Emotional state dynamic monitoring and feedback method

Abstract

The invention discloses an emotional state dynamic monitoring and feedback method, which relates to the technical field of emotional monitoring and comprises the following steps: S1, collecting voice data, S2, monitoring equipment data, S3, determining an emotional score value, S4, calculating a baseline score value and S5, adding label information. According to the method, the accuracy of emotion recognition is improved by combining a time sequence modeling and weighted scoring mechanism of RNN and LSTM, the model not only can extract deep emotion features from sequence data, but also can reflect the actual contribution of different emotion dimensions to the overall state through weight adjustment, so that the score better fits the characteristics of the user, and the user experience is improved. Meanwhile, dynamic calibration and context perception of emotion evaluation are realized by establishing an emotion baseline in a non-interaction state, and a baseline score and a fluctuation value form an'emotion normal state 'reference system of the user, so that the system can distinguish emotion change caused by external interaction from emotion fluctuation of the user, and the user experience is improved. Therefore, the real emotion influence of the interaction behavior can be more accurately judged.

Description

Technical Field

[0001] This invention relates to the field of emotion monitoring technology, specifically to a method for dynamic monitoring and feedback of emotional states. Background Technology

[0002] The dynamic monitoring and feedback method for emotional states collects physiological and behavioral data in real time through wearable devices, dynamically assesses emotional fluctuations, and then pushes personalized intervention strategies to help users achieve self-regulation and improvement of emotions. Patent application number 202410154517.X discloses "a method and system for monitoring emotional states, which involves: acquiring emotional state monitoring data of a target test population; preprocessing the emotional state monitoring data to obtain preprocessed data; determining data indicators related to emotional states; performing feature selection on the preprocessed data based on the data indicators to obtain sample data; constructing a training set based on the sample data; constructing an emotional state monitoring and recognition model based on the AdaBoost ensemble algorithm; inputting the training set into the emotional state monitoring and recognition model for training to obtain a trained emotional state monitoring and recognition model; and inputting the test sample into the trained emotional state monitoring and recognition model to obtain the emotional state result corresponding to the test sample."

[0003] The aforementioned existing technologies have solved the problem of not being able to accurately monitor emotional states. However, when used, they cannot use RNN and LSTM models to jointly analyze emotion scores, resulting in large errors in the evaluation results. Due to the lack of baseline calibration, they cannot distinguish between the user's normal emotions and the real fluctuations caused by specific interactions, which can easily lead to misjudgments. Since the start and end offset values ​​of the interaction process are not determined, it is impossible to quantify the actual impact of a dialogue or multiple interactions on the user's emotions, making it difficult to judge the quality of the interaction experience. Summary of the Invention

[0004] The purpose of this invention is to provide a method for dynamic monitoring and feedback of emotional states to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a method for dynamic monitoring and feedback of emotional states, comprising the following steps:

[0006] S1. Collect voice data: After collecting the user's voice data, the feature vectors in the voice data are transmitted to the RNN model for analysis to obtain the corresponding emotion score.

[0007] S2. Monitoring equipment data: After collecting heart rate variability and skin conductance data, multimodal features are constructed using the collected data and then input into an LSTM model for analysis. The model outputs an emotion score.

[0008] S3. Determine the emotion score: Analyze the corresponding emotion score using RNN and LSTM models, and calculate the actual emotion score of the current user based on the emotion score and the corresponding weight.

[0009] S4. Calculate baseline rating: Collect user emotion data in non-interactive state, and evaluate the baseline rating and rating fluctuation value for all time periods based on the emotion data;

[0010] S5. Add tag information: Determine the starting and ending offset values ​​of the user in the interactive state, count multiple offset values ​​of the same counterpart, and analyze the tag information of the counterpart based on the offset values.

[0011] Preferably, step S1 specifically includes the following steps:

[0012] S101. After collecting the user's voice data through the microphone, preprocess it, including noise reduction, frame segmentation and normalization.

[0013] S102. Obtain the preprocessed speech data, analyze the speech data, and obtain the corresponding fundamental frequency, duration features, energy features, and Mel frequency cepstral coefficients.

[0014] S103. Transmit the fundamental frequency, duration features, energy features, and Mel frequency cepstral coefficients of the speech data to the RNN model, and output the emotion score through the model.

[0015] Preferably, step S2 specifically includes the following steps:

[0016] S201. After collecting heart rate variability and skin conductance data through vital sign monitoring equipment, preprocess the data.

