Digital security and protection system trust regulation and control method and device based on emotion calculation

Abstract

The invention relates to a digital security and protection system trust regulation and control method and device based on emotion calculation. According to the method, multi-modal data such as videos, voices and behaviors are collected and subjected to desensitization and encryption processing, emotion vectors are generated through an emotion recognition model, risk perception indexes are calculated in combination with privacy risks, bias risks and control risks, a trust prediction model is constructed to obtain trust willingness of a user, and when the trust willingness is lower than a preset threshold value, the trust of the user is predicted. And the system automatically executes algorithm transparency improvement and risk communication feedback, and iteratively updates model parameters through a self-learning mechanism, thereby realizing dynamic regulation and self-adaptive optimization of user trust in the digital security and protection system. The application has the effect of improving the security, credibility and user acceptability of the system.

Description

Technical Field

[0001] This application relates to the field of human-computer trust control technology in artificial intelligence, affective computing and digital security systems, and in particular to a trust control method and device for digital security systems based on affective computing. Background Technology

[0002] With the rapid advancement of smart cities, intelligent transportation, and public safety systems, digital security systems have gradually evolved from traditional passive monitoring to proactive perception and intelligent decision-making systems centered on artificial intelligence. Security technologies based on computer vision, speech recognition, and behavioral analysis have made significant progress in areas such as anomaly identification, risk warning, and situational awareness, greatly improving urban operational efficiency and public safety levels. However, while technological capabilities continue to improve, structural problems such as insufficient technological credibility and low public acceptance have gradually emerged in the actual deployment and application of digital security systems.

[0003] On the one hand, the design focus of existing intelligent security systems is mainly on "accurate identification and timely response," emphasizing the optimization of objective security indicators, while generally neglecting the subjective emotional state and psychological feelings of the perceived subjects—the public or users. In an environment of high-density cameras, continuous data collection, and automated algorithmic decision-making, users are prone to feelings of anxiety about being constantly monitored, having their privacy violated, and feeling that the algorithms are uncontrollable. This anxiety does not stem from a single technological flaw, but rather from a combination of factors such as information asymmetry, the uninterpretability of algorithms, and an imbalance in risk perception, directly affecting the public's trust in digital security systems.

[0004] On the other hand, existing digital security systems, in terms of risk management, mostly focus on technical compliance and data security protection, such as reducing the risk of privacy leaks through access control, data encryption, and hierarchical permissions. However, these methods primarily target the objective risks themselves and do not consider the differences in the public's subjective perception of risk. Extensive practice shows that even if a system meets security and compliance requirements at the technical level, users may still lower their trust levels or even exhibit resistance due to concerns about algorithmic bias, opaque decision-making, or unclear data usage, thereby weakening the system's social acceptance and long-term effectiveness.

[0005] Furthermore, existing research on security trust primarily focuses on institutional rules, ethical principles, or static evaluation indicators, typically employing post-event audits or manual intervention for trust management. These methods struggle to address the dynamic changes in user emotions and risk perceptions in real-world application scenarios, failing to achieve real-time perception, immediate feedback, and continuous optimization during system operation. Especially in complex public spaces, where user groups exhibit high heterogeneity, different individuals display significantly different emotional responses and trust levels when faced with the same security strategy, making it difficult for traditional rule-based management methods to effectively differentiate and finely control these differences.

[0006] In the field of artificial intelligence, the rapid development of affective computing technology has provided a new technical path for solving the aforementioned problems. By analyzing multimodal information such as facial expressions, voice features, and behavioral patterns, affective computing can accurately depict users' emotional states and psychological reactions. However, existing affective computing research is mostly applied to areas such as human-computer interaction, intelligent customer service, or medical assistance, and has not yet been deeply integrated with trust control mechanisms in digital security systems.

[0007] Regarding the aforementioned technologies, the inventors believe that there is a lack of technology that can quantitatively model the results of emotion recognition with factors such as risk perception and algorithm transparency, and dynamically adjust the system behavior accordingly. Summary of the Invention

[0008] To address the shortcomings of existing technologies that can quantitatively model emotion recognition results with factors such as risk perception and algorithm transparency, and dynamically adjust system behavior accordingly, this paper provides a trust control method and device for digital security systems based on emotion computing.

