Online and offline linked student ideological and political education effect evaluation system

Abstract

The application discloses an online and offline linkage student ideological and political education effect evaluation system, through offline evidence collection, online behavior burying, evidence alignment and representation, double-channel evidence fusion, counter-fact calibration, index calculation and index generation, portrait construction and evolution, intervention and feedback, and privacy compliance and safety modules, objective, stable and interpretable evaluation of the effect of ideological and political education is realized. The application takes cross-scene consistent representation and bias-free calibration as the core, combines hierarchical portrait and dynamic evolution mechanism, supports personalized intervention and teaching improvement, and gives consideration to evaluation accuracy and data safety compliance.

Description

Technical Field

[0001] This invention relates to the field of educational technology, specifically to an online and offline integrated student ideological and political education effectiveness evaluation system. Background Technology

[0002] With the deep integration of information technology and educational scenarios, ideological and political education is gradually showing a trend of being carried out both online and offline. However, existing methods for evaluating the effectiveness of ideological and political education have significant shortcomings in design and implementation, making it difficult to adapt to the comprehensive evaluation needs of multiple scenarios, multiple modalities, and multiple indicators.

[0003] First, existing assessment systems often focus on measuring knowledge-based outcomes, such as judging students' learning effectiveness through final exam scores or standardized tests. While this method can quantify some cognitive mastery, it cannot comprehensively reflect students' true state in terms of value identification, emotional engagement, and behavioral practice. Especially in ideological and political education, value shaping and emotional resonance are key objectives; relying solely on knowledge assessments leads to distorted evaluations, neglecting students' attitude changes and behavioral performance in real-world situations.

[0004] Secondly, data from online and offline educational activities have long been fragmented. Evidence from offline classrooms, practical activities, and campus-based tasks lacks a unified temporal benchmark and semantic mapping with behavioral data from online courses, discussion forums, and virtual simulation tasks, making it impossible to form a consistent representation across scenarios. This fragmentation makes it difficult for evaluators to track the changing thought patterns of the same student in different environments, and also makes it impossible to analyze the interaction mechanisms between online learning and offline practice.

[0005] Third, existing assessments are susceptible to sampling bias and platform preferences. For example, online learning platforms may be more inclined to record the behavior of users with high-frequency interactions, while ignoring students who participate less frequently but deeply; offline practical activities may suffer from insufficient sample representativeness due to differences in organization methods and participation thresholds. These biases can be amplified in statistical analysis, causing assessment results to deviate from reality and lacking robustness and interpretability.

[0006] Fourth, traditional assessment models often use static indicators and fixed weights, which cannot be dynamically adjusted according to the teaching cycle and students' developmental stages. The effects of ideological and political education are often lagging and cumulative. The educational goals differ across semesters and themes, and a fixed indicator system cannot capture these dynamic changes or provide timely and effective feedback for teaching improvement.

[0007] Fifth, regarding data security and privacy compliance, existing systems often lack robust differential privacy, tiered authorization, and end-to-end auditing mechanisms. Student behavior data involves sensitive information, and vulnerabilities in the collection, transmission, or storage stages could lead to privacy breaches, which contradicts the compliance requirements for educational informatization.

[0008] Furthermore, existing assessment systems are weak in their intervention and feedback mechanisms. Most systems only provide results and fail to automatically generate personalized learning paths or teaching improvement suggestions based on the assessment results. They also lack early warning and referral mechanisms for different risk levels, resulting in a lack of a closed loop between assessment and teaching improvement.

[0009] In summary, existing methods for evaluating the effectiveness of ideological and political education have significant shortcomings in terms of the comprehensiveness of data sources, cross-scenario integration capabilities, bias correction mechanisms, dynamic adaptability, and privacy compliance. There is an urgent need for a new evaluation system that can integrate online and offline multimodal evidence, achieve consistent representation across scenarios, possess bias correction and dynamic evolution capabilities, and meet data security requirements. Summary of the Invention

[0010] The purpose of this invention is to provide an online and offline integrated evaluation system for the effectiveness of ideological and political education for students.

[0011] To achieve the above objectives, this invention provides the following technical solution: an online and offline integrated student ideological and political education effectiveness evaluation system, comprising: an offline evidence collection module for collecting multimodal behavioral evidence of students in physical classrooms, practical activities, and campus scenarios, wherein the evidence includes classroom interaction event sequences, practical activity participation trajectories, situational task decision paths, peer evaluation texts, and facial expression and micro-expression features, and is preprocessed and desensitized locally at edge computing nodes; and an online behavior tracking module for collecting learning behavior sequences, semantic texts, interaction sequences, and task data in online courses, discussion forums, quizzes, and virtual simulation tasks. The system completes path and context selection recording and employs an event tracing mechanism to ensure the integrity and traceability of the behavior sequence. The evidence alignment and representation module aligns offline and online evidence according to student identifiers and timestamps, generating a consistent representation vector across scenarios. This representation vector is obtained by concatenating offline and online evidence sub-vectors and adaptively adjusting channel weights through a gated fusion network. The dual-channel evidence fusion module performs cross-channel weighted fusion of the consistent representation vector based on a cross-channel attention mechanism, outputting a fused evidence vector. The counterfactual calibration module takes the fused evidence vector as input and, based on... The system constructs counterfactual samples based on score matching and dual robust estimation, performs debiasing and calibration on the fused evidence vector, and outputs a calibrated evidence vector. The indicator calculation and index generation module calculates knowledge mastery, value identification, emotional engagement, behavioral practice, and cyber literacy based on the calibrated evidence vector, and generates a comprehensive index of ideological and political education effectiveness based on weighted aggregation. The profile construction and evolution module constructs a hierarchical profile using the comprehensive index and sub-indices as nodes, and dynamically evolves the profile using temporal difference analysis. The intervention and feedback module generates personalized learning interventions and teaching improvement suggestions based on the profile and index. The feedback results are then written back into the teaching process. The privacy compliance and security module performs minimal data collection, differential privacy noise addition, tiered authorization, and end-to-end auditing on the collected and transmitted data. In cross-school or cross-platform assessment scenarios, secure multi-party computation or federated learning is used to achieve data "usable but not visible." The system control and scheduling module adaptively schedules the evidence collection frequency, fusion window, calibration batches, and profile update cycle to ensure the system's real-time performance and stability under large-scale student groups. This defines the overall system architecture and all core modules, forming a complete closed loop from multi-source data collection to assessment, profiling, intervention, and security. Through online and offline dual-channel evidence collection and preprocessing, the comprehensiveness and compliance of the data are guaranteed. Evidence alignment and fusion mechanisms solve the problem of semantic inconsistencies in cross-scenario data. Counterfactual calibration significantly reduces assessment bias. Indicator and index generation enables multi-dimensional comprehensive evaluation. The profiling and intervention modules support personalized teaching improvements. The privacy compliance module ensures data security. The system control module ensures operational efficiency and scalability.The overall solution outperforms existing technologies in terms of comprehensiveness, accuracy, robustness, and security of the assessment.