[0017] The preprocessing of heart rate variability and skin conductance data specifically includes the following steps:

[0018] S2011. Set the exercise intensity threshold, mark and remove heart rate variability and skin conductance data segments during high-intensity exercise, detect signal amplitude, and remove invalid data segments that exceed the physiological range due to poor sensor contact or that show long-term flat lines.

[0019] S2012. In the process of processing heart rate variability data, a bandpass filter is applied to the original photoplethysmography pulse wave signal to suppress high-frequency noise and low-frequency baseline drift. The peak detection algorithm is used to identify the pulse wave peak and generate an RR interval sequence. The sequence is then checked a second time, and the abnormal interval caused by ectopic pulsation is smoothed and corrected by the moving average method.

[0020] S2013. In the process of processing electrodermal activity data, a low-pass filter is applied to the original electrodermal signal to remove high-frequency noise, and a polynomial fitting method is used to estimate and subtract the slowly changing baseline, thereby separating the pure phase-dependent electrodermal response signal.

[0021] S2014. According to the time window, the continuous heart rate variability and skin conductance signals are divided into multiple data segments to ensure that the heart rate variability segments and skin conductance segments from the same time period are precisely aligned on the time axis.

[0022] S202. Through time-domain and frequency-domain analysis, determine the standard deviation of normal heartbeat intervals, root mean square of the difference between adjacent heartbeat intervals, low-frequency power, high-frequency power, and the ratio between low-frequency and high-frequency, and decompose skin electrical activity into long-term baseline and short-term peak components.

[0023] Preferably, step S2 further includes the following steps:

[0024] S203. After constructing multimodal features using the standard deviation of normal heartbeat intervals, the root mean square of the difference between adjacent heartbeat intervals, low-frequency power, high-frequency power, the ratio between low-frequency and high-frequency, long-term baseline and short-term peak components, these features are input into the LSTM model for analysis, and the model outputs an emotion score.

[0025] Preferably, step S3 specifically includes the following steps:

[0026] S301. The corresponding sentiment scores are analyzed using RNN and LSTM models. The Softmax activation function is used to generate dynamic weights for the RNN and LSTM models, with the weights summing to 1. A scoring algorithm is then used to calculate the user's actual sentiment score based on the weights and the model's sentiment score. Specifically, the scoring algorithm is as follows:

[0027]

[0028] in, This represents the actual emotion score. This indicates the confidence level of the LSTM model's prediction. This indicates the confidence level of the RNN model's prediction. Represents the natural constant. This represents the sentiment score output by the RNN model. This represents the sentiment score output by the LSTM model;

[0029] The process of generating dynamic weights for the RNN and LSTM models using the Softmax activation function specifically includes the following steps:

[0030] S3011. During the hidden state calculation process of RNN and LSTM models, the corresponding model will generate an original score vector, which is generated through linear transformation and its dimension is consistent with the number of elements that need to be weighted.

[0031] S3012. Input the original score vector into the Softmax function for normalization. The Softmax function calculates the weight of each element as follows:

[0032]

[0033] in Indicates the first One raw fraction, This represents the total length of the vector. Indicates the first One raw fraction, Represents the natural constant. The parameter represents the operation that transforms the original scores into a probability distribution, ensuring that the sum of all weights is 1 and that each weight value is non-negative.

[0034] S3013. The generated Softmax weights are used as dynamic weights in subsequent calculations. In the attention mechanism, these weights are used to perform a weighted summation of the hidden state and the feature vector.

[0035] S3014. For LSTM models, the generation of dynamic weights is further combined with its gating mechanism. For example, in attention LSTM, the hidden state at the current time step interacts with the hidden states at all time steps to generate attention scores. Then, after normalization by Softmax, the historical hidden states are weighted to finally generate a context vector and input it into the next step of LSTM calculation.

[0036] S302. Determine the corresponding emotional state based on the current user's actual emotional score. The emotional state includes pleasure, calmness, mild stress, moderate stress, and severe stress.

[0037] Preferably, step S3 further includes the following steps:

[0038] S303. When the emotional state is pleasant, encouraging text prompts are displayed through a visual interface.

[0039] S304. When the emotional state is calm, do not perform any operation;

[0040] S305. When the emotional state is mild stress, display emotional regulation text prompts through a visual interface;

[0041] S306. When the emotional state is moderate stress, push professional short videos on emotion regulation.

[0042] S307. When the emotional state is under severe stress, an emotional warning message is automatically sent to the preset emergency contact and the address and contact information of nearby psychological counseling institutions are displayed.

