Emotional valence perception intervention method and system

Abstract

This invention provides a method and terminal for emotional valence perception intervention. The method includes: collecting multimodal data from the user; calculating the user's initial emotional valence value based on the multimodal data; monitoring the initial emotional valence value; generating a multisensory intervention instruction when the initial emotional valence value meets preset triggering conditions; after executing the multisensory intervention instruction, re-collecting the user's multimodal data and calculating a retested emotional valence value; recording the effectiveness of the executed multisensory intervention instruction based on the comparison between the retested and initial emotional valence values; and updating the multisensory intervention instruction based on accumulated historical data of the user, including the effectiveness of the multisensory intervention instruction. This invention achieves continuous quantification of emotional valence value, real-time verification of intervention effects, and adaptive updating of intervention instructions, improving the accuracy and adaptability of intervention and solving the problems of lack of pleasure indicators, unverifiable effects, and unadaptive strategies.

Description

Technical Field

[0001] This invention relates to the field of emotion intervention, and in particular to methods and systems for intervention based on the perception of emotional valence. Background Technology

[0002] Existing emotion intervention systems mostly employ single-point emotion recognition combined with fixed content triggers, exhibiting significant drawbacks: they rely solely on emotion classification and intensity, lacking the core quantitative indicator of emotional valence, thus failing to accurately measure changes in pleasure; interventions are primarily reactive, lacking predictive pre-intervention, making it difficult to prevent emotional deterioration; they are mostly single-level closed loops, unable to balance immediate calming within seconds with long-term personalized adaptation; intervention methods are limited, primarily using voice / text / simple VR (Virtual Reality) technology, lacking multi-sensory immersive interventions involving hearing, vision, taste, and touch; and strategies rely on preset rule mappings, lacking sufficient intelligence and adaptability. Summary of the Invention The technical problem to be solved by this invention is to provide an emotional valence perception intervention method and system that can realize real-time quantification of emotional valence, instant intervention at the second level, and long-term personalized strategy adaptation.

[0003] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows: An intervention method for perceiving emotional valence, characterized by comprising: Collect multimodal data from users and calculate the initial emotional efficacy value of the users based on the multimodal data; Monitor the initial emotional efficacy value, and when the initial emotional efficacy value meets the preset triggering conditions, generate a multi-sensory intervention instruction; After executing the multi-sensory intervention instruction, the user's multimodal data is re-collected and the retested emotional efficacy value is calculated. The effectiveness of executing the multi-sensory intervention instruction is recorded based on the comparison between the retested emotional efficacy value and the initial emotional efficacy value. The multisensory intervention instructions are updated based on the accumulated historical data of the user, including the validity of the multisensory intervention instructions.

[0004] To solve the above-mentioned technical problems, another technical solution adopted by the present invention is as follows: An emotion valence perception intervention system includes a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the various steps of the aforementioned emotion valence perception intervention method.

[0005] The beneficial effects of this invention are as follows: By collecting users' multimodal data to calculate the initial emotional efficacy value, emotional perception is upgraded from discrete classification to continuous quantification, solving the problem of lacking core indicators of pleasure and enabling precise intervention; by monitoring the initial emotional efficacy value and generating multisensory intervention instructions when preset trigger conditions are met, a leap from passive response to active prediction is achieved, preventing emotional deterioration; by re-collecting multimodal data after executing the multisensory intervention instructions to calculate the retested emotional efficacy value and comparing it with the initial emotional efficacy value, the effectiveness of this intervention is recorded, constructing an instant feedback loop of intervention, retesting, and verification, ensuring that the effect of each intervention is quantifiable and verifiable; by accumulating users' historical data to update the multisensory intervention instructions, an evolution from fixed rules to data-driven adaptation is achieved, allowing the intervention instructions to be continuously personalized and optimized with user use, resulting in substantial improvements in intervention accuracy, real-time response, and strategy adaptability. Attached Figure Description

[0006] Figure 1 This is a flowchart of an emotional valence perception intervention method according to an embodiment of the present invention; Figure 2 This is a schematic diagram of an emotion valence perception intervention system according to an embodiment of the present invention. Detailed Implementation

[0007] To explain in detail the technical content, objectives, and effects of the present invention, the following description is provided in conjunction with the embodiments and accompanying drawings.