[0012] Furthermore, the offline evidence collection module includes: a classroom interaction event recording unit, used to record students' speaking frequency, question types, answer depth, group collaboration role switching, and blackboard projection interaction trajectory in class, and analyze the quality of interaction; a practical activity participation trajectory unit, used to record students' check-in path, dwell time, task division, and peer collaboration graph in volunteer service, red study tours, and social surveys, and identify participation and contribution; a situational task decision-making path unit, used to record students' option sequence, hesitation time, reason text, and retrospective modification records in situational tasks such as value clarification, moral dilemmas, and professional ethics, and analyze decision rationality; and a peer evaluation text and facial expression / posture micro-expression feature unit, used to perform semantic parsing on peer evaluation text, extract features from facial expressions and postural micro-expressions during offline peer evaluation, and generate feature vectors of emotional polarity, participation, and respect; the offline evidence collection module performs local preprocessing and desensitization of the original signal through edge computing nodes, and only uploads feature vectors and metadata to the cloud to reduce bandwidth consumption and protect privacy. The document details the specific content and processing methods for offline evidence collection, covering four major scenarios: classroom, practical activities, situational tasks, and peer assessment. It captures student states from multiple dimensions, including cognition, emotion, and behavior. Edge computing preprocessing and anonymization improve data security while reducing cloud computing load. The extraction and analysis of various features make offline evidence more refined, providing high-quality input for subsequent integration and evaluation.

[0013] Furthermore, the online behavior tracking module includes: a learning behavior sequence unit, used to record video viewing completion rate, chapter jump path, number of quiz attempts, posting and replying sequence and citation depth in discussion forums, and analyze learning habits; a semantic text unit, used to perform word segmentation, entity recognition and topic modeling on discussion forums, open-ended quiz questions and reflection logs, generating value stance, factual basis and emotional polarity features; an interaction timing unit, used to record page dwell time, click hotspots, drag and connect operation sequences, generating cognitive load and exploratory behavior features; a task completion path unit, used to record node access order, failed retry and key operation paths in virtual simulation tasks, generating decision rationality and risk preference features; and a scenario selection recording unit, used to record role substitution selection, evidence acceptance order and weighing reasons in contextualized cases, generating evidence-driven value judgment features. The online behavior tracking module adopts an event tracing mechanism to ensure the integrity and traceability of the behavior sequence, providing a reliable foundation for subsequent analysis. The five units and their functions of online behavior tracking are defined in detail, enabling comprehensive capture of students' behavioral patterns and cognitive characteristics in online learning. The event tracing mechanism ensures the integrity and verifiability of the data, making online evidence highly credible and laying a solid foundation for cross-scenario integration and indicator calculation.

[0014] Furthermore, the evidence alignment and representation module includes: a timeline alignment unit, used to align students' offline and online events with a unified time reference, generating a time-sliced ​​event co-occurrence matrix; a cross-scene mapping unit, used to map offline and online evidence to a unified semantic space, defined by a concept ontology and knowledge graph, ensuring semantic comparability of evidence from different scenarios; and a consistent representation generation unit, used to concatenate offline evidence sub-vectors and online evidence sub-vectors, and adaptively adjust channel weights through a gated fusion network to output a consistent representation vector. The calculation formula for the gated fusion network is as follows: , This enables the organic integration of online and offline information. By aligning timelines and semantic mapping, the consistency problem of cross-scenario data is solved. The gated fusion network can dynamically adjust weights based on the quality of evidence, making the fusion results more representative and interpretable, and providing high-quality input vectors for subsequent fusion and calibration.

[0015] Furthermore, the dual-channel evidence fusion module includes a cross-channel attention unit, used to calculate the mutual attention weights between online and offline evidence to highlight complementary information and suppress redundant information, wherein the formula for calculating the mutual attention weights is as follows: , , Similarly, calculate and fusion of evidence vectors Depend on This allows for deep interaction and fusion of online and offline evidence. A cross-channel attention mechanism enables this deep interaction, automatically capturing the correlation and complementarity between evidence from different scenarios, resulting in more accurate fusion results and significantly improving the robustness and interpretability of the assessment.

[0016] Furthermore, the counterfactual calibration module includes: a propensity score estimation unit, used to fuse evidence vectors. Use as input to train the propensity score model ,in Indicates whether a student participated in a specific offline practical activity; the matching and weighting unit is used to perform nearest neighbor matching or kernel matching based on propensity scores, generate a counterfactual sample set, and calculate inverse probability weights. Dual robust estimation units are used to estimate the online evidence subvectors separately. Offline evidence subvectors To estimate the treatment effect for the treatment variable. and The weighted average was used as the effect estimate after bias correction. The loss function of its dual robust estimation is in , These are the true values ​​for online and offline metrics, respectively. , For the outcome regression model; the calibration evidence vector xcal is derived from... Received, among which To calibrate the intensity hyperparameter, It is the L2 norm. It is a symbolic function.

[0017] By constructing counterfactual samples through propensity score matching and dual robust estimation, the influence of platform bias and sample selection bias on the evaluation results is effectively eliminated. The design of the loss function ensures the robustness of the model in both propensity score prediction and outcome regression. The calibration formula can correct biases while preserving the original information, making the evaluation results closer to the truth and significantly improving the scientific rigor and credibility of the evaluation.

[0018] Furthermore, the indicator calculation and index generation module includes: knowledge mastery level. The calculation unit is used to calculate based on the accuracy of online tests, the quality of open-ended argumentation, and the accuracy of offline contextual tasks. in , , Weighting; Value recognition The computational unit is used to calculate the consistency of value stance, sufficiency of evidence citation, and peer review respect based on semantic text. in , , Weighting; Emotional investment The calculation unit is used to calculate based on the emotional polarity of online discussions, offline facial expressions and postures, and the reciprocal of the hesitation duration in the contextual task. in , , Weighting; behavioral compliance The calculation unit is used to calculate based on attendance rate in practical activities, contribution of task division, and consistency of context selection. in , , Weights; Internet literacy The calculation unit is used to calculate based on information verification, copyright awareness, privacy protection behavior, and compliance with discussion forum guidelines. in , , , Weights; Composite Index Computational unit for generating based on weighted aggregation in .

[0019] The system comprehensively covers cognitive, affective, behavioral, value-based, and digital literacy dimensions through five core indicators. Each indicator integrates evidence from multiple sources, both online and offline, avoiding the bias of a single evaluation method. The weighted aggregated comprehensive index can intuitively reflect the overall effectiveness of students' ideological and political education, providing a quantitative basis for subsequent profiling and intervention.

[0020] Furthermore, the profile construction and evolution module includes: a hierarchical profile construction unit, used to construct profiles using a comprehensive index. As the root node, based on knowledge mastery Value recognition Emotional investment Behavioral Practice With digital literacy As a primary indicator, a hierarchical profile is constructed using each indicator's sub-dimension as a leaf node; the temporal difference update unit is used to calculate the indicator increment based on a sliding time window. And update the profile according to the index weighting. in Forgetting factor, The system provides a profile vector for the current time window; a profile interpretation and generation unit, based on contribution analysis and rule extraction, generates an interpretable profile report, including strengths, weaknesses, and intervention priorities; and a profile comparison unit supports longitudinal comparison of individuals, horizontal comparison of groups, and benchmarking against similar courses, outputting differences and significance test results. The hierarchical profile structure clearly demonstrates students' performance in each dimension, the time-series difference update mechanism allows the profile to dynamically reflect students' development and changes, and the interpretation generation and comparison functions provide clear targets for teaching improvement, contributing to personalized education and precise intervention.