[0043] Preferably, step S4 specifically includes the following steps:

[0044] S401. Collect user emotion data in non-interactive state, and evaluate the emotion score value for each time period of the day based on the emotion data.

[0045] S402, Obtain the daily number Mood score for a given time period ,according to Calculate the baseline score for this time period. ,in , Indicates the first day The emotional score for a given time period. Indicates the first day The emotional score for a given time period. Indicates the first day The emotional score for a given time period. Indicates the first day The emotional score for a given time period. Indicates the serial number. Indicates the total number of days. Indicates the time period sequence number;

[0046] S403, to and Analysis was performed, and the result was obtained. Rating fluctuation value over a period of time ,in .

[0047] Preferably, step S5 specifically includes the following steps:

[0048] S501. After determining the emotional score and score fluctuation value for all time periods, determine whether the user is in an interactive state. If not in an interactive state, analyze the emotional score based on the currently acquired emotional data, and update the corresponding baseline score and emotional score according to the emotional score and the corresponding time period.

[0049] S502. If in an interactive state, obtain the emotion score value corresponding to the current emotion data at the start of the interaction. ,according to Calculate the emotional offset value ,in , Indicates the first Baseline score for each time period, Indicates the first The rating fluctuation value over a period of time will As the starting offset value for the current user;

[0050] S503. Obtain the emotion score corresponding to the current emotion data at the end of the interaction. ,according to Calculate the emotional offset value ,in , Indicates the first Baseline score for each time period, Indicates the first The rating fluctuation value over a period of time, according to As the end offset value for the current user, calculate the difference between the start offset value and the end offset value, and use it as the offset difference value;

[0051] S504. Determine the counterpart when the current user is in an interactive state, calculate the offset difference between multiple interactions with the counterpart, analyze the offset difference using hypothesis testing, and determine whether the counterpart will cause the user to have negative emotions. If the counterpart will cause the user to have negative emotions, a negative label will be added to the counterpart's information; otherwise, no label will be added.

[0052] Compared with the prior art, the beneficial effects of the present invention are:

[0053] This invention improves the accuracy and personalization of emotion recognition by combining temporal modeling and weighted scoring mechanisms using RNN and LSTM. The model can not only extract deep emotional features from sequence data, but also reflect the actual contribution of different emotional dimensions to the overall state through weight adjustment, making the scoring more in line with user characteristics. At the same time, by establishing an emotional baseline in the non-interactive state, dynamic calibration and context awareness of emotion assessment are achieved. The baseline score and fluctuation value constitute a reference system for the user's "emotional norm", enabling the system to distinguish between emotional changes caused by external interactions and the user's own emotional fluctuations, thereby more accurately judging the true emotional impact of interactive behavior. Furthermore, by analyzing the start and end offset values ​​in the interaction process and statistically analyzing the corresponding party's label information, a quantitative insight into the dynamics of interaction and the impact on relationships is achieved. This method can not only capture the emotional trajectory of a single interaction, but also reveal the stable influence patterns of specific interacting parties through long-term data aggregation. Attached Figure Description

[0054] Figure 1 An overall method flowchart is provided for embodiments of the present invention. Detailed Implementation

[0055] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0056] Example:

[0057] Please see Figure 1 This invention provides a technical solution: a method for dynamic monitoring and feedback of emotional states, comprising the following steps:

[0058] S1. Collect voice data: After collecting the user's voice data, the feature vectors in the voice data are transmitted to the RNN model for analysis to obtain the corresponding emotion score.

[0059] S2. Monitoring equipment data: After collecting heart rate variability and skin conductance data, multimodal features are constructed using the collected data and then input into an LSTM model for analysis. The model outputs an emotion score.

[0060] S3. Determine the emotion score: Analyze the corresponding emotion score using RNN and LSTM models, and calculate the actual emotion score of the current user based on the emotion score and the corresponding weight.

[0061] S4. Calculate baseline rating: Collect user emotion data in non-interactive state, and evaluate the baseline rating and rating fluctuation value for all time periods based on the emotion data;

[0062] S5. Add tag information: Determine the starting and ending offset values ​​of the user in the interactive state, count multiple offset values ​​of the same counterpart, and analyze the tag information of the counterpart based on the offset values.

[0063] S1 specifically includes the following steps:

[0064] S101. After collecting the user's voice data through the microphone, preprocess it, including noise reduction, frame segmentation and normalization.