[0009] This application aims to address the challenges of maintaining public trust dynamically and the lack of system interpretability in digital security systems. By constructing a closed-loop model of risk perception, emotional response, trust determination, and dual-regulation feedback, it achieves real-time monitoring and evaluation of public sentiment and trust levels, intelligent adjustment of algorithm transparency and risk communication signals, and a system self-learning optimization and long-term trust stability mechanism.

[0010] This application provides a trust control method and device for a digital security system based on affective computing, which adopts the following technical solution: Firstly, a trust control method for a digital security system based on affective computing includes the following steps: Step 1: Multimodal Data Acquisition Information such as video, audio, and behavioral data is collected, and then anonymized and encrypted to ensure privacy and security. Step 2: Emotion Recognition and Feature Extraction The CNNLSTM fusion model is used to extract sentiment features and generate sentiment vectors (ER). Step 3: Risk Perception Calculation The risk perception index is obtained by weighting and summing the privacy risk r1, bias risk r2, and control risk r3: RP = Σwᵢ·rᵢ (i=1…n); Step 4: Trust Prediction Model Establish the structural equation: TW = β0 + β1·RP + β2·ER + γ3·(RP×AT) + ε Where TW represents trust intention, AT represents algorithm transparency adjustment factor, β0 represents constant term (bias), β1 represents risk perception coefficient, β2 represents emotional response coefficient, and ε represents residual term; Step 5: Dual-adjustment feedback mechanism When TW falls below the threshold T0, the system automatically performs the following: 1. Increased transparency – displaying the algorithm logic and data usage instructions; 2. Risk communication signal output – voice or text reassurance prompts; Step Six: Self-Learning Optimization Trust changes are input into the learning unit, which automatically updates the model parameters, enabling the system to adapt and iterate.

[0011] By employing the above technical solutions, the secure collection and de-identification encryption of multimodal data such as video, voice, and behavior enable privacy-friendly perception of user status, providing a reliable data foundation for trust regulation. Utilizing a CNN-LSTM fusion model for temporal modeling of emotional features allows for real-time quantification of user emotional changes, enhancing the security system's ability to identify public psychological states. Introducing privacy risk, bias risk, and control risk to construct a risk perception index enables a comprehensive assessment of users' subjective risk perception. Combining emotion vectors with algorithmic transparency adjustment factors establishes a trust prediction model, enabling dynamic calculation and interpretable evaluation of trust intention. When trust intention falls below a threshold, the system automatically implements transparency enhancement and risk communication feedback, proactively alleviating user anxiety and restoring trust. Continuous updating of model parameters through a self-learning optimization mechanism enables adaptive evolution of trust regulation strategies, thereby improving the credibility, stability, and long-term operational effectiveness of the digital security system.

[0012] Optionally, in the multimodal data acquisition step, video information includes facial expressions, gaze direction, and micro-expression features; voice information includes speech rate, tone, and intensity changes; and behavioral information includes dwell time, path trajectory, and interaction frequency. During the acquisition phase, the multimodal data replaces personal identity information with local anonymization identifiers.

[0013] By adopting the above technical solution, and by collecting facial expressions, gaze direction, micro-expressions, speech rate, tone, volume, and behavioral characteristics such as dwell time, path trajectory, and interaction frequency, the system can quantify the user's psychological state and behavioral intentions, improve the accuracy of emotion recognition and trust assessment, and at the same time, replace personal identity information with local anonymized identifiers to effectively protect privacy and security, reduce risks, and improve the compliance and public trust of digital security systems.

[0014] Optionally, in the emotion recognition and feature extraction step, the CNN-LSTM fusion model uses a convolutional neural network to encode spatial features and a long short-term memory network to model time-series emotion changes, outputting a normalized emotion vector ER, which includes at least tension, reassurance and resistance components.