[0021] Furthermore, the intervention and feedback module includes: a personalized learning intervention generation unit, used to generate personalized learning paths based on profile shortcomings and index levels, including online micro-lessons, offline situational tasks, peer support groups, and reflective writing topics; a teaching improvement suggestion generation unit, used to generate teaching improvement suggestions based on the distribution of indicators at the classroom and course levels, including content restructuring, activity design optimization, and interaction strategy adjustment; and an early warning and referral unit, used when an individual or group falls short in value identification level VI and behavioral practice level. or digital literacy When a sustained decline occurs, an early warning is triggered, and the individual is referred to a counselor or psychological counseling center. A feedback loop recording unit is used to write the intervention implementation and changes in effectiveness into a feedback log, serving as input for the next round of profile updates, thus forming a closed loop for assessment and teaching improvement. This achieves a direct link between assessment results and teaching intervention; personalized pathways and teaching suggestions enhance the pertinence and effectiveness of education; the early warning and referral mechanism can promptly address potential risks; and the feedback loop ensures continuous tracking and optimization of intervention effects.

[0022] Furthermore, the privacy compliance and security module includes: a minimum collection unit, used to collect only the necessary data directly related to the evaluation target, and to perform feature extraction and desensitization at the collection end; and a differential privacy noise addition unit, used to add Laplacian or Gaussian noise to sensitive features during the aggregation statistics and model training stages, to meet the requirements of privacy. - Differential privacy; Tiered authorization and auditing unit, used to set tiered data access permissions for students, teachers, administrators, and third-party institutions, and record end-to-end access and operation audit logs; Secure multi-party computation and federated learning unit, used in cross-school or cross-platform evaluation scenarios to achieve data "usable but not visible" through secure multi-party computation or federated learning, with its federated learning objective function being: in For the number of participants, For the first Square sample size The total sample size is... For the first Local experience loss of the party, These are global model parameters. For local model parameters, This is a regularization coefficient; the data lifecycle management unit is used to implement data retention periods, periodic cleanup, and revocable consent mechanisms to ensure data compliance and minimize exposure. Through multi-layered privacy protection measures, student data security is protected to the maximum extent while ensuring assessment accuracy. Differential privacy and federated learning mechanisms are particularly critical in cross-institutional assessments. Tiered authorization and auditing ensure the controllability and traceability of data use. Data lifecycle management complies with legal and regulatory requirements, enhancing the system's compliance and social acceptability.

[0023] This invention provides an online and offline integrated student ideological and political education effectiveness evaluation system, which has the following beneficial effects: The online and offline integrated student ideological and political education effectiveness evaluation system proposed in this invention significantly outperforms existing technologies in multiple dimensions through the integration of multi-module collaboration and advanced algorithms. The specific advantages are as follows.

[0024] First, this system achieves comprehensive coverage of multimodal evidence both online and offline during the data collection phase. The offline evidence collection module records classroom interaction sequences, practical activity participation trajectories, situational task decision-making paths, peer evaluation texts, and facial expression / posture micro-expression features in physical classrooms, practical activities, and campus settings, ensuring that information at the cognitive, emotional, and behavioral levels is included in the evaluation scope. The online behavior tracking module collects learning behavior sequences, semantic text, interaction timing, task completion paths, and situational selection records in online courses, discussion forums, quizzes, and virtual simulation tasks, enabling a fine-grained portrayal of students' online learning status. This dual-channel, full-scenario data collection method fundamentally overcomes the limitations of traditional methods that rely on only a single data source, making the evaluation results more comprehensive and multi-dimensional.

[0025] Second, this system precisely aligns offline and online evidence using student identifiers and timestamps through an evidence alignment and representation module. It then maps these evidence to a unified semantic space using cross-scenario mapping, ultimately generating a consistent representation vector. This vector is formed by concatenating offline and online evidence sub-vectors, and the channel weights are adaptively adjusted through a gated fusion network. This process highlights complementary information and suppresses redundant information. This mechanism effectively solves the problems of semantic inconsistency and temporal asynchrony between online and offline data, laying a solid foundation for subsequent fusion and analysis.

[0026] Third, this system introduces a cross-channel attention mechanism in the dual-channel evidence fusion module to calculate the mutual attention weights between online and offline evidence, and generate a fused evidence vector based on attention weighting. This attention-based fusion method can not only capture the correlation between evidence in different scenarios, but also dynamically adjust the fusion strategy according to the context, making the fusion results more targeted and interpretable, and significantly improving the accuracy and robustness of the evaluation.

[0027] Fourth, in the counterfactual calibration module, this system employs propensity score matching and dual robust estimation methods to construct counterfactual samples and debias-calibrate the fused evidence vector. The probability of students participating in specific offline practical activities is estimated using a propensity score model, and calibration weights are generated based on inverse probability weighting. Parameter optimization is then performed using the loss function of dual robust estimation, effectively reducing the impact of platform bias and sample selection bias on the evaluation results. The calibrated evidence vector more realistically reflects the effectiveness of students' ideological and political education, making the evaluation results more scientific and credible.

[0028] Fifth, this system defines five core indicators in its indicator calculation and index generation module: knowledge mastery, value identification, emotional engagement, behavioral practice, and digital literacy. A comprehensive index is generated based on weighted aggregation. The calculation of each indicator integrates multi-source evidence from both online and offline sources. For example, knowledge mastery combines test accuracy, the quality of open-ended argumentation, and the accuracy of contextual tasks; value identification combines semantic textual consistency, the sufficiency of evidence citation, and peer review respect. This multi-dimensional, multi-source integrated indicator design comprehensively reflects students' overall performance in ideological and political education, avoiding the one-sidedness of a single indicator.

[0029] Sixth, this system constructs a hierarchical profile within its profile building and evolution module, using a comprehensive index as the root node, sub-indicators as first-level nodes, and sub-dimensions as leaf nodes, forming a clearly structured profile system. Simultaneously, based on a sliding time window and temporal differential update mechanism, the profile can dynamically evolve with the teaching cycle and student development, supporting individual longitudinal comparisons, group horizontal comparisons, and benchmarking against similar courses. This dynamic profile mechanism not only helps identify students' strengths and weaknesses but also provides precise targets for teaching improvement.

[0030] Seventh, this system achieves a closed-loop connection between assessment and teaching in the intervention and feedback module. Based on the profile's shortcomings and index levels, the system can automatically generate personalized learning paths, including online micro-lessons, offline situational tasks, peer support groups, and reflective writing topics; it also generates teaching improvement suggestions, covering content restructuring, activity design optimization, and interaction strategy adjustments. For individuals or groups whose value identification, behavioral practice, or digital literacy continues to decline, the system will trigger an alert and refer them to counselors or psychological counseling centers to ensure timely risk intervention. The feedback closed-loop recording unit writes the intervention implementation and changes in effect into a log, providing a basis for the next round of profile updates, thereby achieving continuous improvement.

[0031] Eighth, this system employs multiple security safeguards in its privacy compliance and security modules. The minimized data collection unit ensures that only necessary data directly related to the evaluation objective is collected, and performs feature extraction and anonymization at the collection end; the differential privacy noise-adding unit adds noise to sensitive features during aggregation statistics and model training to meet differential privacy requirements; the hierarchical authorization and auditing unit sets data access permissions for different roles and records full-link operation logs; the secure multi-party computation and federated learning unit achieves "usable but not visible" data in cross-school or cross-platform evaluations, avoiding privacy risks associated with centralized raw datasets; and the data lifecycle management unit implements data retention periods, periodic cleanup, and revocable consent mechanisms to ensure data compliance and minimized exposure.

