A college student career development risk intelligent evaluation and grading intervention method and system based on multi-modal behavior trajectory

Abstract

This invention discloses a method and system for intelligent assessment and tiered intervention of career development risks for college students based on multimodal behavioral trajectories, belonging to the fields of artificial intelligence and educational informatization technology. The method includes: collecting students' explicit behavioral data, implicit ability-related data, psychological and motivational-related data, and market environment data; and performing cleaning, standardization, temporal alignment, and windowed fusion on these data; extracting joint features from structured and unstructured data to obtain integrated career development features; identifying risk-driving factors based on causal inference, establishing a career development risk assessment model, and outputting a comprehensive risk level; generating tiered intervention strategies based on the comprehensive risk level, driving factors, historical intervention records, and behavioral response characteristics; and updating the risk assessment model and intervention strategies based on intervention feedback. This invention enables dynamic assessment, tiered early warning, and closed-loop intervention of career development risks for college students.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence and educational informatization, and in particular to a method and system for assessing and grading the career development risks of college students by utilizing multi-source heterogeneous data, causal inference, and intelligent decision-making.

[0002] The proposed technical solution is applicable to scenarios such as college employment guidance, regional education cloud, and career development service platforms. It can be used to dynamically identify and intervene in abnormal behaviors, ability deviations, and psychological fluctuations in students' career development process. Background Technology

[0003] Existing technologies for college student employment services primarily focus on job recommendations, career profile building, or academic early warning analysis. These technologies typically model data based on grades, resumes, or single behavioral patterns, enabling job matching or anomaly alerts, but they struggle to continuously, dynamically, and multidimensionally identify potential risks in the career development process.

[0004] Career development risks are influenced by factors such as behavioral patterns, skill reserves, psychological state, and external employment environment. Existing solutions are still insufficient in terms of data dimensions, analysis depth, and intervention linkage, especially lacking the causal identification of risk-driving factors and the generation mechanism of intervention strategies that are appropriate to the risk level.

[0005] Furthermore, consultation records, assessment results, and behavioral trajectories in university settings are highly sensitive to privacy concerns, and centralized processing models are not conducive to collaborative applications by multiple stakeholders. Therefore, how to form a closed-loop technical solution of "risk identification—tiered early warning—precise intervention—feedback optimization" while ensuring privacy protection has become an urgent technical problem to be solved.

[0006] To address the aforementioned problems, this invention is proposed. Summary of the Invention

[0007] The purpose of this invention is to provide a method and system for intelligent assessment and graded intervention of career development risks for college students based on multimodal behavioral trajectories, in order to solve the problems of insufficient dimensions for career development risk identification, difficulty in characterizing risk driving factors, and lack of closed-loop optimization in the intervention process in the existing technology.

[0008] To achieve the above objectives, this invention provides a method for intelligent assessment and tiered intervention of career development risks for college students based on multimodal behavioral trajectories. The method includes: collecting multimodal data related to career development and performing cleaning, standardization, temporal alignment, and windowing processing to construct a sequence of students' career development states; extracting joint features from structured and unstructured data to obtain integrated career development features; constructing a career development risk assessment model based on the integrated features, identifying risk drivers through causal inference, and establishing multiple risk sub-models to output risk values, driving factors, and comprehensive risk levels; generating and executing tiered intervention strategies based on the comprehensive risk level, risk drivers, historical intervention records, and behavioral response characteristics; and collecting intervention feedback data to update the career development risk assessment model and / or tiered intervention strategies.

[0009] In some implementations, the explicit behavioral data includes course grades, GPA, major ranking, failing grades, internship experience, number of resume submissions, submission-to-interview conversion rate, interview-to-hire conversion rate, campus activity records, and campus card tracking data; the implicit competency-related data includes one or more of the following: course papers, lab reports, competition plans, project documents, and career assessment reports.