[0015] By adopting the above technical solution, the system can accurately identify the spatiotemporal dynamics of user emotions. By encoding the spatial features of video, voice and behavioral data through convolutional neural networks and modeling time series changes using long short-term memory networks, the system can capture emotional fluctuations and subtle changes, and generate a normalized emotion vector ER containing components such as tension, sense of security and resistance. This enhances the real-time quantification capability of digital security systems of users' psychological states and provides reliable data support for trust assessment and dynamic control.

[0016] Optionally, in the risk perception calculation step, the weights wᵢ of privacy risk r1, bias risk r2, and control risk r3 are dynamically adjusted based on historical trust feedback data, and the risk factors are standardized by the constraint Σwᵢ=1.

[0017] By adopting the above technical solution, dynamic quantification and personalized assessment of multidimensional risks can be achieved. By combining privacy risks, bias risks, and control risks, and dynamically adjusting the weights wᵢ of each risk based on historical trust feedback data, the system can reflect the user's actual sensitivity to different risk factors. At the same time, through the standardization constraint of Σwᵢ=1, the risk factors are calculated uniformly under the same dimension, generating a comparable risk perception index. This provides accurate and reliable risk input data for trust prediction and adaptive control, thereby improving the credibility and intelligence level of the digital security system.

[0018] Optionally, the algorithm transparency adjustment factor AT is set in a hierarchical manner according to the information level that the system opens to users, including the basic information explanation layer, the model logic visualization layer, and the decision path explanation layer, and AT increases monotonically with the increase of the transparency level.

[0019] By adopting the above technical solution, the system achieves hierarchical control and dynamic feedback of algorithm transparency. By mapping the transparency adjustment factor AT to the information levels that the system opens to users, including basic information explanations, model logic visualization, and decision path explanations, the system can flexibly adjust the transparency according to user needs or trust levels. AT increases monotonically with the transparency level, enabling users to intuitively understand the system's operating logic and data usage, thereby enhancing user trust, reducing anxiety, and providing technical guarantees for the interpretability and controllability of digital security systems.

[0020] Optionally, in the dual-adjustment feedback mechanism, when the trust intention TW is lower than the threshold T0, the system selects different strengths of transparency enhancement strategies and risk communication signal output strategies according to the magnitude of the risk perception index RP, so as to achieve graded intervention and differentiated trust repair. The trust control method for digital security systems based on affective computing is characterized in that: in the self-learning optimization step, the model parameters are updated using an incremental learning approach, and the coefficients β1, β2, and γ3 in the trust prediction model are iteratively corrected by combining the user's individual historical trust trajectory, thereby improving the accuracy and stability of trust prediction in long-term operation of the system.

[0021] Using the above technical solution, when the trust intention TW is lower than the threshold T0, the system automatically selects different levels of transparency enhancement and risk communication strategies based on the risk perception index RP, achieving graded intervention and personalized trust repair. At the same time, through incremental self-learning optimization, the trust prediction model parameters β1, β2, and γ3 are iteratively updated in combination with the user's historical trust trajectory, enabling the system to continuously improve the accuracy and stability of trust prediction in long-term operation, and enhance the sustainable reliability and adaptability of the digital security system.

[0022] Secondly, a device for trust regulation in a digital security system based on affective computing includes a data acquisition unit, an affective recognition unit, a risk assessment unit, a trust determination unit, a regulation unit, and a learning unit, wherein the data feedback of the learning unit is connected to the data acquisition unit; Data is transmitted sequentially from the acquisition unit to the emotion recognition unit, risk assessment unit, trust determination unit, control unit, and learning unit.

[0023] By adopting the above technical solution, modular and real-time processing of trust control in digital security systems is achieved. The acquisition unit is responsible for the secure acquisition of multimodal data, providing basic input for the system; the emotion recognition unit extracts user emotional features and generates emotion vectors; the risk assessment unit calculates a comprehensive risk index; the trust determination unit combines emotion and risk to predict trust willingness; the control unit implements transparency and risk communication feedback based on the prediction results; and the learning unit continuously updates the parameters of each module through data feedback, realizing the system's adaptive loop and dynamic optimization, thereby improving the intelligence, reliability, and user acceptance of the security system.

[0024] Optionally, the acquisition unit includes a video acquisition module, a voice acquisition module, and a behavior acquisition module. The acquisition unit is equipped with a privacy protection module, which is used to desensitize and encrypt the acquired multimodal data before transmitting it to the emotion recognition unit.