[0032] Ninth, this system boasts advantages in computational efficiency and scalability. Each module employs a parallelizable design; for example, evidence collection can be preprocessed locally on edge computing nodes, fusion and calibration can be performed in batches on the cloud, and profile updates can utilize incremental computation, thus meeting the real-time assessment needs of large-scale student groups. Furthermore, the system supports flexible configuration across semesters, courses, and groups, allowing for adjustments to indicator weights and profile structures based on different academic levels and professional scenarios, demonstrating excellent versatility and scalability.

[0033] Tenth, this system excels in industrial applicability and licensability. Its technical solution is innovative in data acquisition, fusion, calibration, profiling, and intervention, especially the combination of counterfactual calibration and dual-channel attention fusion, which is not found in existing technologies, effectively circumventing existing technical paths and possessing a high probability of patent licensing. Furthermore, the system's effectiveness has been verified in university courses, with significant improvements in comprehensive indices and related indicators, demonstrating its feasibility and value in real-world teaching environments.

[0034] In summary, this system, through a complete chain of multi-source data collection, cross-scenario consistent representation, bias correction and calibration, dynamic profiling, and closed-loop intervention, significantly outperforms existing technologies in terms of comprehensiveness, accuracy, robustness, dynamism, and security. It can provide a scientific, reliable, and operable assessment solution for online and offline integrated ideological and political education. Attached Figure Description

[0035] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings in the following description are merely exemplary, and those skilled in the art can derive other embodiments based on the provided drawings without creative effort.

[0036] Figure 1 This is the overall operational logic diagram of the system of the present invention; Figure 2 This is an internal logic diagram of the offline evidence collection module of the present invention; Figure 3 This is the internal logic diagram of the online behavior tracking module of the present invention; Figure 4 This is the internal logic diagram of the evidence alignment and characterization module of the present invention; Figure 5 This is the internal logic diagram of the dual-channel evidence fusion module of the present invention. Detailed Implementation

[0037] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numerals in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this disclosure. Rather, they are merely examples of apparatuses consistent with some aspects of this disclosure as detailed in the appended claims.

[0038] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0039] How to use: The offline evidence collection module is used in physical classrooms, practical activities, and campus settings. It collects student speaking frequency, question types, answer depth, group collaboration role switching, and whiteboard / screen projection interaction trajectories through the classroom interaction event recording unit. The practical activity participation trajectory unit records student check-in paths, dwell time, task assignments, and peer collaboration maps during volunteer service, red tourism study tours, and social surveys. The situational task decision-making path unit records student option sequences, hesitation durations, reasoning texts, and backtracking modification records in situational tasks such as value clarification, moral dilemmas, and professional ethics. The peer review text and facial expression / micro-expression feature unit performs semantic analysis on the peer review text and extracts facial expression / micro-expression features to generate emotional polarity, participation, and respect feature vectors. All raw signals are preprocessed and anonymized locally on edge computing nodes; only feature vectors and metadata are uploaded to the cloud.

[0040] The online behavior tracking module is used in online courses, discussion forums, quizzes, and virtual simulation tasks. It records video completion rates, chapter navigation paths, quiz attempts, and the timing and depth of posts and replies in discussion forums through the behavior sequence unit. The semantic text unit performs word segmentation, entity recognition, and topic modeling on discussion forums, open-ended quiz questions, and reflection logs to generate value stances, factual evidence, and emotional polarity features. The interaction timing unit records page dwell time, click hotspots, drag-and-drop, and connection operation sequences to generate cognitive load and exploratory behavior features. The task completion path unit records the node access order, failed retry attempts, and key operation paths in virtual simulation tasks to generate decision rationality and risk preference features. The scenario selection recording unit records role-playing choices, evidence acceptance order, and weighing reasons in contextualized cases to generate evidence-driven value judgment features. The tracking process employs an event tracing mechanism to ensure the integrity and traceability of the behavior sequences.

[0041] The evidence alignment and representation module is used as follows: After data collection, the timeline alignment unit is activated to align offline and online events using a unified time benchmark, generating a time-sliced ​​event co-occurrence matrix. The cross-scene mapping unit maps offline and online evidence to a unified semantic space defined by the concept ontology and knowledge graph. The consistent representation generation unit concatenates offline and online evidence sub-vectors and adaptively adjusts channel weights through a gated fusion network to output a consistent representation vector, achieving the organic integration of online and offline information.

[0042] The use of the dual-channel evidence fusion module: After obtaining the consistent representation vector, the cross-channel attention unit is enabled to calculate the mutual attention weight between online and offline evidence, highlight complementary information and suppress redundant information, generate a fused evidence vector, and realize the deep interaction and fusion of online and offline evidence.

[0043] The counterfactual calibration module works as follows: The fused evidence vector is input into the propensity score estimation unit to train the propensity score model and estimate the probability of students participating in specific offline activities. The matching and weighting unit performs nearest neighbor matching or kernel matching based on the propensity score, generates a counterfactual sample set, and calculates inverse probability weights. The dual robust estimation unit uses online and offline evidence sub-vectors as processing variables respectively, estimates the processing effects, and takes a weighted average to obtain a bias-calibrated effect estimate. This calibrates the fused evidence vector and outputs a calibrated evidence vector.

[0044] The use of the indicator calculation and index generation module: Based on the calibration evidence vector, the module calculates the degree of knowledge mastery, value recognition, emotional investment, behavioral practice and digital literacy, and generates a comprehensive index of ideological and political education effectiveness based on weighted aggregation, forming a multi-dimensional quantitative evaluation result.

[0045] The use of the profile construction and evolution module: Using the comprehensive index and sub-indicators as nodes, a hierarchical profile is constructed. The index increment is calculated based on the sliding time window, and the profile is updated by index weighting to generate an interpretable report. It supports vertical comparison of individuals, horizontal comparison of groups, and benchmarking against similar courses, reflecting the dynamic changes in students' ideological and political education performance.

[0046] Use of the intervention and feedback module: Based on the profile and index, personalized learning paths and teaching improvement suggestions are generated. Early warnings are triggered for a continuous decline in value recognition, behavioral practice or digital literacy and referrals are made to counselors or psychological counseling centers. The implementation of intervention and changes in effect are recorded in the feedback log as input for the next round of profile updates, forming a closed loop of assessment and teaching improvement.

[0047] Privacy compliance and security module usage: During the collection and transmission process, minimized collection and feature extraction anonymization are enabled; during the aggregation statistics and model training stages, noise is added to sensitive features to meet differential privacy requirements; hierarchical data access permissions are set and full-link audit logs are recorded; when conducting cross-school or cross-platform evaluations, secure multi-party computation or federated learning is used to achieve data usability without visibility; data retention periods, regular cleanup, and revocable consent mechanisms are implemented to ensure data compliance and minimized exposure.

[0048] The system control and scheduling module is used to adaptively schedule the evidence collection frequency, fusion window, calibration batches and profile update cycle to ensure the real-time performance and stability of the system under a large student population, and to dynamically adjust the operating parameters according to the teaching cycle and assessment requirements.

[0049] Example: Example 1: Blended Teaching Scenario for College Ideological and Political Theory Courses Usage Environment Designed for undergraduate ideological and political theory courses, this program covers physical classrooms, red tourism research and social practice, extracurricular activities, and online course platforms. Classrooms are equipped with multimodal data acquisition devices, and practical sessions include attendance and task milestones. The online platform hosts video courses, discussion forums, quizzes, and virtual simulation tasks. Edge computing nodes are deployed in teaching buildings and training centers, while an evaluation engine and user profile database are built in the cloud. A campus data platform provides unified identity and access control.