[0065] S102. Obtain the preprocessed speech data, analyze the speech data, and obtain the corresponding fundamental frequency, duration features, energy features, and Mel frequency cepstral coefficients.

[0066] S103. Transmit the fundamental frequency, duration features, energy features, and Mel frequency cepstral coefficients of the speech data to the RNN model, and output the emotion score through the model;

[0067] S2 specifically includes the following steps:

[0068] S201. After collecting heart rate variability and skin conductance data through vital sign monitoring equipment, preprocess the data.

[0069] The specific steps involved in preprocessing heart rate variability and skin conductance data are as follows:

[0070] S2011. Set the exercise intensity threshold, mark and remove heart rate variability and skin conductance data segments during high-intensity exercise, detect signal amplitude, and remove invalid data segments that exceed the physiological range due to poor sensor contact or that show long-term flat lines.

[0071] S2012. In the process of processing heart rate variability data, a bandpass filter is applied to the original photoplethysmography pulse wave signal to suppress high-frequency noise and low-frequency baseline drift. The peak detection algorithm is used to identify the pulse wave peak and generate an RR interval sequence. The sequence is then checked a second time, and the abnormal interval caused by ectopic pulsation is smoothed and corrected by the moving average method.

[0072] S2013. In the process of processing electrodermal activity data, a low-pass filter is applied to the original electrodermal signal to remove high-frequency noise, and a polynomial fitting method is used to estimate and subtract the slowly changing baseline, thereby separating the pure phase-dependent electrodermal response signal.

[0073] S2014. According to the time window, the continuous heart rate variability and skin conductance signals are divided into multiple data segments to ensure that the heart rate variability segments and skin conductance segments from the same time period are precisely aligned on the time axis.

[0074] S202. Through time-domain and frequency-domain analysis, determine the standard deviation of normal heartbeat intervals, root mean square of the difference between adjacent heartbeat intervals, low-frequency power, high-frequency power, and the ratio between low-frequency and high-frequency, and decompose skin electrical activity into long-term baseline and short-term peak components.

[0075] S2 further includes the following steps:

[0076] S203. After constructing multimodal features using the standard deviation of normal heartbeat interval, root mean square of the difference between adjacent heartbeat intervals, low-frequency power, high-frequency power, the ratio between low-frequency and high-frequency, long-term baseline and short-term peak components, these features are input into the LSTM model for analysis, and the model outputs an emotion score.

[0077] S3 specifically includes the following steps:

[0078] S301. The corresponding sentiment scores are analyzed using RNN and LSTM models. The Softmax activation function is used to generate dynamic weights for the RNN and LSTM models, with the weights summing to 1. A scoring algorithm is then used to calculate the user's actual sentiment score based on the weights and the model's sentiment score. The scoring algorithm is as follows:

[0079]

[0080] in, This represents the actual emotion score. This indicates the confidence level of the LSTM model's prediction. This indicates the confidence level of the RNN model's prediction. Represents the natural constant. This represents the sentiment score output by the RNN model. This represents the sentiment score output by the LSTM model;

[0081] Generating dynamic weights for RNN and LSTM models using the Softmax activation function includes the following steps:

[0082] S3011. During the hidden state calculation process of RNN and LSTM models, the corresponding model will generate an original score vector, which is generated through linear transformation and its dimension is consistent with the number of elements that need to be weighted.

[0083] S3012. Input the original score vector into the Softmax function for normalization. The Softmax function calculates the weight of each element as follows:

[0084]

[0085] in Indicates the first One raw fraction, This represents the total length of the vector. Indicates the first One raw fraction, Represents the natural constant. The parameter represents the operation that transforms the original scores into a probability distribution, ensuring that the sum of all weights is 1 and that each weight value is non-negative.

[0086] S3013. The generated Softmax weights are used as dynamic weights in subsequent calculations. In the attention mechanism, these weights are used to perform a weighted summation of the hidden state and the feature vector.

[0087] S302. Determine the corresponding emotional state based on the current user's actual emotional score. Emotional states include pleasure, calmness, mild stress, moderate stress, and severe stress.

[0088] S3 also includes the following steps:

[0089] S303. When the emotional state is pleasant, encouraging text prompts are displayed through a visual interface.

[0090] S304. When the emotional state is calm, do not perform any operation;

[0091] S305. When the emotional state is mild stress, display emotional regulation text prompts through a visual interface;

[0092] S306. When the emotional state is moderate stress, push professional short videos on emotion regulation.