[0025] By adopting the above technical solution, secure and efficient collection and preprocessing of user multimodal data can be achieved. The video acquisition module, voice acquisition module, and behavior acquisition module respectively acquire facial expressions, voice features, and behavioral trajectories to achieve comprehensive perception of user status. The privacy protection module desensitizes and encrypts the multimodal data before data output to effectively prevent the leakage of personal information. The processed data is transmitted to the emotion recognition unit to provide a reliable and secure data foundation for subsequent emotion analysis, risk assessment, and trust regulation, thereby improving the privacy protection capabilities and credibility of the digital security system.

[0026] Optionally, the trust determination unit has a built-in trust prediction model, which is used to generate a trust willingness value based on the risk perception index output by the risk assessment unit and the emotion vector output by the emotion recognition unit. When the trust willingness value is lower than a preset threshold, the control unit is triggered to perform algorithm transparency improvement and risk communication feedback operations, and the learning unit adaptively updates the parameters of the trust prediction model according to the control results. The emotion recognition unit includes a feature extraction submodule and an emotion fusion submodule. The feature extraction submodule is used to encode features of video, speech and behavioral data respectively. The emotion fusion submodule is used to fuse multimodal features based on time series to generate an emotion vector for trust determination. The learning unit includes a parameter update module and a model storage module. The parameter update module is used to iteratively adjust the model parameters in the trust determination unit based on the results of changes in trust willingness. The model storage module is used to save the updated model parameters and provide calling support to the trust determination unit to achieve continuous adaptive operation of the system.

[0027] The above technical solution realizes a closed-loop function of trust prediction and adaptive control in the digital security system. The trust determination unit generates a trust willingness value based on the risk perception index and emotion vector. When the value is lower than the threshold, the control unit is triggered to perform transparency improvement and risk communication feedback to achieve real-time trust repair. The emotion recognition unit provides accurate emotion input through feature extraction and multimodal emotion fusion. The learning unit iteratively updates the model parameters based on feedback to ensure continuous adaptive optimization of the system, improve the accuracy of trust prediction and the stability and reliability of long-term operation.

[0028] Thirdly, a computer-readable storage medium storing a computer program thereon, said computer program, when executed by a processor, is used to implement the trust control method for a digital security system based on emotion computing as described in any one of claims 1 to 7, comprising: Collect multimodal data that has undergone anonymization and encryption. Emotion vectors are generated based on an emotion recognition model; The risk perception index is obtained by weighting multiple risk factors. Trust willingness value is calculated based on risk perception index, sentiment vector and algorithm transparency adjustment factor; When the trust willingness value falls below a preset threshold, transparency adjustment and risk communication feedback are triggered. The model parameters are adaptively updated based on the changes in trust to achieve closed-loop optimization of trust regulation.

[0029] By adopting the above technical solution, the program on the storage medium collects de-identified and encrypted multimodal data, generates emotion vectors, and calculates risk perception index and trust willingness value to achieve quantitative assessment of user emotions and risks. When the trust willingness is lower than the threshold, transparency enhancement and risk communication feedback are automatically triggered. At the same time, combined with an adaptive parameter update mechanism, a closed-loop optimization of trust prediction and regulation is achieved, improving the system's intelligence, executability, and long-term operational reliability.

[0030] In summary, this application includes at least one of the following beneficial technical effects: 1. This project introduces affective computing into digital security trust regulation for the first time, achieving the integration of public psychology and system security.

[0031] 2. A dual adjustment mechanism (algorithm transparency AT and risk communication signal RCS) is proposed to significantly reduce trust fluctuations and improve user acceptance.

[0032] 3. The system has a self-learning function, which can optimize the model and automatically update the feedback strategy based on real-time data.

[0033] 4. Verification of technical effectiveness: The average accuracy rate of emotion recognition is over 88%, and the trust recovery time is shortened by 30%.

[0034] 5. It can be extended to human-computer trust scenarios such as intelligent transportation, public safety, and financial risk control. Attached Figure Description

[0035] Figure 1 This is a schematic diagram of a trust control device for a digital security system based on emotion computing, according to an embodiment of this application.