[0050] Implementation steps The offline evidence collection module was activated: In the classroom, it recorded speaking frequency, question types, answer depth, and group collaboration and blackboard interaction patterns; in red-themed study tours and practical activities, it collected sign-in paths, dwell time, task division, and peer collaboration maps; in situational tasks, it recorded option sequences, hesitation duration, reasoning text, and retrospective modifications; and in peer review sessions, it collected peer review texts and facial expression / posture micro-expression features. Edge nodes underwent local preprocessing and anonymization before being uploaded.

[0051] Activate the online behavior tracking module: record video completion rate, chapter jump path, number of quiz attempts, and the posting and replying sequence and citation depth in the discussion forum; perform word segmentation, entity recognition, and topic modeling on the discussion forum, open-ended questions, and reflection logs; collect page dwell time, click hotspots, drag and connect operation sequences; record node access order, failed retry attempts, and key operation paths in virtual simulation tasks; and record role-playing selection, evidence acceptance order, and weighing reasons in contextualized case learning.

[0052] Run the evidence alignment and representation module: align offline and online events on a unified time base to generate an event co-occurrence matrix; map offline and online evidence to a unified semantic space; splice offline evidence sub-vectors and online evidence sub-vectors, adaptively adjust channel weights through a gating fusion network, and output a consistent representation vector.

[0053] Run the dual-channel evidence fusion module: calculate the mutual attention weights between online and offline evidence, highlight complementary information and suppress redundant information, and generate a fused evidence vector.

[0054] Run the counterfactual calibration module: train the propensity score model to estimate the probability of students participating in specific offline practical activities; match based on propensity scores and generate inverse probability weighted weights; perform dual robust estimation with online evidence sub-vectors and offline evidence sub-vectors as processing variables respectively to obtain the effect estimate after bias calibration, and calibrate the fused evidence vector.

[0055] The module for calculating operational indicators and generating indices comprehensively calibrates evidence vectors to calculate knowledge mastery, value recognition, emotional investment, behavioral practice, and online literacy, and then aggregates these metrics to generate a comprehensive index of the effectiveness of ideological and political education.

[0056] The profile building and evolution module constructs hierarchical profiles using comprehensive indices and sub-indices as nodes. It calculates indicator increments based on sliding time windows and updates profiles by index weighting, generating interpretable reports and supporting vertical comparison of individuals, horizontal comparison of groups, and benchmarking against similar courses.

[0057] The intervention and feedback module generates personalized learning paths and teaching improvement suggestions based on the profile's shortcomings and index levels; it triggers early warnings and referrals for individuals or groups whose value recognition, behavioral practice, or digital literacy continues to decline; and it records the implementation of interventions and changes in effectiveness in the feedback log to form a closed loop.

[0058] The privacy compliance and security module is implemented as follows: it performs minimal data collection and feature desensitization; it adds noise during the aggregation statistics and model training phases to meet differential privacy; it sets up hierarchical authorization and end-to-end auditing; it enables secure multi-party computation or federated learning during cross-school or cross-platform evaluations; and it implements data retention periods, periodic cleanup, and revocable consent mechanisms.

[0059] The system control and scheduling module adaptively schedules the evidence collection frequency, fusion window, calibration batches, and profile update cycle to ensure real-time performance and stability under large-scale student groups.

[0060] Expected results In blended learning, we achieve a panoramic portrayal of cognitive, emotional, behavioral, and value dimensions, deeply integrate online and offline evidence, significantly reduce assessment bias, make teaching improvement suggestions more targeted, provide timely early warning and referral responses, and ensure that the entire data process is compliant and controllable.

[0061] Example 2: Online Ideological and Political Education Scenario in Higher Vocational Colleges Usage Environment This system targets the online ideological and political education system of higher vocational colleges, covering a three-stage teaching approach: "small theoretical classroom + large social scenario + online cloud interaction," an on-campus new media matrix, and off-campus red tourism sites and enterprise practice bases. Data collection devices are deployed in classrooms and training rooms, behavioral points are embedded in the campus portal and new media platforms, and mobile data collection terminals and task nodes are configured for off-campus practice. A cloud-based evaluation engine and data platform work together to support consistent evaluation across different scenarios.

[0062] Implementation steps The offline evidence collection module was activated: in classroom interactions, the frequency of speaking, the type of questions asked, the depth of answers, and the interaction trajectory of group collaboration and blackboard writing were recorded; in large-scale social practice scenarios, the check-in path, the duration of stay, the division of tasks, and the peer collaboration map were collected; in situational tasks, the sequence of options, the duration of hesitation, the reason text, and the retrospective modification were recorded; and in the peer evaluation process, the peer evaluation text and facial expression and micro-expression features were collected.

[0063] The online behavior tracking module is activated to record video course completion rates, chapter navigation paths, quiz attempts, and the timing and citation depth of posts and replies in discussion forums; it performs word segmentation, entity recognition, and topic modeling on discussion forums, open-ended questions, and reflection logs; it collects page dwell time, click hotspots, drag-and-drop and connection operation sequences; it records node access order, failed retry attempts, and key operation paths in virtual simulation tasks; and it records role selection, evidence acceptance order, and justifications in contextualized case learning.

[0064] Run the evidence alignment and representation module: align offline and online events on a unified time base to generate an event co-occurrence matrix; map offline and online evidence to a unified semantic space; splice offline evidence sub-vectors and online evidence sub-vectors, adaptively adjust channel weights through a gating fusion network, and output a consistent representation vector.

[0065] Run the dual-channel evidence fusion module: calculate the mutual attention weights between online and offline evidence, highlight complementary information and suppress redundant information, and generate a fused evidence vector.

[0066] Run the counterfactual calibration module: train the propensity score model to estimate the probability of students participating in specific offline practical activities; match based on propensity scores and generate inverse probability weighted weights; perform dual robust estimation with online evidence sub-vectors and offline evidence sub-vectors as processing variables respectively to obtain the effect estimate after bias calibration, and calibrate the fused evidence vector.

[0067] The module for calculating operational indicators and generating indices comprehensively calibrates evidence vectors to calculate knowledge mastery, value recognition, emotional investment, behavioral practice, and online literacy, and then aggregates these metrics to generate a comprehensive index of the effectiveness of ideological and political education.

[0068] The profile building and evolution module constructs hierarchical profiles using comprehensive indices and sub-indices as nodes. It calculates indicator increments based on sliding time windows and updates profiles by index weighting, generating interpretable reports and supporting vertical comparison of individuals, horizontal comparison of groups, and benchmarking against similar courses.

[0069] The intervention and feedback module generates personalized learning paths and teaching improvement suggestions based on the profile's shortcomings and index levels; it triggers early warnings and referrals for individuals or groups whose value recognition, behavioral practice, or digital literacy continues to decline; and it records the implementation of interventions and changes in effectiveness in the feedback log to form a closed loop.

[0070] The privacy compliance and security module is implemented as follows: it performs minimal data collection and feature desensitization; it adds noise during the aggregation statistics and model training phases to meet differential privacy; it sets up hierarchical authorization and end-to-end auditing; it enables secure multi-party computation or federated learning during cross-school or cross-platform evaluations; and it implements data retention periods, periodic cleanup, and revocable consent mechanisms.

[0071] The system control and scheduling module adaptively schedules the evidence collection frequency, fusion window, calibration batches, and profile update cycle to ensure real-time performance and stability under large-scale student groups.