[0010] In some implementations, the risk sub-model includes at least one or more of the following: an employment procrastination risk model, a career herd risk model, a skills lag risk model, and an employment anxiety risk model. The employment procrastination risk model uses causal graphs and instrumental variables to identify the causal effects of procrastination tendencies. The career herd risk model uses the difference-in-differences method to identify the impact of career guidance activities on risk. The skills lag risk model uses semantic matching of competency requirements and time-series prediction to assess changes in skills gaps. The employment anxiety risk model uses multimodal fusion of text sentiment features, behavioral anomaly features, and consumption anomaly features to identify anxiety risk.

[0011] In some implementations, the tiered intervention strategy is trained using reinforcement learning. Its state space includes risk level, historical intervention records, and behavioral response characteristics, while its action space includes intervention type, timing of delivery, and delivery channel. The reward function comprehensively considers risk reduction, student feedback scores, and cost penalties. During the strategy execution phase, the execution channel and maximum reach frequency can be determined based on students' historical active periods, channel response rates, and risk levels. Feedback adjustments or manual upgrades can be made based on reach results, appointment completion results, course completion results, and risk change results.

[0012] In some implementations, multi-school collaborative training adopts a federated learning approach, where each participant retains the original data locally and only uploads model parameters or gradients. The central aggregation node performs global model aggregation and adds differential privacy noise during the upload phase. Sensitive data is protected using homomorphic encryption or ciphertext computation.

[0013] Compared with existing technologies, this invention has at least the following beneficial effects: by using temporal fusion and joint feature extraction of multimodal data, it can improve the completeness of career development risk characterization; by identifying risk driving factors through causal inference, it can improve the interpretability and stability of risk assessment results; by using hierarchical intervention strategies and feedback update mechanisms, it can form a closed-loop optimization of risk identification and intervention execution; and in multi-agent collaborative scenarios, it can also reduce the risks brought about by direct sharing of original sensitive data through federated learning and privacy protection mechanisms. Attached Figure Description

[0014] Figure 1 This is a schematic diagram of the overall system architecture in an embodiment of the present invention.

[0015] Figure 2 This is a flowchart of risk modeling based on causal inference in an embodiment of the present invention.

[0016] Figure 3 This is a schematic diagram of multi-dimensional risk identification in an embodiment of the present invention.

[0017] Figure 4 This is a flowchart of the hierarchical intervention decision engine in an embodiment of the present invention.

[0018] Figure 5 This is a diagram of the federated learning deployment architecture in an embodiment of the present invention.

[0019] Figure 6 This is a schematic diagram illustrating the implementation effect evaluation and closed-loop optimization in an embodiment of the present invention. Detailed Implementation

[0020] The present invention will be further described below with reference to the accompanying drawings and specific embodiments. It should be understood that the following embodiments are only used to explain the present invention and are not intended to limit the scope of protection of the present invention.

[0021] like Figure 1 As shown, the system of this invention generally includes a data acquisition layer, a risk modeling layer, an intervention decision-making layer, and an effect evaluation layer. The data acquisition layer is responsible for collecting multimodal data from the academic affairs system, employment system, campus card system, psychological counseling system, and external recruitment platforms; the risk modeling layer is responsible for establishing four types of risk models; the intervention decision-making layer is responsible for formulating tiered intervention plans; and the effect evaluation layer is responsible for conducting A / B testing, effect analysis, and online updates.

[0022] During the data collection and fusion phase, explicit behavioral data, implicit ability data, psychological and motivational data, and market environment data are uniformly coded. Cross-system data association is completed based on student identifiers, mapping data at different time granularities onto a unified timeline to form a time-series archive of student career development. Missing values ​​are filled using statistical or model-based imputation; outliers are identified, corrected, or removed using threshold rules, statistical rules, or anomaly detection models. Furthermore, time-series features such as frequency, rate of change, volatility, and trend are extracted by week, month, or job-seeking stage window.

[0023] For unstructured texts such as course papers, lab reports, competition plans, and project documents, large language models or natural language processing models can be used to extract implicit features such as logical thinking, innovation ability, expression ability, collaboration ability, and technical depth. The extracted results, along with the structured features, can then be input into the subsequent risk model.