[0036] Figure 2 This is a schematic diagram of the trust model path in an embodiment of this application.

[0037] Figure 3 Algorithm adjustment feedback sequence diagram of embodiments of this application. Detailed Implementation

[0038] The following is in conjunction with the appendix Figure 1-3 This application will be described in further detail.

[0039] This application discloses a method and apparatus for trust control in a digital security system based on affective computing. (Refer to...) Figure 2 , Figure 3 This includes the following steps: Step 1: Multimodal Data Acquisition Information such as video, audio, and behavioral data is collected, and then anonymized and encrypted to ensure privacy and security. Step 2: Emotion Recognition and Feature Extraction The CNNLSTM fusion model is used to extract sentiment features and generate sentiment vectors (ER). Step 3: Risk Perception Calculation The risk perception index is obtained by weighting and summing the privacy risk r1, bias risk r2, and control risk r3: RP = Σwᵢ·rᵢ (i=1…n); Step 4: Trust Prediction Model Establish the structural equation: TW = β0 + β1·RP + β2·ER + γ3·(RP×AT) + ε Where TW represents trust intention, AT represents algorithm transparency adjustment factor, β0 represents constant term (bias), β1 represents risk perception coefficient, β2 represents emotional response coefficient, and ε represents residual term; Step 5: Dual-adjustment feedback mechanism When TW falls below the threshold T0, the system automatically performs the following: 1. Increased transparency – displaying the algorithm logic and data usage instructions; 2. Risk communication signal output – voice or text reassurance prompts; Step Six: Self-Learning Optimization Trust changes are input into the learning unit, and the model parameters are automatically updated to achieve system adaptive looping; The installation relationship is: data collection → identification → evaluation → judgment → regulation → learning, forming a closed-loop structure; The working principle is that risk perception and emotional response interact to influence the willingness to trust, and the trust gap is dynamically repaired through two moderating variables.

[0040] A trust regulation device for a digital security system based on affective computing, referring to Figure 1 It includes a data acquisition unit, an emotion recognition unit, a risk assessment unit, a trust determination unit, a control unit, and a learning unit. The data feedback from the learning unit is connected to the data acquisition unit. Data is transmitted sequentially from the acquisition unit to the emotion recognition unit, risk assessment unit, trust determination unit, control unit, and learning unit. The acquisition unit includes a video acquisition module, a voice acquisition module, and a behavior acquisition module. The acquisition unit is equipped with a privacy protection module, which is used to desensitize and encrypt the acquired multimodal data before transmitting it to the emotion recognition unit. The trust determination unit has a built-in trust prediction model, which is used to generate a trust willingness value based on the risk perception index output by the risk assessment unit and the emotion vector output by the emotion recognition unit. When the trust willingness value is lower than the preset threshold, the control unit is triggered to perform algorithm transparency improvement and risk communication feedback operations, and the learning unit adaptively updates the parameters of the trust prediction model according to the control results. The emotion recognition unit includes a feature extraction submodule and an emotion fusion submodule. The feature extraction submodule is used to encode features from video, speech and behavioral data respectively. The emotion fusion submodule is used to fuse multimodal features based on time series to generate an emotion vector for trust determination. The learning unit includes a parameter update module and a model storage module. The parameter update module is used to iteratively adjust the model parameters in the trust determination unit based on the results of changes in trust willingness. The model storage module is used to save the updated model parameters and provide calling support to the trust determination unit to achieve continuous adaptive operation of the system. Each unit of the device enables real-time, modular, and adaptive closed-loop operation of trust regulation in the digital security system. Video, voice, and behavioral information are transmitted to the emotion recognition unit after being desensitized and encrypted, ensuring data integrity and availability while protecting user privacy. The emotion recognition unit generates an emotion vector reflecting the user's psychological state through feature extraction and multimodal emotion fusion, providing accurate input for trust determination. The trust determination unit combines the emotion vector and risk perception index to generate a trust willingness value. When the value falls below a threshold, the regulation unit is triggered to implement transparency enhancement and risk communication feedback, achieving real-time trust repair and tiered intervention. The learning unit iteratively updates and saves the model parameters based on trust changes, ensuring that the system continuously optimizes the trust prediction accuracy and regulation strategy during long-term operation, improving system stability, intelligence level, and user trust.