[0072] Expected results It supports full-chain assessment of the "three-stage" teaching approach, connects the three spaces of classroom, society and the internet, promotes precise resource supply and personalized growth support, and enhances the attractiveness and effectiveness of online ideological and political education.

[0073] Example 3: Digital Ideological and Political Education Scenario in a One-Stop Student Community in Universities Usage Environment Aimed at providing a "one-stop" student community, it covers dormitory areas, Party and Youth League building, psychological and employment services, extracurricular activities, social practice, and an online service hall. The student community deploys access control and sensing devices, while online portals and lightweight applications support services throughout the entire learning and living process. A data platform integrates systems such as student affairs, academic affairs, logistics, access control, finance, and the library, with assessment engines and user profile databases deployed to the community level, supporting real-time analysis and feedback.

[0074] Implementation steps The offline evidence collection module was activated: in classroom interactions, the frequency of speaking, the type of questions asked, the depth of answers, and the interaction trajectory of group collaboration and blackboard writing were recorded; in community volunteer service and social practice, the check-in path, the duration of stay, the division of tasks, and the peer collaboration map were collected; in situational tasks, the sequence of options, the duration of hesitation, the reason text, and the retrospective modification were recorded; and in the peer evaluation process, the peer evaluation text and facial expression and micro-expression features were collected.

[0075] The online behavior tracking module is activated to record video course completion rates, chapter navigation paths, quiz attempts, and the timing and citation depth of posts and replies in discussion forums; it performs word segmentation, entity recognition, and topic modeling on discussion forums, open-ended questions, and reflection logs; it collects page dwell time, click hotspots, drag-and-drop and connection operation sequences; it records node access order, failed retry attempts, and key operation paths in virtual simulation tasks; and it records role selection, evidence acceptance order, and justifications in contextualized case learning.

[0076] Run the evidence alignment and representation module: align offline and online events on a unified time base to generate an event co-occurrence matrix; map offline and online evidence to a unified semantic space; splice offline evidence sub-vectors and online evidence sub-vectors, adaptively adjust channel weights through a gating fusion network, and output a consistent representation vector.

[0077] Run the dual-channel evidence fusion module: calculate the mutual attention weights between online and offline evidence, highlight complementary information and suppress redundant information, and generate a fused evidence vector.

[0078] Run the counterfactual calibration module: train the propensity score model to estimate the probability of students participating in specific offline practical activities; match based on propensity scores and generate inverse probability weighted weights; perform dual robust estimation with online evidence sub-vectors and offline evidence sub-vectors as processing variables respectively to obtain the effect estimate after bias calibration, and calibrate the fused evidence vector.

[0079] The module for calculating operational indicators and generating indices comprehensively calibrates evidence vectors to calculate knowledge mastery, value recognition, emotional investment, behavioral practice, and online literacy, and then aggregates these metrics to generate a comprehensive index of the effectiveness of ideological and political education.

[0080] The profile building and evolution module constructs hierarchical profiles using comprehensive indices and sub-indices as nodes. It calculates indicator increments based on sliding time windows and updates profiles by index weighting, generating interpretable reports and supporting vertical comparison of individuals, horizontal comparison of groups, and benchmarking against similar courses.

[0081] The intervention and feedback module generates personalized learning paths and teaching improvement suggestions based on the profile's shortcomings and index levels; it triggers early warnings and referrals for individuals or groups whose value recognition, behavioral practice, or digital literacy continues to decline; and it records the implementation of interventions and changes in effectiveness in the feedback log to form a closed loop.

[0082] The privacy compliance and security module is implemented as follows: it performs minimal data collection and feature desensitization; it adds noise during the aggregation statistics and model training phases to meet differential privacy; it sets up hierarchical authorization and end-to-end auditing; it enables secure multi-party computation or federated learning during cross-school or cross-platform evaluations; and it implements data retention periods, periodic cleanup, and revocable consent mechanisms.

[0083] The system control and scheduling module adaptively schedules the evidence collection frequency, fusion window, calibration batches, and profile update cycle to ensure real-time performance and stability under large-scale student groups.

[0084] Expected results By deeply embedding ideological and political assessment into the daily life of student communities, and by connecting management and service data, a synergistic effect of "education through management, education through service, and education through culture" can be formed, supporting the continuous operation of accurate identification, accurate service, and accurate evaluation.

[0085] Example 4: Scenario of the Big Data Platform for Precision Ideological and Political Education in Higher Education Institutions Usage Environment This platform is designed for universities' "Smart Student Affairs + Integrated Educational Big Data" initiatives, covering multiple departmental business systems, course platforms, practical components, and new media matrix. It establishes a data middleware platform and evaluation engine, unifying identity management and access control, and supporting cross-system data aggregation and governance. Evaluation tasks are executed in batches and via designated windows, with user profiles and index results fed back to teaching and management staff.

[0086] Implementation steps The offline evidence collection module was launched: in classroom interactions, the frequency of speaking, the type of questions asked, the depth of answers, and the interaction trajectory of group collaboration and blackboard writing were recorded; in volunteer service, social surveys and industrial practice, the check-in path, the duration of stay, the division of tasks and the peer collaboration map were collected; in situational tasks, the sequence of options, the duration of hesitation, the reason text and retrospective modification were recorded; in the peer evaluation process, the peer evaluation text and facial expression and micro-expression features were collected.

[0087] The online behavior tracking module is activated to record video course completion rates, chapter navigation paths, quiz attempts, and the timing and citation depth of posts and replies in discussion forums; it performs word segmentation, entity recognition, and topic modeling on discussion forums, open-ended questions, and reflection logs; it collects page dwell time, click hotspots, drag-and-drop and connection operation sequences; it records node access order, failed retry attempts, and key operation paths in virtual simulation tasks; and it records role selection, evidence acceptance order, and justifications in contextualized case learning.

[0088] Run the evidence alignment and representation module: align offline and online events on a unified time base to generate an event co-occurrence matrix; map offline and online evidence to a unified semantic space; splice offline evidence sub-vectors and online evidence sub-vectors, adaptively adjust channel weights through a gating fusion network, and output a consistent representation vector.

[0089] Run the dual-channel evidence fusion module: calculate the mutual attention weights between online and offline evidence, highlight complementary information and suppress redundant information, and generate a fused evidence vector.

[0090] Run the counterfactual calibration module: train the propensity score model to estimate the probability of students participating in specific offline practical activities; match based on propensity scores and generate inverse probability weighted weights; perform dual robust estimation with online evidence sub-vectors and offline evidence sub-vectors as processing variables respectively to obtain the effect estimate after bias calibration, and calibrate the fused evidence vector.

[0091] The module for calculating operational indicators and generating indices comprehensively calibrates evidence vectors to calculate knowledge mastery, value recognition, emotional investment, behavioral practice, and online literacy, and then aggregates these metrics to generate a comprehensive index of the effectiveness of ideological and political education.

[0092] The profile building and evolution module constructs hierarchical profiles using comprehensive indices and sub-indices as nodes. It calculates indicator increments based on sliding time windows and updates profiles by index weighting, generating interpretable reports and supporting vertical comparison of individuals, horizontal comparison of groups, and benchmarking against similar courses.

[0093] The intervention and feedback module generates personalized learning paths and teaching improvement suggestions based on the profile's shortcomings and index levels; it triggers early warnings and referrals for individuals or groups whose value recognition, behavioral practice, or digital literacy continues to decline; and it records the implementation of interventions and changes in effectiveness in the feedback log to form a closed loop.