[0008] In existing technologies, emotion intervention systems typically employ single-point emotion recognition combined with fixed content triggers. These systems primarily rely on emotion classification (e.g., happiness, sadness, anger) and intensity levels for judgment, lacking emotional valence—a core quantitative indicator for measuring pleasure—making it difficult to accurately capture subtle changes in a user's emotional state. Intervention mechanisms are mostly reactive, initiating intervention only after emotions have significantly worsened, lacking predictive identification of negative deterioration trends and failing to proactively intervene before emotions escalate. System architectures often employ single-level closed-loop structures, either emphasizing immediate response but lacking long-term adaptability, or focusing on strategy optimization but with significant response latency, making it difficult to balance instantaneous calming with long-term personalized adaptation. Intervention methods are relatively limited, mostly consisting of voice, text, or simple virtual reality content, lacking immersive intervention methods that integrate multiple senses such as hearing, vision, touch, and smell. Strategy generation relies on preset rule mappings, lacking adaptive learning capabilities based on user data, and failing to achieve personalized interventions tailored to each individual.

[0009] To address at least the aforementioned issues, this approach involves collecting multimodal data from users and calculating an initial emotional efficacy value. The initial emotional efficacy value is monitored, and a multisensory intervention instruction is generated when preset trigger conditions are met. After executing the intervention instruction, data is recollected to calculate the retested emotional efficacy value, which is then compared with the initial value to record the effectiveness of the intervention. The multisensory intervention instruction is updated based on accumulated historical data. This method enables continuous quantitative perception of emotional efficacy value, elevating emotion assessment from discrete classification to a continuous quantitative indicator. It allows for a shift from passive response to proactive prediction, enabling immediate intervention when emotions show a negative deterioration trend. It constructs an instant feedback loop for intervention, retesting, and verification, ensuring that the effect of each intervention is quantifiable and verifiable. Furthermore, it allows for adaptive updates of intervention instructions based on accumulated historical data, achieving personalized optimization of intervention strategies and substantially improving intervention accuracy, real-time response, and strategy adaptability.

[0010] The following details an intervention method for perceiving emotional valence in this invention. Please refer to [link / reference]. Figure 1 The method 100 includes steps 110 to 140: Step 110: Collect the user's multimodal data and calculate the user's initial emotional efficacy value based on the multimodal data.

[0011] For example, collect users' multimodal data and calculate the initial emotional effect value based on the acquired multimodal data.

[0012] Step 120: Monitor the initial emotional efficacy value. When the initial emotional efficacy value meets the preset triggering conditions, generate a multi-sensory intervention instruction.

[0013] For example, the initial emotional efficacy value is monitored, and when the initial emotional efficacy value reaches a preset trigger condition, a multi-sensory intervention instruction is generated. The multi-sensory intervention instruction is used to simultaneously engage at least two senses to provide an immersive intervention to the user, in order to bring the user's emotional efficacy value back to a normal range.

[0014] Step 130: After executing the multi-sensory intervention instruction, re-collect the user's multimodal data and calculate the retested emotional efficacy value. Record the effectiveness of this execution of the multi-sensory intervention instruction based on the comparison results between the retested emotional efficacy value and the initial emotional efficacy value.

[0015] For example, after executing the generated multi-sensory intervention instructions, the user's multimodal data is collected again and the retested emotional efficacy value is calculated. The retested emotional efficacy value is compared with the initial emotional efficacy value, and the effectiveness of this intervention is recorded based on the comparison results.

[0016] Step 140: Update the multisensory intervention instructions based on the accumulated historical data of the user, including the effectiveness of the multisensory intervention instructions.

[0017] For example, multi-sensory intervention instructions can be updated based on accumulated user history data, which includes records of the effectiveness of each intervention.

[0018] As described above, this invention first improves emotional perception from discrete classification to continuous quantification by collecting users' multimodal data and calculating initial emotional efficacy value, thus solving the problem of lacking core indicators of pleasure. Secondly, by monitoring the initial emotional efficacy value, multisensory intervention instructions are generated when preset trigger conditions are met, achieving a leap from passive response to proactive prediction, enabling immediate intervention when emotions show a negative deterioration trend. Thirdly, by re-collecting data and calculating the retested emotional efficacy value after executing the multisensory intervention instruction, the effectiveness of this multisensory intervention instruction execution is recorded after comparing it with the initial emotional efficacy value, constructing an instant feedback loop of intervention, retesting, and verification, ensuring that the effect of each intervention is quantifiable and verifiable. Finally, the multisensory intervention instructions are updated based on accumulated historical data, including intervention effectiveness, achieving an evolution from fixed rules to data-driven adaptation, allowing the multisensory intervention instructions to be continuously personalized and optimized with user usage. This invention substantially improves intervention accuracy, real-time response, and strategy adaptability.

[0019] In one alternative implementation, step 110 includes steps 111 and 112.