[0093] S307. When the emotional state is under severe stress, automatically send an emotional warning message to the preset emergency contact and display the address and contact information of nearby psychological counseling institutions;

[0094] S4 specifically includes the following steps:

[0095] S401. Collect user emotion data in non-interactive state, and evaluate the emotion score value for each time period of the day based on the emotion data.

[0096] S402, Obtain the daily number Mood score for a given time period ,according to Calculate the baseline score for this time period. ,in , Indicates the first day The emotional score for a given time period. Indicates the first day The emotional score for a given time period. Indicates the first day The emotional score for a given time period. Indicates the first day The emotional score for a given time period. Indicates the serial number. Indicates the total number of days. Indicates the time period sequence number;

[0097] S403, to and Analysis was performed, and the result was obtained. Rating fluctuation value over a period of time ,in ;

[0098] S5 specifically includes the following steps:

[0099] S501. After determining the emotional score and score fluctuation value for all time periods, determine whether the user is in an interactive state. If not in an interactive state, analyze the emotional score based on the currently acquired emotional data, and update the corresponding baseline score and emotional score according to the emotional score and the corresponding time period.

[0100] S502. If in an interactive state, obtain the emotion score value corresponding to the current emotion data at the start of the interaction. ,according to Calculate the emotional offset value ,in , Indicates the first Baseline score for each time period, Indicates the first The rating fluctuation value over a period of time will As the starting offset value for the current user;

[0101] S503. Obtain the emotion score corresponding to the current emotion data at the end of the interaction. ,according to Calculate the emotional offset value ,in , Indicates the first Baseline score for each time period, Indicates the first The rating fluctuation value over a period of time, according to As the end offset value for the current user, calculate the difference between the start offset value and the end offset value, and use it as the offset difference value;

[0102] S504. Determine the counterpart when the current user is in an interactive state, calculate the offset difference between multiple interactions with the counterpart, analyze the offset difference using hypothesis testing, and determine whether the counterpart will cause the user to have negative emotions. If the counterpart will cause the user to have negative emotions, a negative label will be added to the counterpart's information; otherwise, no label will be added.

[0103] The analysis of the offset difference using hypothesis testing specifically includes the following steps:

[0104] S5041. Define the assumptions and set the null hypothesis. To ensure that the corresponding side does not cause negative emotions in users, i.e., the mean of the offset difference. Equal to zero, alternative hypothesis The corresponding party will cause negative emotions in users, that is Not equal to zero, select a significance level It is usually taken as 0.05;

[0105] S5042. Obtain the offset difference data of multiple interactions between the current user and the corresponding party. If the data conforms to a normal distribution, then use a parametric test (e.g., ...). (test), otherwise use a non-parametric test (such as...) (Sign-rank test)

[0106] S5043, Regarding The test is performed, and the mean and standard deviation of the deviation differences are calculated to obtain... Statistics, determining degrees of freedom, querying The distribution table yields the critical value for... Test the signed-rank sum statistic, calculate the rank-sum test, and compare it with the critical value;

[0107] S5044. If the absolute value of the test statistic is greater than the critical value, then the null hypothesis is rejected and the alternative hypothesis is accepted, and it is determined that the corresponding party will cause the current user to have negative emotions; otherwise, the null hypothesis is not rejected.

[0108] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

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

Claims

1. A method for dynamic monitoring and feedback of emotional states, characterized in that, The method includes the following steps: S1. Collect voice data: After collecting the user's voice data, the feature vectors in the voice data are transmitted to the RNN model for analysis to obtain the corresponding emotion score. S2. Monitoring equipment data: After collecting heart rate variability and skin conductance data, multimodal features are constructed using the collected data and then input into an LSTM model for analysis. The model outputs an emotion score. S3. Determine the emotion score: Analyze the corresponding emotion score using RNN and LSTM models, and calculate the actual emotion score of the current user based on the emotion score and the corresponding weight. S4. Calculate baseline rating: Collect user emotion data in non-interactive state, and evaluate the baseline rating and rating fluctuation value for all time periods based on the emotion data; S5. Add tag information: Determine the starting and ending offset values ​​of the user in the interactive state, count multiple offset values ​​of the same counterpart, and analyze the tag information of the counterpart based on the offset values.