[0041] A computer-readable storage medium storing a computer program, which, when executed by a processor, implements a trust control method for a digital security system based on emotion computing. The method includes: collecting de-identified and encrypted multimodal data; generating an emotion vector based on an emotion recognition model; calculating a risk perception index by weighting multiple risk factors; calculating a trust willingness value based on the risk perception index, the emotion vector, and an algorithm transparency adjustment factor; triggering transparency adjustment and risk communication feedback when the trust willingness value falls below a preset threshold; and adaptively updating model parameters based on trust changes to achieve closed-loop optimization of trust control. The method involves collecting de-identified and encrypted multimodal data to assess user emotions and behaviors related to security; generating an emotion vector based on an emotion recognition model and calculating a risk perception index based on multiple risk factors to quantitatively assess user psychological state and risk perception; calculating a trust willingness value using a trust prediction model and automatically triggering transparency enhancement and risk communication feedback when the value falls below a threshold for real-time trust repair; and combining an adaptive parameter update mechanism to enable the system to continuously optimize trust prediction and control strategies, thereby improving the intelligence, executability, and long-term operational stability of the digital security system.

[0042] 1. The emotion recognition accuracy data in this technical solution is shown in Table 1 below: Table 1 2. The algorithm execution log sample of this technical solution is shown in Table 2 below: Table 2 In Table 2, each log entry contains one record and can be saved as CSV or TXT format, using "UTF8" encoding.

[0043] 3. Risk Communication and Notification Script System notification: To protect your privacy and security, this algorithm has activated transparent mode. The current analysis is used only for public safety risk identification and does not involve the storage of personal data.

[0044] Please feel free to use it; the system complies with data compliance and ethical standards throughout the entire process.

[0045] Example of a voice API call (JSON): { "signal_id": "RCS_001", "signal_type": "voice", "content": "The system is currently performing a security check. Please do not worry about privacy leaks." "language": "zh-CN", "volume": 0.9, "emotion_tone": "calm" } The key technical points of this technical solution are: • 1. A dynamic trust prediction algorithm is formed by fusing risk perception (RP) and emotional response (ER) in a model. 2. Algorithm transparency and risk communication signals constitute a dual-regulation closed-loop feedback mechanism; 3. The self-learning module enables continuous updates of model parameters.

[0046] The main protection points of this technical solution are: 1. Overall process and system architecture of trust regulation method based on affective computing; 2. Trust prediction model and threshold-triggered feedback logic; 3. Methods for implementing transparent algorithm display and risk communication signals; 4. Self-learning module and parameter update algorithm.

[0047] Alternative solutions to this technical solution: To ensure that the present invention can achieve the same technical effect under different implementation environments and hardware conditions, the following equivalent technical implementation methods are allowed. However, all alternative solutions should follow the core logical framework of "risk perception - emotional response - trust judgment - dual adjustment feedback" and should not affect the overall structure and functional relationship of the system.

[0048] Alternative implementation of the emotion recognition module In addition to CNNLSTM models, Transformer, TemporalConvNet, or HybridTransformer can be used; When video / audio is insufficient, physiological signals (pulse rate, ECG, skin conductance response) can be used to identify emotions, maintaining an accuracy rate of ≥85%. Attention-weighted or multi-task shared fusion layers can be used to optimize feature extraction; Alternative implementation of trust model and risk perception module In addition to linear models, Bayesian trust networks or fuzzy hierarchical analysis models can be used to calculate TW; Risk indicators can be expanded to include delay risk, system reliability risk, or social sentiment risk.

[0049] Alternative Implementation of Algorithm Transparency and Risk Communication Signaling Module Transparency can be displayed through decision path visualization, feature weight ranking, or natural language description. Communication output can be voice, text, video, graphics, or vibration alerts; The triggering condition can be based on the TW threshold or the trend change of ΔTW / Δt.

