate samples : wherein, denotes the sample matching degree between the i-th historical graduate sample and the target graduate, denotes the professional category similarity, denotes the post category similarity, denotes the signing unit type similarity, denotes the signing time node similarity, , , , respectively denote the corresponding weight coefficients, and satisfy ;​ The similarity of professional categories, job categories, and types of contracting units are determined according to preset classification codes or preset classification hierarchical relationships, and the similarity of contract signing time nodes is determined according to the time interval between the contract signing of historical graduate samples and target graduates. The value range of each similarity is 0 to 1. According to the sample matching degree And the historical behavior value of the historical graduate sample at the current compliance stage, the stage behavior baseline of the i-th behavior dimension at the i-th compliance stage is calculated: The stage behavior baseline of the i-th behavior dimension at the i-th compliance stage is calculated: The stage behavior baseline of the i-th behavior dimension at the i-th compliance stage is calculated: wherein, represents the stage behavior baseline of the i-th behavior dimension at the j-th implementation stage, represents the historical behavior value of the i-th behavior dimension at the j-th implementation stage of the i-th historical graduate sample, represents the number of historical graduate samples in the same sample set.​​​​​ 5. The multi-dimensional graduate behavior-driven early warning and intervention method for resignation breach of contract as described in claim 4, characterized in that, The multidimensional behavioral data of the target graduate in the current performance stage is compared with the staged behavioral baseline to obtain the behavioral offset features corresponding to each behavioral dimension, including: obtaining the target graduate's behavior data in the current performance stage. The first phase of the performance Actual behavioral values ​​of each behavioral dimension Based on the actual behavior value Phased behavioral baseline and the standard deviation of similar sample sets under the corresponding behavioral dimension. Calculate the behavioral offset intensity: in, Indicates the first The first phase of the performance Behavioral offset strength of each behavioral dimension For the first sample in the same type of sample set The first phase of the performance Standard deviation of historical behavioral values ​​for each behavioral dimension Indicates the smoothing coefficient; Calculate the comprehensive behavioral offset index based on the behavioral offset intensity across multiple behavioral dimensions: in, Indicates the target graduates in the The comprehensive behavioral deviation index under each performance phase Indicates the first The weight coefficients of each behavioral dimension, Indicates the number of behavioral dimensions; When the behavior offset intensity Exceeding the single-dimensional offset threshold of the corresponding behavioral dimension, or the comprehensive behavioral offset index. When the overall offset threshold for the corresponding performance stage is exceeded, the corresponding behavior will be identified as a single point of abnormality.

6. The multi-dimensional graduate behavior-driven early warning and intervention method for resignation breach of contract as described in claim 5, characterized in that, Constructing an abnormal behavior chain based on the occurrence time, behavior dimension, and performance stage of a single abnormal behavior includes: representing each single abnormal behavior as an abnormal behavior node, wherein the abnormal behavior node includes at least the occurrence time, the performance stage to which it belongs, the behavior dimension to which it belongs, and the abnormality intensity; sorting multiple abnormal behavior nodes according to their occurrence time, and selecting two abnormal behavior nodes whose time interval is within a preset association time window as abnormal behavior nodes to be associated; calculating the chain association strength based on the time proximity, behavior dimension association, and performance stage continuity between the two abnormal behavior nodes to be associated. in, Indicates a single point of abnormal behavior Single point of abnormal behavior The strength of the chain association between them Indicates temporal proximity. Indicates the correlation between behavioral dimensions. Indicates the continuity of the performance phase. , , Let represent the corresponding weight coefficients, and satisfy . ; The temporal proximity is determined based on the time interval between the occurrence of two single-point abnormal behaviors, and decreases as the time interval increases; the behavioral dimension correlation is determined based on a preset behavioral dimension correlation matrix, which is an m×m matrix with matrix elements ranging from 0 to 1, used to represent the degree of correlation between any two behavioral dimensions, and the preset behavioral dimension correlation matrix is ​​determined based on the co-occurrence relationship of historical abnormal behaviors or preset correlation rules. The continuity of the performance stage is determined based on the stage interval between the performance stages of the two single-point abnormal behaviors, and decreases as the stage interval increases; when the chain association strength is greater than the preset association threshold, the corresponding two single-point abnormal behaviors are connected as adjacent nodes in the same abnormal behavior chain.