[0024] like Figure 2 As shown, in the risk modeling stage, a career development causal graph is first constructed based on students' historical job-seeking behavior, course participation, consultation texts, and market environment information; then, instrumental variables, natural experimental factors, or control group information are selected to identify key driving factors that have a causal effect on risk; and then multidimensional risk models are established to output risk scores and their confidence levels for each dimension.

[0025] In modeling the risk of job procrastination, job expectations, self-efficacy, anxiety levels, and time management skills can be used as candidate factors, while the concentration of coursework deadlines can be used as an instrumental variable to estimate the causal effect of procrastination tendency on abnormal resume submission behavior. If students consistently exhibit a high concentration of submissions before deadlines accompanied by low response rates, their risk of job procrastination can be considered elevated.

[0026] In the modeling of career herd risk, the concentration of job applications, the difference between individual and job matching, and the deviation of salary expectations are calculated. The difference-in-differences method is used to analyze the risk changes of students before and after participating in career guidance courses or consulting services, so as to identify whether the herd risk is caused by information asymmetry, comparison mentality, or insufficient career cognition.

[0027] In the modeling of skills lag risk, the skill tags required for the target position are extracted from the recruitment text and semantically matched with the skills, course achievements and project experience in the student's resume to obtain the ability-demand difference. At the same time, the time series prediction model is used to predict the trend of the skill difference change within a preset time window in the future, so as to output the skills lag risk index and the list of skills gaps.

[0028] In the modeling of employment anxiety risk, sentiment analysis is performed on career counseling records to obtain anxiety intensity, degree of confusion, and negative emotional characteristics. At the same time, data on abnormal fluctuations in job-seeking behavior and abnormal campus card consumption are integrated to identify stressors such as family expectations, peer pressure, and previous failures, thereby outputting the employment anxiety risk level.

[0029] like Figure 3 As shown, the risks of employment delay, career herding, skills lag, and employment anxiety are weighted and integrated to obtain a comprehensive risk value R. This value can be calculated using the formula R = α·R1 + β·R2 + γ·R3 + δ·R4, where each weight is determined based on expert experience and historical samples. An early warning is triggered when any dimension of risk reaches a high level or when the comprehensive risk value exceeds a threshold.

[0030] After an alert is triggered, the system enters a tiered intervention phase. For example... Figure 4 As shown, the student's current risk level, historical intervention records, and behavioral response characteristics serve as state inputs, and the reinforcement learning strategy model selects intervention actions. These intervention actions can be information dissemination measures, consultation service measures, or resource matching measures; further, the timing of the dissemination, the execution channel, and the maximum reach frequency can be determined by combining the student's historical active periods, channel response rates, and risk levels.

[0031] In one embodiment, low-risk students are prioritized to receive information push notifications such as industry trend reports, career competency profiles, and course recommendations; medium-risk students, in addition to information push notifications, may also trigger services such as AI mock interviews and career counseling appointment reminders; high-risk students receive intensive interventions including human counseling, alumni resource matching, and psychological support advice. The implementation channels can be one or more of the following: applications, mini-programs, SMS, email, or human contact, with different upper limits on the frequency of contact set according to different risk levels.

[0032] To avoid student resentment caused by excessive intervention frequency, the reward function of the reinforcement learning model considers risk reduction, student feedback scores, and intervention cost penalties simultaneously, and is trained using a proximate strategy optimization algorithm. As intervention samples accumulate, the system can automatically learn optimal intervention combinations for different student groups at different risk stages. Simultaneously, it monitors reach results, open results, click results, appointment completion results, course completion results, and risk change results. When the response rate falls below a threshold or the risk level fails to decrease within a consecutive preset period, it automatically adjusts the execution channels, push timing, and intervention combination, and triggers escalation of manual consultation when conditions are met.

[0033] like Figure 5As shown, in a multi-university collaborative training scenario, each university retains its original local student data, completes model training locally only, and uploads parameters or gradients to the central aggregation node. The central aggregation node then distributes the updated model parameters after completing global model aggregation. To meet privacy protection requirements, differential privacy noise is added during the upload process, and sensitive data such as consultation records and psychological assessments are protected using homomorphic encryption or ciphertext computation.