[0050] Alternative implementation of the learning and feedback module Learning algorithms can employ reinforcement learning, transfer learning, or meta-learning. The feedback path can be extended to hardware sensor self-calibration; System Implementation and Interface Extension The deployment platform can be in the cloud, at the edge, or on an embedded terminal; The communication protocol can be HTTP, MQTT, or ROS, and the data format can be JSON / XML / Protobuf; Safety certification complies with ISO / IEC 27001 or GB / T 35273-2020 standards.

[0051] (vi) Explanation of the equivalent conditions of the alternative solutions All alternative solutions must guarantee: emotion recognition accuracy ≥ 85%, trust prediction error ≤ 5%, response latency ≤ 500ms, and maintain the core algorithm logic and dual adjustment mechanism unchanged.

[0052] The implementation principle of the trust control method and device for a digital security system based on emotion computing in this application embodiment is as follows: The core idea of ​​the technical solution is to acquire user video, voice and behavioral information through multimodal data acquisition, and ensure privacy and security through desensitization and encryption processing; the emotion recognition unit uses a deep learning model (such as CNN-LSTM or an alternative Transformer model) to extract spatial and temporal features and generate an emotion vector ER reflecting the user's psychological state; the risk assessment unit performs weighted summation on multiple risk factors such as privacy risk, bias risk and control risk to form a risk perception index RP; the trust determination unit fuses the emotion vector ER and the risk perception index RP, and predicts the trust willingness value TW based on structural equation model or alternative model. When TW is lower than the threshold T0, the trust determination unit determines the trust willingness value. At the same time, the control unit automatically executes dual-adjustment feedback: on the one hand, it enhances the transparency of the algorithm by visualizing the decision path, ranking feature weights, or describing it in natural language to help users understand the system logic; on the other hand, it outputs risk communication signals by providing voice, text, graphics, or vibration prompts to reassure users. The learning unit adaptively updates the parameters of the trust prediction model based on trust change data, and can use incremental learning, reinforcement learning, or transfer learning to continuously optimize the trust prediction accuracy and control strategy in long-term operation. The whole system achieves a closed-loop flow of collection, identification, evaluation, judgment, control, and learning through modular design, ensuring that the digital security system can perceive, quantify, and proactively repair user trust in a dynamic environment, thereby improving the system's intelligence, interpretability, privacy security, and long-term operational stability.

[0053] The above are all preferred embodiments of this application, and are not intended to limit the scope of protection of this application. Therefore, all equivalent changes made in accordance with the structure, shape and principle of this application should be covered within the scope of protection of this application.

Claims

1. A trust control method for a digital security system based on affective computing, characterized in that: Includes the following steps: Step 1: Multimodal Data Acquisition Information such as video, audio, and behavioral data is collected, and then anonymized and encrypted to ensure privacy and security. Step 2: Emotion Recognition and Feature Extraction The CNNLSTM fusion model is used to extract sentiment features and generate sentiment vectors (ER). Step 3: Risk Perception Calculation The risk perception index is obtained by weighting and summing the privacy risk r1, bias risk r2, and control risk r3: RP = Σwᵢ·rᵢ (i=1…n); Step 4: Trust Prediction Model Establish the structural equation: TW = β0 + β1·RP + β2·ER + γ3·(RP×AT) + ε Where TW represents trust intention, AT represents algorithm transparency adjustment factor, β0 represents constant term (bias), β1 represents risk perception coefficient, β2 represents emotional response coefficient, and ε represents residual term; Step 5: Dual-adjustment feedback mechanism When TW falls below the threshold T0, the system automatically performs the following:

1. Increased transparency – displaying the algorithm logic and data usage instructions; 2. Risk communication signal output – voice or text reassurance prompts; Step Six: Self-Learning Optimization Trust changes are input into the learning unit, which automatically updates the model parameters, enabling the system to adapt and iterate.

2. The trust control method for a digital security system based on affective computing according to claim 1, characterized in that: In the multimodal data acquisition step, video information includes facial expressions, gaze direction, and micro-expression features; voice information includes speech rate, tone, and intensity changes; and behavioral information includes dwell time, path trajectory, and interaction frequency. During the acquisition phase, the multimodal data replaces personal identity information with local anonymization identifiers.