[0020] Step 111: Collect at least one of the user's physiological data, behavioral data, and environmental data as multimodal data.

[0021] For example, one or more of the user's physiological data, behavioral data, or environmental data can be obtained as multimodal data.

[0022] Step 112: Input the multimodal data into the regression network to obtain continuous values, and use the continuous values ​​as the initial emotional effect value.

[0023] For example, multimodal data can be input into a regression network for processing, and a continuous value can be output, which can then be used as the user's initial emotional efficacy value.

[0024] As described above, the process first involves collecting at least one of physiological, behavioral, and environmental data as multimodal data, enabling flexible acquisition of multi-source data. Physiological data reflects internal states, behavioral data captures external manifestations, and environmental data provides contextual information; these three complement each other, providing a rich data foundation for subsequent calculations. Secondly, the multimodal data is input into a regression network to obtain continuous values ​​as initial emotional efficacy values, elevating emotion perception from discrete classification to continuous quantification. This allows for precise characterization of subtle changes in user emotions along the pleasure dimension, such as a deterioration trend from -0.2 to -0.5 or an improvement trend from 0.3 to 0.6. Continuous emotional efficacy values ​​provide quantifiable and comparable core indicators for trigger condition judgment, intervention effect evaluation, and strategy updates, shifting emotion intervention from experience-driven to data-driven, thus improving the accuracy and scientific rigor of the intervention.

[0025] In one alternative implementation, step 112 includes steps 1121 to 1124.

[0026] Step 1121: Perform feature fusion on the multimodal data to obtain a fused feature vector.

[0027] For example, the collected physiological, behavioral, and environmental data can be fused at the feature level. Specifically, methods such as splicing, weighted summation, or attention mechanisms can be used to integrate feature vectors from different modalities into a unified fused feature vector.

[0028] Step 1122: Receive the fused feature vector through the input layer of the regression network.

[0029] For example, the generated fused feature vector can be used as input data and fed into the input layer of a regression network, with the number of neurons in the input layer matching the dimension of the fused feature vector.

[0030] Step 1123: Extract features from the fused feature vector through the hidden layer of the regression network to obtain deep features.

[0031] For example, the hidden layers of a regression network perform layer-by-layer nonlinear transformations on the input fused feature vector. Each layer calculates the output of the previous layer through a weight matrix and activation function, abstracting higher-level feature representations layer by layer, and finally outputting deep features.

[0032] Step 1124: The output layer of the regression network outputs continuous values ​​based on deep features as the initial sentiment effect value.

[0033] For example, the output layer of a regression network receives deep features extracted from the hidden layer. After a linear transformation, these features are mapped to a numerical range from negative one to positive one using a hyperbolic tangent activation function. The output layer then outputs a continuous numerical value as the initial emotional effect value, where negative one represents extremely negative emotion and positive one represents extremely positive emotion. The closer the value is to positive one, the more pleasant the emotion; the closer it is to negative one, the more unpleasant the emotion.

[0034] As described above, the process first involves feature fusion of multimodal data, integrating data from different sources such as physiology, behavior, and environment into a single fused feature vector, thus resolving the problem of heterogeneous multi-source data and the difficulty in unified processing. Secondly, the input layer of the regression network receives this fused feature vector, achieving effective integration between multimodal data and the deep learning network. Next, the hidden layers of the regression network extract features layer by layer from the fused feature vector to obtain deep features, automatically mining deep patterns related to emotional valence using the nonlinear mapping capability of the neural network. Finally, the output layer of the regression network outputs continuous values ​​based on the deep features as initial emotional valence values, completing the end-to-end mapping from multimodal data to a continuous quantifiable indicator. Through this process, the originally discrete and heterogeneous multimodal data is transformed into continuous and quantifiable emotional valence values, providing accurate and unified numerical data for subsequent emotion monitoring, intervention triggering, and effect evaluation.

[0035] In one alternative implementation, step 120 includes steps 121 and 122.

[0036] Step 121: When the initial emotional efficacy value is lower than the preset threshold or when the initial emotional efficacy value is detected to be showing a negative deterioration trend, it is determined that the initial emotional efficacy value meets the preset triggering condition.

[0037] For example, the calculated initial emotional efficacy value is compared with a pre-set threshold. If the initial emotional efficacy value is lower than the threshold, the trigger condition is determined to be met. Alternatively, the change in the initial emotional efficacy value is continuously monitored over a period of time. If a continuously declining negative deterioration trend is detected, the trigger condition is determined to be met even if the value has not yet fallen below the threshold, thus achieving predictive intervention.