7. The multi-dimensional graduate behavior-driven early warning and intervention method for resignation breach of contract as described in claim 6, characterized in that, Based on the duration, stage span, abnormal intensity, and behavioral sequence of the abnormal behavior chain, the performance risk status of the target graduate is determined, including: determining the chain abnormal intensity of the abnormal behavior chain by weighting the number of abnormal behavior nodes in the chain, the abnormal intensity of each abnormal behavior node, the chain correlation strength between adjacent abnormal behavior nodes, the number of performance stages spanned by the abnormal behavior chain, and the duration of the abnormal behavior chain; determining the performance risk index of the target graduate by weighting the chain abnormal intensity, the comprehensive behavioral deviation index under the current performance stage, the proportion of uncompleted key performance tasks under the current performance stage, and historical ineffective intervention factors; wherein, the The proportion of uncompleted key performance tasks is the ratio of the number of uncompleted key performance tasks in the current performance stage to the total number of key performance tasks that should be completed in the current performance stage; the historical intervention ineffective factor is determined based on the number of times the risk status of the target graduate did not decrease after historical intervention and the total number of historical interventions in the current employment performance cycle; the performance risk status is divided into normal performance status, attention status, early warning status, high-risk status and intervention escalation status, and assigned status levels of 0, 1, 2, 3 and 4 respectively; based on the comparison result of the performance risk index and the preset status threshold, the performance risk status of the target graduate is determined as the performance risk status of the corresponding status level.

8. The multi-dimensional graduate behavior-driven early warning and intervention method for resignation breach of contract as described in claim 7, characterized in that, The system collects feedback data after the intervention task is executed, and updates the performance risk status of the target graduates based on this data. This includes at least one type of feedback data: graduate confirmation feedback, material supplementation feedback, company communication feedback, school follow-up feedback, stable employment status feedback, and contract termination / rescheduling feedback. The feedback data is then converted into feedback evaluation values ​​ranging from 0 to 1, and an intervention effectiveness index is calculated based on multiple feedback evaluation values. in, Indicating the effectiveness index of the intervention, This indicates the graduate's confirmed feedback value. This represents the enterprise communication feedback value. This indicates that the school is following up on the feedback values. This indicates feedback value for supplementary material submission or completion of key performance tasks. , , , Let represent the corresponding weight coefficients, and satisfy . ; When the intervention effectiveness index meets the preset effective conditions and no new single-point abnormal behavior is added within the preset observation window after the intervention task is executed, the performance risk status level of the target graduate is reduced according to the preset adjustment range; when the intervention effectiveness index does not meet the preset effective conditions, or a new single-point abnormal behavior is added within the preset observation window after the intervention task is executed, the performance risk status level of the target graduate is maintained or increased; when the updated performance risk status level is less than the status level corresponding to the warning status, the warning is lifted; when the updated performance risk status level is greater than or equal to the status level corresponding to the warning status, the corresponding intervention task continues to be generated.

9. A multi-dimensional graduate behavior-driven early warning and intervention system for pre-exit breach of contract, characterized in that: include: The data acquisition module is used to acquire multidimensional behavioral data generated by the target graduates during their employment contract fulfillment period, and to organize the multidimensional behavioral data according to time windows to obtain a behavioral feature sequence. The phase determination module is used to determine the performance phase of the target graduate based on their current employment status and current time point. The behavioral baseline construction module is used to construct a phased behavioral baseline corresponding to the current performance stage based on the target graduate's major category, job category, type of contracting unit, contract signing time, and historical performance data of similar graduates. The behavior offset analysis module is used to compare the multidimensional behavior data of the target graduate in the current performance stage with the staged behavior baseline to obtain the behavior offset features corresponding to each behavior dimension. An abnormal behavior identification module is used to identify single-point abnormal behavior based on the behavior offset features; The abnormal behavior chain construction module is used to construct abnormal behavior chains according to the occurrence time, behavior dimension, and performance stage of a single abnormal behavior. The risk status determination module is used to determine the performance risk status of the target graduate based on the duration, stage span, abnormal intensity, and behavior sequence of the abnormal behavior chain. An intervention task generation module is used to match corresponding intervention strategies based on the performance risk status and generate intervention tasks for at least one of the graduate, school, and enterprise ends. The feedback update module is used to collect feedback data after the intervention task is executed, and update the performance risk status of the target graduates based on the feedback data. The early warning output module is used to output early warning results for resignation default or to lift the early warning result based on the updated performance risk status.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method described in any one of claims 1 to 8.

Images