[0094] The privacy compliance and security module is implemented as follows: it performs minimal data collection and feature desensitization; it adds noise during the aggregation statistics and model training phases to meet differential privacy; it sets up hierarchical authorization and end-to-end auditing; it enables secure multi-party computation or federated learning during cross-school or cross-platform evaluations; and it implements data retention periods, periodic cleanup, and revocable consent mechanisms.

[0095] The system control and scheduling module adaptively schedules the evidence collection frequency, fusion window, calibration batches, and profile update cycle to ensure real-time performance and stability under large-scale student groups.

[0096] Expected results The "Smart Student Affairs" system enables comprehensive governance and intelligent analysis of ideological and political data, supporting a closed-loop process of precise identification, precise education, precise service, and precise evaluation, thereby improving governance efficiency and the quality of education.

[0097] Example 5: Digital Ideological and Political Education Convergence Media Communication and Teaching Scenarios in Higher Education Institutions Usage Environment Aimed at university digital ideological and political education media centers, this system covers course platforms, short video and live streaming platforms, virtual simulation laboratories, red resource databases, and a campus new media matrix. Classrooms and studios are equipped with data acquisition devices, online platforms are embedded with behavioral data points, and virtual simulation tasks are set with process nodes. The evaluation engine and content platform work in tandem to support integrated evaluation of dissemination effectiveness and educational outcomes.

[0098] Implementation steps The offline evidence collection module was activated: in classroom interactions, the frequency of speaking, the type of questions asked, the depth of answers, and the interaction trajectory of group collaboration and blackboard writing were recorded; in red study tours and themed practice activities, the sign-in path, the duration of stay, the division of tasks, and the peer collaboration map were collected; in situational tasks, the sequence of options, the duration of hesitation, the reason text, and the retrospective modification were recorded; in the peer evaluation process, the peer evaluation text and facial expression and micro-expression features were collected.

[0099] The online behavior tracking module is activated to record video course completion rates, chapter navigation paths, quiz attempts, and the timing and citation depth of posts and replies in discussion forums; it performs word segmentation, entity recognition, and topic modeling on discussion forums, open-ended questions, and reflection logs; it collects page dwell time, click hotspots, drag-and-drop and connection operation sequences; it records node access order, failed retry attempts, and key operation paths in virtual simulation tasks; and it records role selection, evidence acceptance order, and justifications in contextualized case learning.

[0100] Run the evidence alignment and representation module: align offline and online events on a unified time base to generate an event co-occurrence matrix; map offline and online evidence to a unified semantic space; splice offline evidence sub-vectors and online evidence sub-vectors, adaptively adjust channel weights through a gating fusion network, and output a consistent representation vector.

[0101] Run the dual-channel evidence fusion module: calculate the mutual attention weights between online and offline evidence, highlight complementary information and suppress redundant information, and generate a fused evidence vector.

[0102] Run the counterfactual calibration module: train the propensity score model to estimate the probability of students participating in specific offline practical activities; match based on propensity scores and generate inverse probability weighted weights; perform dual robust estimation with online evidence sub-vectors and offline evidence sub-vectors as processing variables respectively to obtain the effect estimate after bias calibration, and calibrate the fused evidence vector.

[0103] The module for calculating operational indicators and generating indices comprehensively calibrates evidence vectors to calculate knowledge mastery, value recognition, emotional investment, behavioral practice, and online literacy, and then aggregates these metrics to generate a comprehensive index of the effectiveness of ideological and political education.

[0104] The profile building and evolution module constructs hierarchical profiles using comprehensive indices and sub-indices as nodes. It calculates indicator increments based on sliding time windows and updates profiles by index weighting, generating interpretable reports and supporting vertical comparison of individuals, horizontal comparison of groups, and benchmarking against similar courses.

[0105] The intervention and feedback module generates personalized learning paths and teaching improvement suggestions based on the profile's shortcomings and index levels; it triggers early warnings and referrals for individuals or groups whose value recognition, behavioral practice, or digital literacy continues to decline; and it records the implementation of interventions and changes in effectiveness in the feedback log to form a closed loop.

[0106] The privacy compliance and security module is implemented as follows: it performs minimal data collection and feature desensitization; it adds noise during the aggregation statistics and model training phases to meet differential privacy; it sets up hierarchical authorization and end-to-end auditing; it enables secure multi-party computation or federated learning during cross-school or cross-platform evaluations; and it implements data retention periods, periodic cleanup, and revocable consent mechanisms.

[0107] The system control and scheduling module adaptively schedules the evidence collection frequency, fusion window, calibration batches, and profile update cycle to ensure real-time performance and stability under large-scale student groups.

[0108] Expected results By integrating curriculum dissemination with teaching practice, we can achieve collaborative evaluation of content supply, interactive participation, and value internalization, promote the secondary dissemination and in-depth transformation of high-quality resources, and enhance the attractiveness, appeal, and guiding power of ideological and political content.

[0109] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An online and offline integrated student ideological and political education effectiveness evaluation system, characterized in that: include: The offline evidence collection module is used to collect multimodal behavioral evidence of students in physical classrooms, practical activities and campus scenarios. The evidence includes classroom interaction event sequences, practical activity participation trajectories, situational task decision-making paths, peer evaluation texts and facial / postural micro-expression features. The online behavior tracking module is used to collect learning behavior sequences, semantic text, interaction timing, task completion paths, and context selection records in online courses, discussion forums, quizzes, and virtual simulation tasks. The evidence alignment and representation module is used to align the offline evidence and online evidence according to student identifiers and timestamps, and generate a consistent representation vector across scenarios. The representation vector is obtained by concatenating the offline evidence sub-vector and the online evidence sub-vector and then fusing them through gating. A dual-channel evidence fusion module is used to perform cross-channel weighted fusion of the consistent representation vector based on an attention mechanism and output a fused evidence vector. The counterfactual calibration module is used to construct counterfactual samples based on bias score matching and dual robust estimation, take the fused evidence vector as input, perform debias calibration on the fused evidence vector, and output the calibrated evidence vector. The indicator calculation and index generation module is used to calculate knowledge mastery, value recognition, emotional investment, behavioral practice and network literacy based on the calibration evidence vector, and generate a comprehensive index of ideological and political education effectiveness based on weighted aggregation. The profile construction and evolution module is used to construct a hierarchical profile using the comprehensive index and sub-indicators as nodes, and to dynamically evolve the profile by combining temporal difference. The intervention and feedback module is used to generate personalized learning intervention and teaching improvement suggestions based on the profile and index, and to write the feedback results back to the teaching process. The privacy compliance and security module is used to perform minimal data collection, differential privacy noise addition, hierarchical authorization, and end-to-end auditing on the collected and transmitted data. The system control and scheduling module is used to adaptively schedule the evidence collection frequency, fusion window, calibration batches, and profile update cycle.

2. The online and offline integrated student ideological and political education effectiveness evaluation system according to claim 1, characterized in that, The offline evidence collection module includes: The classroom interaction event recording unit is used to record students' speaking frequency, question types, answer depth, group collaboration role switching, and blackboard / screen projection interaction trajectory in class; The practical activity participation trajectory unit is used to record students' check-in path, stay duration, task division and peer collaboration map in volunteer service, red study tours, and social surveys; The contextual task decision path unit is used to record students' choice sequence, hesitation time, reason text, and backtracking modification records in contextual tasks such as value clarification, moral dilemmas, and professional ethics. Peer evaluation text and facial / postural micro-expression feature units are used to perform semantic parsing on peer evaluation text, extract features of facial expressions and postural micro-expressions during offline peer evaluation, and generate feature vectors of emotional polarity, participation and respect. The offline evidence collection module performs local preprocessing and desensitization of the original signal through edge computing nodes, and only uploads the feature vector and metadata to the cloud.