[0038] Step 122: Generate multisensory intervention instructions containing interventions for at least two sensory types.

[0039] For example, based on the fulfillment of triggering conditions, an intervention instruction involving multiple senses can be generated. This instruction may include intervention content for at least two of the senses, such as dynamic music, light color temperature adjustment, breathing guidance, and micro-vibration.

[0040] As described above, a dual triggering condition is first established: First, an immediate trigger when the initial emotional efficacy value falls below a preset threshold, ensuring immediate intervention when the user's emotions have significantly deviated from the normal range; second, a predictive trigger when a negative deterioration trend is detected in the initial emotional efficacy value. Even if the current efficacy value has not yet fallen below the threshold, intervention can be initiated in advance as long as a continuous downward trend is monitored, preventing further deterioration of the emotion. Through the complementary combination of these two triggering conditions, both timely response to existing emotional problems and proactive prediction of potential deterioration risks are ensured, thus completing the leap from traditional passive response to proactive predictive intervention. Secondly, this step generates multi-sensory intervention instructions containing at least two sensory types. By simultaneously mobilizing multiple senses—hearing, sight, touch, and smell—an immersive intervention environment is created. Compared to single-sensory intervention, multi-sensory synergistic intervention has a stronger sense of immersion and therapeutic effect, acting on the user's emotional state from multiple dimensions simultaneously, accelerating the recovery of positive emotions.

[0041] In one alternative implementation, step 130 includes steps 131 to 133.

[0042] Step 131: After executing the multi-sensory intervention instruction, re-collect the user's multimodal data and recalculate the retest emotional efficacy value.

[0043] For example, after the multi-sensory intervention instructions are executed, the user's physiological data, behavioral data, and environmental data are collected again as multimodal data. The newly collected multimodal data is then input into a regression network to calculate the retested emotional efficacy value after the intervention.

[0044] Step 132: Compare the retested emotional efficacy value with the initial emotional efficacy value to obtain the comparison results.

[0045] For example, the calculated retested emotional efficacy value is compared with the initial emotional efficacy value recorded before the intervention, and the difference or magnitude of change between the two is calculated as the comparison result.

[0046] Step 133: Record the effectiveness of this multi-sensory intervention instruction based on the comparison results.

[0047] For example, based on the comparison results, determine whether the intervention has achieved the expected effect. If the retested emotional efficacy value is significantly improved compared to the initial emotional efficacy value or has reached the normal range, then record the intervention as effective; otherwise, record it as ineffective, so that the intervention instructions can be optimized based on the accumulated effectiveness records.

[0048] As described above, the process involves three steps: First, after executing the multi-sensory intervention instructions, the user's multimodal data is re-collected, and the retested emotional efficacy value is calculated to obtain a quantitative value of the emotional state after the intervention. Second, the retested emotional efficacy value is compared with the initial emotional efficacy value to obtain a comparison result of the difference between the two, thus quantitatively assessing the emotional changes brought about by this intervention. Finally, the effectiveness of this intervention is recorded based on the comparison results, and the effect of a single intervention is saved in a recordable and traceable form. Through these three steps, a complete feedback loop of intervention, retesting, comparison, and recording is constructed, enabling the effect of each intervention to be quantitatively assessed and recorded, providing a reliable data foundation for subsequent optimization of intervention instructions based on historical data.

[0049] In an alternative implementation, step 133 is followed by step 1331.

[0050] Step 1331: Compare the retested emotional efficacy value with the preset threshold. If the retested emotional efficacy value is lower than the preset threshold, increase the intensity of the multisensory intervention instruction or change the multisensory intervention instruction.

[0051] For example, after recording the effectiveness of this intervention, the retested emotional efficacy value is compared with the preset threshold. If the retested emotional efficacy value is still lower than the preset threshold, it means that this intervention has not achieved the desired effect. At this time, it is necessary to further adjust the intervention strategy: the intensity of the current multi-sensory intervention instructions can be increased, such as increasing the volume, increasing the vibration, adjusting the brightness of the lights, etc.; or another combination of multi-sensory intervention instructions can be used, such as changing the music type from soothing to cheerful, adjusting the breathing guidance rhythm, etc., to try different intervention methods until the emotional efficacy value reaches the target.

[0052] As described above, after recording the effectiveness of the intervention, the retested emotional efficacy value is compared with a preset threshold. If the retested emotional efficacy value is still lower than the preset threshold, a progressive intervention is implemented by increasing the intensity of the intervention or changing the intervention combination. This is an immediate adjustment mechanism that dynamically optimizes the current intervention strategy based on the actual effect of a single intervention, ensuring that adjustments can be made immediately when the effect of a single intervention is unsatisfactory, thus avoiding the continuation of ineffective interventions. Through this progressive intervention mechanism, the inner feedback loop is optimized in real time, making the intervention process more flexible and efficient until the user's emotional efficacy value returns to the normal range.