2. The method for dynamic monitoring and feedback of emotional state according to claim 1, characterized in that: S1 specifically includes the following steps: S101. After collecting the user's voice data through the microphone, preprocess it, including noise reduction, frame segmentation and normalization. S102. Obtain the preprocessed speech data, analyze the speech data, and obtain the corresponding fundamental frequency, duration features, energy features, and Mel frequency cepstral coefficients. S103. Transmit the fundamental frequency, duration features, energy features, and Mel frequency cepstral coefficients of the speech data to the RNN model, and output the emotion score through the model.

3. The method for dynamic monitoring and feedback of emotional state according to claim 1, characterized in that: S2 specifically includes the following steps: S201. After collecting heart rate variability and skin conductance data through vital sign monitoring equipment, preprocess the data. S202. Through time-domain and frequency-domain analysis, determine the standard deviation of normal heartbeat intervals, root mean square of the difference between adjacent heartbeat intervals, low-frequency power, high-frequency power, and the ratio between low-frequency and high-frequency, and decompose skin electrical activity into long-term baseline and short-term peak components.

4. The method for dynamic monitoring and feedback of emotional state according to claim 3, characterized in that: S2 further includes the following steps: S203. After constructing multimodal features using the standard deviation of normal heartbeat intervals, the root mean square of the difference between adjacent heartbeat intervals, low-frequency power, high-frequency power, the ratio between low-frequency and high-frequency, long-term baseline and short-term peak components, these features are input into the LSTM model for analysis, and the model outputs an emotion score.

5. The method for dynamic monitoring and feedback of emotional state according to claim 1, characterized in that: S3 specifically includes the following steps: S301. Analyze the corresponding sentiment scores using RNN and LSTM models, use the Softmax activation function to generate dynamic weights for the RNN and LSTM models, and sum the weights to 1. Calculate the current user's actual sentiment score using the weights and the model's sentiment score. S302. Determine the corresponding emotional state based on the current user's actual emotional score. The emotional state includes pleasure, calmness, mild stress, moderate stress, and severe stress.

6. The method for dynamic monitoring and feedback of emotional state according to claim 5, characterized in that: S3 further includes the following steps: S303. When the emotional state is pleasant, encouraging text prompts are displayed through a visual interface. S304. When the emotional state is calm, do not perform any operation; S305. When the emotional state is mild stress, display emotional regulation text prompts through a visual interface; S306. When the emotional state is moderate stress, push professional short videos on emotion regulation. S307. When the emotional state is under severe stress, an emotional warning message is automatically sent to the preset emergency contact and the address and contact information of nearby psychological counseling institutions are displayed.

7. The method for dynamic monitoring and feedback of emotional state according to claim 1, characterized in that: S4 specifically includes the following steps: S401. Collect user emotion data in non-interactive state, and evaluate the emotion score value for each time period of the day based on the emotion data. S402, Obtain the daily number Mood score for a given time period ,according to Calculate the baseline score for this time period. ,in , Indicates the first day The emotional score for a given time period. Indicates the first day The emotional score for a given time period. Indicates the first day The emotional score for a given time period. Indicates the first day The emotional score for a given time period. Indicates the serial number. Indicates the total number of days. Indicates the time period sequence number; S403, to and Analysis was performed, and the result was obtained. Rating fluctuation value over a period of time ,in .

8. The method for dynamic monitoring and feedback of emotional state according to claim 1, characterized in that: S5 specifically includes the following steps: S501. After determining the emotional score and score fluctuation value for all time periods, determine whether the user is in an interactive state. If not in an interactive state, analyze the emotional score based on the currently acquired emotional data, and update the corresponding baseline score and emotional score according to the emotional score and the corresponding time period. S502. If in an interactive state, obtain the emotion score value corresponding to the current emotion data at the start of the interaction. ,according to Calculate the emotional offset value ,in , Indicates the first Baseline score for each time period, Indicates the first The rating fluctuation value over a period of time will Used as the starting offset value for the current user. Indicates the time period sequence number; S503. Obtain the emotion score corresponding to the current emotion data at the end of the interaction. ,according to Calculate the emotional offset value ,in , Indicates the first Baseline score for each time period, Indicates the first The rating fluctuation value over a period of time. Indicates the date. Indicates the time period sequence number, according to As the end offset value for the current user, calculate the difference between the start offset value and the end offset value, and use it as the offset difference value; S504. Determine the counterpart when the current user is in an interactive state, calculate the offset difference between multiple interactions with the counterpart, analyze the offset difference using hypothesis testing, and determine whether the counterpart will cause the user to have negative emotions. If the counterpart will cause the user to have negative emotions, a negative label will be added to the counterpart's information; otherwise, no label will be added.

Images