[0034] like Figure 6 As shown, the system can evaluate the intervention effect through an A / B testing framework. Students are divided into experimental and control groups, and indicators such as early warning accuracy, recall, F1 score, employment rate, job search cycle, student satisfaction, and complaint rate are compared. The evaluation results are fed back into the risk model and intervention strategy model for periodic updates.

[0035] Example 1: In a single-school deployment scenario, the system connects with the academic affairs system, campus card system, employment information network, and psychological counseling system to collect historical samples of recent graduates for offline training; after going online, the system assesses student risk on a weekly basis and triggers one-on-one guidance appointments or resource matching services for high-risk students.

[0036] Example 2: In the scenario of a regional education cloud platform, multiple universities are involved in federated learning. The central cloud platform aggregates the global model weekly, and each university completes local fine-tuning based on its own student data. This approach can improve the generalization ability of the model while avoiding the sharing of original sensitive data across universities.

[0037] Those skilled in the art will understand that equivalent substitutions or modifications can be made to the above embodiments without departing from the spirit and essence of the present invention, and such substitutions or modifications should all fall within the protection scope defined by the claims of the present invention.

Claims

1. A method for intelligent assessment and graded intervention of career development risks for college students based on multimodal behavioral trajectories, characterized in that, Includes the following steps: S1. Collect multimodal data related to career development. The multimodal data includes at least explicit behavioral data, implicit ability-related data, psychological and motivation-related data, and market environment data. Clean, standardize, and time-series align the various types of data, and construct a sequence of students' career development status according to a preset observation window. S2. Perform joint feature extraction on the structured and unstructured data in the career development state sequence to obtain fusion features that represent students' behavior patterns, ability representations, psychological states and external environment, wherein implicit ability features are extracted from the unstructured data. S3. Based on the fusion features, construct a career development risk assessment model. The career development risk assessment model identifies risk drivers through causal inference and establishes one or more risk sub-models, including at least the risks of job delay, career conformity, skills lag, and job anxiety. The model outputs the risk value, driving factors, and comprehensive risk level of each risk sub-model. S4. Based on the comprehensive risk level, the driving factors, historical intervention records and behavioral response characteristics, determine the combination of intervention measures and the timing of intervention corresponding to the current student from the pre-built intervention measure library, generate a graded intervention strategy and execute it. S5. Collect feedback data after the intervention is implemented, and update the parameters of the career development risk assessment model and / or the graded intervention strategy based on the feedback data to form a closed-loop optimization process of risk identification, strategy generation, intervention implementation and effect feedback.

2. The method according to claim 1, characterized in that, The explicit behavioral data includes one or more of the following: course grades, GPA, major ranking, failing grades, internship experience, number of resume submissions, submission-to-interview conversion rate, interview-to-hire conversion rate, campus activity records, and campus card tracking data; the implicit ability-related data includes one or more of the following: course papers, lab reports, competition plans, project documents, and career assessment reports.

3. The method according to claim 1, characterized in that, The data processing in step S1 includes at least the following sub-steps: associating data from different business systems based on student identifiers; performing duplicate record removal and field format standardization; completing missing values ​​in structured data using statistical imputation or model prediction; identifying, correcting, or removing outliers using preset threshold rules, box plot rules, or anomaly detection models; performing desensitization, word segmentation, or vectorization on unstructured text before feature extraction; mapping data at different time granularities to a unified time axis, generating career development status sequences according to weekly, monthly, or job-seeking stage windows, and extracting frequency features, rate of change features, volatility features, and trend features as time-series input features.

4. The method according to claim 1, characterized in that, In step S2, one or more implicit ability features from logical thinking ability, innovation ability, written expression ability, teamwork ability, problem-solving ability and technical depth are extracted from unstructured data through a large language model or natural language processing model, and then spliced, weighted and fused or fused with structured features.