[0053] In one alternative implementation, step 140 includes steps 141 to 143.

[0054] Step 141: Accumulate historical data on the user's initial emotional efficacy value and the effectiveness records of multi-sensory intervention instructions according to a preset cycle.

[0055] For example, on a minute-by-minute or day-by-day basis, the user's initial emotional efficacy value, the context information at the time of the intervention, and the record of whether the intervention was effective can be accumulated and stored to form a user's historical database.

[0056] Step 142: Construct a user's emotional profile based on accumulated historical data, evaluate the effectiveness of the records, and obtain the evaluation results.

[0057] For example, based on accumulated historical data, we can analyze in which scenarios users are prone to emotional fluctuations, which types of intervention instructions are more effective for the user, and the patterns of changes in the user's emotional efficacy value, thus constructing a personalized emotional profile; at the same time, we can conduct statistical analysis on historical effectiveness records, evaluate indicators such as the success rate and average effect of different intervention instructions, and obtain evaluation results.

[0058] Step 143: Update the multisensory intervention instructions based on the assessment results.

[0059] For example, the generation strategy for multi-sensory intervention instructions can be adjusted based on the evaluation results, such as increasing the weight of effective intervention types, adjusting the threshold of triggering conditions, and optimizing the combination of interventions, so that the subsequently generated intervention instructions are more adapted to the user's personalized needs.

[0060] As described above, the system first accumulates historical data on users' initial emotional efficacy, scenario data, and intervention effectiveness records according to a preset cycle, providing a data foundation for long-term optimization. Secondly, it constructs personalized emotional profiles of users based on the accumulated historical data and evaluates the effectiveness records to uncover patterns in user emotional fluctuations and the distribution of intervention strategy effectiveness. Finally, it updates multi-sensory intervention instructions based on the evaluation results, continuously optimizing the instructions as users use them, achieving an evolution from fixed rules to data-driven adaptive approaches. This results in personalized intervention strategies tailored to each individual and long-term, with intervention effectiveness continuously improving over time.

[0061] In one alternative implementation, step 143 includes step 1431.

[0062] Step 1431: Using a reinforcement learning model, the positive change between the retested emotional efficacy value and the initial emotional efficacy value in the effectiveness record is used as a reward signal, and the multi-sensory intervention instructions are iteratively updated based on the evaluation results.

[0063] For example, a reinforcement learning model can be constructed, taking the user's current state (such as initial emotional efficacy value, scene information, historical intervention records, etc.) as input, the combination of multi-sensory intervention instructions that can be generated as the action space, and the increase in the retest emotional efficacy value relative to the initial emotional efficacy value in the effectiveness records as the reward signal. Combined with the user profile and intervention effectiveness distribution in the evaluation results, the model can be trained by methods such as deep Q-networks to continuously iterate and update the generation strategy of multi-sensory intervention instructions.

[0064] As described above, the reinforcement learning model uses the positive change in the retested emotional efficacy value relative to the initial emotional efficacy value as a reward signal, and the user profile and effectiveness distribution in the evaluation results as prior information to iteratively update the multi-sensory intervention instructions. The reinforcement learning model can continuously learn through trial and error in interaction with the environment, automatically discovering the optimal intervention strategy for a specific user by maximizing long-term cumulative rewards. With the increase in the number of interventions, the model's grasp of user emotional patterns becomes increasingly accurate, and the generated multi-sensory intervention instructions become increasingly personalized, achieving a substantial leap from fixed rules to data-driven approaches, and from general strategies to personalized solutions for each individual.

[0065] In an alternative implementation, the method further includes step 150.

[0066] Step 150,

[0067] Where R is the reward signal. To retest the emotional efficacy value, Let t represent the initial emotional efficacy value, and t represent the intervention duration of the multi-sensory intervention instruction. The reward weighting for the initial emotional effect value. The penalty weight for the duration of intervention.

[0068] For example, when the post-intervention emotional efficacy value shows a significant improvement compared to the initial emotional efficacy value, the reward signal is positive, encouraging the model to continue using similar intervention strategies. Conversely, when the intervention period is too long and the improvement is small, the subtracted penalty term will lower the reward signal, prompting the model to shorten the intervention period as much as possible while maintaining effectiveness. This can be achieved through adjustments. and The value of can balance the importance of intervention effect and intervention efficiency.