3. The online and offline integrated student ideological and political education effectiveness evaluation system according to claim 1, characterized in that, The online behavior tracking module includes: Learning behavior sequence units are used to record video viewing completion rate, chapter jump path, number of quiz attempts, and the timing and citation depth of posts / replies in the discussion forum; Semantic text units are used to perform word segmentation, entity recognition, and topic modeling on discussion forums, open-ended quizzes, and reflection logs to generate value stances, factual evidence, and sentiment polarity features. Interactive timing units are used to record page dwell time, click hotspots, drag and connect operation sequences, and generate cognitive load and exploratory behavior characteristics; The task completion path unit is used to record the node access order, failure retry points and key operation paths in the virtual simulation task, and to generate decision rationality and risk preference characteristics. The context selection recording unit is used to record the role selection, evidence acceptance order and weighing reasons in contextualized cases, and generate evidence-driven value judgment features. The online behavior tracking module employs an event tracing mechanism to ensure the integrity and traceability of the behavior sequence.

4. The online and offline integrated student ideological and political education effectiveness evaluation system according to claim 1, characterized in that, The evidence alignment and characterization module includes: The timeline alignment unit is used to align students' offline and online events with a unified time reference, generating a time-sliced ​​event co-occurrence matrix. A cross-scenario mapping unit is used to map offline evidence and online evidence to a unified semantic space, which is defined by a concept ontology and a knowledge graph. The uniform representation generation unit is used to concatenate offline evidence subvectors with online evidence subvectors, and adaptively adjust the channel weights through a gated fusion network to output a uniform representation vector. ; The gated fusion network is calculated in the following manner: in, For online evidence subvectors, For offline evidence sub-vectors, and For learnable parameters, For the Sigmoid function, This is an element-wise multiplication.

5. The online and offline integrated student ideological and political education effectiveness evaluation system according to claim 1, characterized in that, The dual-channel evidence fusion module includes: Cross-channel attention unit is used to calculate the mutual attention weights between online and offline evidence to highlight complementary information and suppress redundant information; The mutual attention weight and Calculated using the following formula: Fusion Evidence Vector We obtain it from the following formula: in, , , , , , For learnable parameters, For attention dimension, LayerNorm is for layer normalization.

6. The online and offline integrated student ideological and political education effectiveness evaluation system according to claim 1, characterized in that, The counterfactual calibration module includes: Propensity score estimation unit, used to fuse evidence vectors Use as input to train the propensity score model ,in Indicates whether a student participates in a specific offline practical activity; The matching and weighting unit is used to perform nearest neighbor matching or kernel matching based on propensity scores, generate a counterfactual sample set, and calculate inverse probability weights. ; Dual robust estimation units are used to estimate the online evidence subvectors separately. Offline evidence subvectors To estimate the treatment effect for the treatment variable. and The weighted average was used as the effect estimate after bias correction. ; The loss function of the dual robust estimation is defined as: in, and These are the true values ​​for online and offline metrics, respectively. and For the outcome regression model; Calibration Evidence Vector We obtain it from the following formula: in, To calibrate the intensity hyperparameter, It is the L2 norm. It is a symbolic function.

7. The online and offline integrated student ideological and political education effectiveness evaluation system according to claim 1, characterized in that, The indicator calculation and index generation module includes: Knowledge mastery The calculation unit is used to calculate knowledge mastery based on online test accuracy, open-ended question argument quality, and offline contextual task accuracy. : in, , , As weight, To test accuracy, For the quality of the open-ended argument, For contextual task accuracy; Value recognition The computational unit is used to calculate value consensus based on the consistency of value stance, sufficiency of evidence citation, and peer review respect in semantic text. : in, , , As weight; Emotional investment The calculation unit is used to calculate emotional engagement based on the emotional polarity of online discussions, offline facial expressions / posture participation, and the reciprocal of the hesitation time in the contextual task. : in, , , As weight, Standardize the duration of hesitation; The Behavioral Performance BP (BP) calculation unit is used to calculate the Behavioral Performance BP based on attendance rate in practical activities, contribution of task allocation, and consistency of context selection. in, , , As weight; Digital literacy The calculation unit is used to calculate online literacy scores based on information discernment, copyright awareness, privacy protection behavior, and compliance with forum guidelines. : in, , , , As weight; Composite Index The calculation unit is used to generate a comprehensive index of the effectiveness of ideological and political education based on weighted aggregation. in, , , , , Let be the weight, and satisfy... .

8. The online and offline integrated student ideological and political education effectiveness evaluation system according to claim 1, characterized in that, The portrait construction and evolution module includes: Hierarchical profile building units are used to construct profiles based on comprehensive indices. As the root node, based on knowledge mastery Value recognition Emotional investment Behavioral Practice With digital literacy As a primary indicator, each indicator's sub-dimension is used as a leaf node to construct a hierarchical profile; The time-series differential update unit is used to calculate the increment of the index based on the sliding time window. And update the profile according to the index weighting: in, Forgetting factor, Image vector for the current time window; The profile interpretation and generation unit is used to generate an interpretable profile report based on contribution analysis and rule extraction. The report includes strengths, weaknesses and intervention priorities. The profile comparison unit is used to support longitudinal comparison of individuals, horizontal comparison of groups, and benchmarking against similar courses, and output the differences and significance test results.

9. The online and offline integrated student ideological and political education effectiveness evaluation system according to claim 1, characterized in that, The intervention and feedback module includes: The personalized learning intervention generation unit is used to generate personalized learning paths based on the profile's shortcomings and index level. The paths include online micro-lessons, offline situational tasks, peer support groups, and reflective writing topics. The teaching improvement suggestion generation unit is used to generate teaching improvement suggestions based on the distribution of indicators at the classroom and course levels. The suggestions include content restructuring, activity design optimization, and interaction strategy adjustment. The early warning and referral unit is used when an individual or group has a certain level of value consensus. Behavioral Practice or digital literacy If a sustained decline occurs, an alert will be triggered and the individual will be referred to a counselor or mental health center. The feedback loop recording unit is used to write the intervention implementation and changes in effect into the feedback log, which serves as input for the next round of profile updates.

10. The online and offline integrated student ideological and political education effectiveness evaluation system according to claim 1, characterized in that, The privacy compliance and security module includes: Minimize the acquisition unit to collect only the necessary data directly related to the evaluation target, and perform feature extraction and desensitization at the acquisition end; Differential privacy noise-adding units are used to add Laplacian or Gaussian noise to sensitive features during the aggregation statistics and model training phases, satisfying the following requirements: -Differential privacy; The tiered authorization and auditing unit is used to set tiered data access permissions for students, teachers, administrators, and third-party organizations, and to record full-link access and operation audit logs. The secure multi-party computation and federated learning unit is used to achieve "usable but invisible" data in cross-institutional or cross-platform evaluation scenarios by employing secure multi-party computation or federated learning. The objective function of the federated learning is defined as follows: in, For the number of participants, For the first Square sample size The total sample size is... For the first Local experience loss of the party, These are global model parameters. For local model parameters, The regularization coefficient is used. The data lifecycle management unit is used to implement data retention periods, periodic cleanup, and revocable consent mechanisms to ensure data compliance and minimize exposure.

Images