5. The method according to claim 1, characterized in that, The career development risk assessment model described in step S3 includes at least one of the following risk sub-models: the job procrastination risk model, the career conformity risk model, the skills lag risk model, and the job anxiety risk model. The job procrastination risk model constructs a causal graph influencing job application behavior and uses the concentration of coursework deadlines as an instrumental variable to estimate the causal effect of job procrastination tendency on abnormal job application behavior. The career conformity risk model calculates the concentration of applied positions, the difference between individual and job matching, and the deviation from salary expectations, and uses the difference-in-differences method to identify the impact of participation in career guidance courses on career conformity risk. The skills lag risk model extracts the skill requirements of target positions from recruitment data, performs semantic matching with students' resume skills and learning outcomes to obtain the ability-requirement difference, and uses a time-series prediction model to predict the changing trend of the skill difference within a preset time window. The job anxiety risk model performs sentiment analysis on career counseling records and combines abnormal fluctuations in job application behavior and abnormal campus card spending characteristics for multimodal fusion. The comprehensive career development risk value is calculated according to the following formula: R = α·R1 + β·R2 + γ·R3 + δ·R4, R1, R2, R3, and R4 represent the risk values ​​of employment delay risk, career herd risk, skills lag risk, and employment anxiety risk, respectively, with α, β, γ, and δ being the corresponding weights. An early warning is triggered when the risk level corresponding to any risk sub-model reaches the preset high-risk level, or when the comprehensive career development risk value exceeds the preset early warning threshold.

6. The method according to claim 1, characterized in that, The tiered intervention strategy described in step S4 is generated through a reinforcement learning strategy model. The state space of the reinforcement learning strategy model includes at least the student's current risk level, historical intervention records, and behavioral response characteristics. The action space includes at least the type of intervention measure, the timing of the push, and the push channel. The reward function includes at least the amount of risk reduction, the student's feedback score, and the intervention cost penalty. For low-risk students, information push intervention is implemented. For medium-risk students, a combination of information push and intelligent consultation intervention is implemented. For high-risk students, manual consultation intervention or resource priority matching intervention is implemented.

7. The method according to claim 6, characterized in that, In steps S4 and S5, based on students' historical active periods, channel response rates, and current risk levels, at least one execution channel is selected from applications, mini-programs, SMS, email, or manual contact channels, and a corresponding maximum reach frequency is set. The collected feedback data includes at least reach results, open results, click results, appointment completion results, course completion results, and risk change results. When the response rate is lower than the threshold or the risk level does not decrease within a consecutive preset period, the execution channel, push timing, and intervention measure combination are automatically adjusted, and manual consultation escalation processing is triggered when the escalation conditions are met. The method also includes a privacy-protected collaborative modeling step: in a multi-school or multi-organization collaborative training scenario, each participant trains a risk identification model locally, uploads model parameters or gradients only to the central aggregation node, and adds differential privacy noise during the upload process. Sensitive data is protected using homomorphic encryption or ciphertext computation.

8. A smart assessment and graded intervention system for college students' career development risks based on multimodal behavioral trajectories, characterized in that, include: The data acquisition and time series fusion module is used to collect multimodal data related to career development, and to perform data association, duplicate record removal, field format standardization, missing value completion, outlier identification and correction or removal, time series alignment and windowing processing on various types of data in order to construct a sequence of students' career development status. The feature extraction module is used to perform joint feature extraction on structured and unstructured data to obtain fused features that represent students' behavioral patterns, ability representations, psychological states, and external environment. The risk assessment and classification module is used to construct a career development risk assessment model based on the fusion features, identify risk drivers through causal inference, and establish one or more risk sub-models that include at least the risks of job delay, career conformity, skills lag, and job anxiety, so as to output the risk value, driving factors, and comprehensive risk level of each risk sub-model. The intervention decision-making and execution module is used to determine the combination of intervention measures and the timing of intervention from a pre-built intervention measure library based on the comprehensive risk level, risk driving factors, historical intervention records and behavioral response characteristics, and to generate and execute a graded intervention strategy. The intervention decision-making and execution module includes a channel selection unit, a frequency control unit, a feedback monitoring unit and an escalation processing unit. The feedback optimization module is used to collect feedback data after the intervention is implemented and to update the career development risk assessment model and / or the graded intervention strategy based on the feedback data.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the computer program is executed by the processor, it implements the method as described in any one of claims 1 to 7.

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 as described in any one of claims 1 to 7.

Images