[0069] Based on this, a Deep Q-Network (DQN) is used as the reinforcement learning model, and the objective function is set as follows:

[0070] in The Q value is the current output of the network. In this application, it represents the expected positive emotional stimulus that can be obtained by taking a specified multi-sensory intervention instruction A under the user's current emotional state S. Q is the output of the target network, used to stabilize the training objective; R is the reward value, calculated by the reward function, representing the actual improvement in the user's emotional efficacy value after this intervention. This is a discount factor, ranging from 0.9 to 0.95, used to balance the importance of current rewards and future rewards; This represents the expectation of randomly sampling from the experience pool. Through this objective function, the network's predicted Q-value is continuously fitted to the target value, thereby learning the optimal intervention strategy that maximizes the user's long-term emotional benefits.

[0071] Update network parameters using gradient descent:

[0072] Where η is the learning rate, a very small positive number used to control the step size of each update; The gradient of the loss function with respect to the parameter θ points in the direction of the fastest increase in loss; updating in the opposite direction can gradually reduce the loss. Through this method, the model aims to maximize long-term cumulative reward and iteratively optimizes the intervention strategy.

[0073] As described above, by setting a reward function, the positive change in emotional efficacy before and after the intervention is used as the reward, and the intervention duration is used as the penalty. This allows the reinforcement learning model to quantitatively evaluate the quality of each intervention during iterative training. A larger reward value indicates that the intervention achieved a better emotional improvement effect in a shorter time; a smaller or negative reward value indicates a poor intervention effect or excessive time consumption. This reward function provides a clear optimization objective for the reinforcement learning model, guiding it to continuously evolve towards improving intervention effectiveness and shortening intervention time, ultimately generating efficient and personalized multi-sensory intervention instructions.

[0074] The following is a detailed description of an embodiment of the present invention, specifically the process of the emotional valence perception intervention method of the present invention in an emotional intervention scenario, including S21 to S24.

[0075] S21. First, multimodal data acquisition and initial emotional efficacy value calculation are performed. At least one of the following is collected as multimodal data: user physiological signals (e.g., heart rate, heart rate variability, skin conductance, electroencephalography), speech, text, eye movement, facial micro-expressions, and environmental information. Feature fusion is performed on the collected multimodal data to obtain a fused feature vector. The fused feature vector is input into a regression network, which includes an input layer, hidden layers, and a single-neuron output layer. The hidden layers use an activation function and are configured with dropout regularization to suppress overfitting. The fused feature vector is received through the input layer of the regression network, and deep features are extracted from the fused feature vector through the hidden layers. The output layer outputs a continuous value normalized to the range of negative one to positive one based on the deep features, which serves as the initial emotional efficacy value. Here, negative one represents extremely negative emotion, and positive one represents extremely positive emotion. This is equivalent to steps 110, 111, 112, 1121, and 1124 above.

[0076] S22. Monitor the initial emotional efficacy value and generate multi-sensory intervention instructions. Compare the initial emotional efficacy value with a preset threshold. When the initial emotional efficacy value is lower than the preset threshold or a continuous downward trend in the initial emotional efficacy value is detected, a preset trigger condition is determined to be met. In response to meeting the trigger condition, generate and output multi-sensory intervention instructions containing at least two types of music, light, breathing guidance, and vibration, creating an immersive healing field through multi-sensory synergy. This is equivalent to steps 120, 121, and 122 above.

[0077] S23. After implementing the intervention, conduct a retest, compare the effects, record the effectiveness, and implement progressive intervention. After executing the multi-sensory intervention instructions, re-collect the user's multimodal data and recalculate the retest emotional efficacy value; compare the retest emotional efficacy value with the initial emotional efficacy value to obtain the comparison results and quantitatively evaluate the emotional changes brought about by this intervention; record the effectiveness of this intervention based on the comparison results. If the retest emotional efficacy value is significantly improved compared to the initial emotional efficacy value or has reached the normal range, it is recorded as effective; otherwise, it is recorded as ineffective; compare the retest emotional efficacy value with a preset threshold. If the retest emotional efficacy value is still lower than the preset threshold, increase the intensity of the multi-sensory intervention instructions or change the combination of multi-sensory intervention instructions until the emotional efficacy value reaches the target. This is equivalent to steps 130, 131 to 133, and 1331 above.