3. The trust control method for a digital security system based on affective computing according to claim 1, characterized in that: In the emotion recognition and feature extraction steps, the CNN-LSTM fusion model uses a convolutional neural network to encode spatial features and a long short-term memory network to model time-series emotion changes, outputting a normalized emotion vector ER. The emotion vector includes at least tension, reassurance, and resistance components.

4. The trust control method for a digital security system based on affective computing according to claim 1, characterized in that: In the risk perception calculation step, the weights wᵢ of privacy risk r1, bias risk r2, and control risk r3 are dynamically adjusted based on historical trust feedback data, and the risk factors are standardized through the constraint Σwᵢ=1.

5. The trust control method for a digital security system based on affective computing according to claim 1, characterized in that: The algorithm transparency adjustment factor AT is set in a hierarchical manner according to the information level that the system opens to users, including the basic information explanation layer, the model logic visualization layer, and the decision path explanation layer, and AT increases monotonically with the increase of the transparency level.

6. The trust control method for a digital security system based on affective computing according to claim 1, characterized in that: In the dual-adjustment feedback mechanism, when the trust intention TW is lower than the threshold T0, the system selects different strengths of transparency enhancement strategies and risk communication signal output strategies according to the magnitude of the risk perception index RP, so as to achieve graded intervention and differentiated trust repair. The trust control method for digital security systems based on affective computing is characterized in that: in the self-learning optimization step, the model parameters are updated using an incremental learning approach, and the coefficients β1, β2, and γ3 in the trust prediction model are iteratively corrected by combining the user's individual historical trust trajectory, thereby improving the accuracy and stability of trust prediction in long-term operation of the system.

7. A device for trust regulation in a digital security system based on affective computing, characterized in that: It includes a data acquisition unit, an emotion recognition unit, a risk assessment unit, a trust determination unit, a control unit, and a learning unit, wherein the data feedback of the learning unit is connected to the data acquisition unit; Data is transmitted sequentially from the acquisition unit to the emotion recognition unit, risk assessment unit, trust determination unit, control unit, and learning unit.

8. The device for trust regulation in a digital security system based on affective computing according to claim 8, characterized in that: The acquisition unit includes a video acquisition module, a voice acquisition module, and a behavior acquisition module. The acquisition unit is equipped with a privacy protection module, which is used to desensitize and encrypt the acquired multimodal data before transmitting it to the emotion recognition unit.

9. The device for trust regulation in a digital security system based on affective computing according to claim 8, characterized in that: The trust determination unit has a built-in trust prediction model, which is used to generate a trust willingness value based on the risk perception index output by the risk assessment unit and the emotion vector output by the emotion recognition unit. When the trust willingness value is lower than a preset threshold, the control unit is triggered to perform algorithm transparency improvement and risk communication feedback operations, and the learning unit adaptively updates the parameters of the trust prediction model according to the control results. The emotion recognition unit includes a feature extraction submodule and an emotion fusion submodule. The feature extraction submodule is used to encode features of video, speech and behavioral data respectively. The emotion fusion submodule is used to fuse multimodal features based on time series to generate an emotion vector for trust determination. The learning unit includes a parameter update module and a model storage module. The parameter update module is used to iteratively adjust the model parameters in the trust determination unit based on the results of changes in trust willingness. The model storage module is used to save the updated model parameters and provide calling support to the trust determination unit to achieve continuous adaptive operation of the system.

10. A computer-readable storage medium having a computer program stored thereon, the computer program, when executed by a processor, being used to implement the trust control method for a digital security system based on affective computing as described in any one of claims 1 to 7, characterized in that: include: Collect multimodal data that has undergone anonymization and encryption. Emotion vectors are generated based on an emotion recognition model. The risk perception index is obtained by weighting multiple risk factors. Trust willingness scores are calculated based on risk perception index, sentiment vector, and algorithm transparency adjustment factor. When the trust willingness value falls below a preset threshold, transparency adjustment and risk communication feedback are triggered. The model parameters are adaptively updated based on the changes in trust to achieve closed-loop optimization of trust regulation.

Images