[0078] S24. Update multi-sensory intervention instructions based on accumulated historical data. Accumulate historical data on the user's initial emotional efficacy value, scenario data, and intervention effectiveness records according to a preset cycle (minutes or days); construct a personalized emotional profile of the user based on the accumulated historical data, analyze in which scenarios the user is prone to emotional fluctuations, which types of intervention instructions are more effective for the user, and the pattern of emotional efficacy value changes; perform statistical analysis on the effectiveness records, evaluate the success rate and average effect of different intervention instructions, and obtain evaluation results; use a reinforcement learning model (e.g., deep Q-network or policy gradient method), taking the user's current state (e.g., initial emotional efficacy value, scenario information, historical intervention records, etc.) as input, and the generated multi-sensory intervention instruction combinations as the action space, using the positive change between the retested emotional efficacy value and the initial emotional efficacy value as the reward signal, and iteratively update the multi-sensory intervention instructions based on the evaluation results; set the reward function as follows. Where R is the reward signal, To retest the emotional efficacy value, The initial emotional efficacy value is given by t, where t is the intervention duration. The reward weighting for the initial emotional effect value. The penalty weight for intervention duration aims to maximize long-term cumulative rewards, enabling the model to output optimal intervention instructions for different users and scenarios. This is equivalent to steps 140, 141 to 143, 1431, and 150 above.

[0079] Through the above steps, this invention completes the collection, feature fusion, and regression network calculation of multimodal data before intervention begins, realizing end-to-end mapping from multi-source heterogeneous data to continuously quantified emotional efficacy value, solving the problem of existing technologies lacking core indicators of pleasure. By monitoring the initial emotional efficacy value and setting dual triggering conditions (threshold triggering and deterioration trend triggering), it achieves a leap from passive response to active prediction, enabling immediate intervention when emotions show a negative deterioration trend and preventing further deterioration. By performing post-intervention retesting, effect comparison, effectiveness recording, and progressive intervention, a complete feedback loop of intervention, retesting, comparison, recording, and progression is constructed, ensuring that the effect of each intervention is quantifiable and verifiable, and allowing for immediate adjustments when the effect is unsatisfactory. By periodically accumulating historical data, constructing emotional profiles, evaluating effectiveness records, and using reinforcement learning models to iteratively update intervention instructions, it achieves an evolution from fixed rules to data-driven adaptiveness, allowing intervention instructions to be continuously personalized and optimized as users use them. The above mechanisms work together to effectively solve the problems of lacking quantifiable indicators of pleasure, unverifiable intervention effects, and inability to adaptively optimize strategies in existing technologies, resulting in substantial improvements in intervention accuracy, real-time response, and strategy adaptability.

[0080] Please refer to Figure 2The present invention also provides an emotion valence perception intervention system 200, including a memory 201, a processor 202, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the various steps of an emotion valence perception intervention method as described above.

[0081] In summary, this invention provides a method and terminal for intervention based on emotional valence perception. First, multimodal data of the user is collected. Through feature fusion and regression network calculation, a continuous value is output as the initial emotional valence value, elevating emotional perception from discrete classification to continuous quantification, thus solving the problem of existing technologies lacking core indicators of pleasure. Based on this, the initial emotional valence value is monitored. When the initial emotional valence value is lower than a preset threshold or a negative deterioration trend is detected, a preset trigger condition is determined to be met, generating a multisensory intervention instruction containing at least two sensory types. This achieves a leap from passive response to active prediction, enabling immediate intervention when emotions show a negative deterioration trend to prevent further deterioration. After executing the multisensory intervention instruction, multimodal data is re-collected and the re-measured emotional valence value is calculated and compared with the initial emotional valence value. The effectiveness of this intervention is recorded based on the comparison results, constructing an instant feedback loop of intervention, re-measurement, comparison, and recording, ensuring that the effect of each intervention is quantifiable and verifiable. If the re-measured emotional valence value is still lower than the preset threshold, the intervention intensity is increased or the intervention combination is changed, achieving progressive intervention. By accumulating historical data on initial emotional efficacy value, scenario data, and intervention effectiveness records according to a preset cycle, a user emotional profile is constructed and the effectiveness records are evaluated. A reinforcement learning model is adopted, with positive changes in the retest emotional efficacy value compared to the initial emotional efficacy value as a reward signal. Based on the evaluation results, the multi-sensory intervention instructions are iteratively updated, realizing the evolution from fixed rules to data-driven adaptive behavior, so that the intervention instructions are continuously personalized and optimized as users use them.

[0082] Furthermore, in the context of emotion intervention, the accuracy, immediate response capability, and long-term adaptability of emotion intervention are maintained through the coordinated operation of multiple stages, including multimodal data collection and feature fusion, regression network computation, dual trigger condition judgment, multisensory intervention instruction generation, post-intervention retesting and effect comparison, effectiveness recording and progressive intervention, periodic accumulation and emotion profile construction, and reinforcement learning model iterative update.

[0083] The above description is merely an embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent modifications made based on the content of the present invention specification and drawings, or direct or indirect applications in related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A method for intervention based on perceived emotional valence, characterized in that, include: Collect multimodal data from users and calculate the initial emotional efficacy value of the users based on the multimodal data; Monitor the initial emotional efficacy value, and when the initial emotional efficacy value meets the preset triggering conditions, generate a multi-sensory intervention instruction; After executing the multi-sensory intervention instruction, the user's multimodal data is re-collected and the retested emotional efficacy value is calculated. The effectiveness of executing the multi-sensory intervention instruction is recorded based on the comparison between the retested emotional efficacy value and the initial emotional efficacy value. The multisensory intervention instructions are updated based on the accumulated historical data of the user, including the validity of the multisensory intervention instructions.

2. The emotional valence perception intervention method according to claim 1, characterized in that, The process of collecting multimodal data from users, calculating and outputting the user's initial emotional efficacy value based on the multimodal data, includes: The user's physiological data, behavioral data, and environmental data are collected as the multimodal data. The multimodal data is input into a regression network to obtain continuous values, which are then used as the initial emotional effect value.

3. The emotional valence perception intervention method according to claim 2, characterized in that, The step of inputting the multimodal data into a regression network to obtain continuous values, and using the continuous values ​​as the initial sentiment effect value, includes: The multimodal data is fused to obtain a fused feature vector; The fused feature vector is received through the input layer of the regression network; Deep features are obtained by extracting features from the fused feature vector through the hidden layers of the regression network; The output layer of the regression network outputs the continuous value as the initial emotional effect value based on the deep features.

4. The emotional valence perception intervention method according to claim 1, characterized in that, The monitoring of the initial emotional efficacy value, when the initial emotional efficacy value meets preset triggering conditions, generates a multi-sensory intervention instruction, including: When the initial emotional efficacy value is lower than the preset threshold or when the initial emotional efficacy value is detected to be showing a negative deterioration trend, it is determined that the initial emotional efficacy value meets the preset triggering condition. Generate the multisensory intervention instructions that include interventions on at least two sensory types.

5. The emotional valence perception intervention method according to claim 1, characterized in that, After executing the multi-sensory intervention instruction, the user's multimodal data is re-collected and the retested emotional efficacy value is calculated. The effectiveness of this execution of the multi-sensory intervention instruction is recorded based on the comparison between the retested emotional efficacy value and the initial emotional efficacy value, including: After executing the multi-sensory intervention instruction, the user's multimodal data is re-collected and the retested emotional efficacy value is recalculated; The comparison result is obtained by comparing the retested emotional efficacy value with the initial emotional efficacy value. The effectiveness of executing the multisensory intervention command is recorded based on the comparison results.

6. The emotional valence perception intervention method according to claim 5, characterized in that, After recording the effectiveness of executing the multi-sensory intervention command based on the comparison results, the method further includes: The retested emotional efficacy value is compared with a preset threshold. If the retested emotional efficacy value is lower than the preset threshold, the intensity of the multisensory intervention instruction is increased or the multisensory intervention instruction is replaced.

7. The emotional valence perception intervention method according to claim 1, characterized in that, Updating the multisensory intervention instruction based on the accumulated historical data of the user, including the validity of the multisensory intervention instruction, includes: Historical data on the user's initial emotional efficacy value and the effectiveness records of the multi-sensory intervention instructions are accumulated according to a preset period. The user's emotional profile is constructed based on the accumulated historical data, and the validity of the records is evaluated to obtain the evaluation results; The multisensory intervention instructions are updated based on the assessment results.

8. The emotional valence perception intervention method according to claim 7, characterized in that, The step of updating the multisensory intervention instruction based on the evaluation results includes: A reinforcement learning model is used, with the positive change between the retested emotional efficacy value and the initial emotional efficacy value in the validity record as the reward signal, and the multi-sensory intervention instructions are iteratively updated based on the evaluation results.

9. The emotional valence perception intervention method according to claim 8, characterized in that, Also includes: Wherein, R is the reward signal. The retested emotional efficacy value, Let t be the initial emotional efficacy value, and t be the intervention duration of the multi-sensory intervention instruction. The reward weight for the initial emotional effect value. The penalty weight is the duration of the intervention.

10. An emotion valence perception intervention system, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements each step of the emotional valence perception intervention method according to any one of claims 1 to 9.

Images