Personalized emotion adjusting system and method based on music therapy

Abstract

The invention discloses a personalized emotion regulation system and method based on music therapy, and relates to the technical field of intelligent emotion intervention, and the method comprises the steps: collecting a real-time emotion index of a user based on an individual high-sensitivity emotion regulation time period, predicting an emotion perturbation index through a weighted time attenuation integral function, and recognizing a mild emotion fluctuation state of the user; constructing an embedded feature projection model, and generating an intervention music parameter set according to the mild emotion fluctuation state of the user; playing intervention music according to the intervention music parameter set, collecting feedback signals in real time, and calculating an intervention response score by constructing an intervention response score function; and constructing a historical intervention response data set based on the intervention data set, and generating an intervention music recommendation strategy through an intervention situation parameter mapping method. According to the method, an intervention music recommendation strategy is generated through an intervention situation parameter mapping method, and historical forward intervention samples are clustered into intervention response categories.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent emotion intervention, and in particular to a personalized emotion regulation system and method based on music therapy. Background Art

[0002] In the field of emotion regulation, music therapy, as a non-drug, non-invasive intervention, has garnered widespread attention in recent years. Research has shown that musical stimulation can improve an individual's emotional state by influencing autonomic nervous system activity, EEG rhythmic changes, and hormone secretion. Current technologies typically rely on subjective questionnaire responses, matching pre-set music libraries, or making fixed recommendations based on general user characteristics (such as gender and age). These conventional methods typically assess the user's emotional state before intervention and select a number of broadly applicable music clips to play in the hope of achieving emotional relief. In terms of data collection, commonly used physiological indicators include heart rate variability (HRV), electrodermal conduction (EDA), and electroencephalogram (EEG) activity, which support emotion recognition and intervention effectiveness assessment. With the development of wearable devices and artificial intelligence algorithms, this field is gradually transitioning from a rule-driven to a data-driven approach, improving the intelligence level of personalized intervention.

[0003] However, existing emotion regulation methods based on rules or static feature matching still face two limitations. First, most current intervention systems fail to dynamically perceive users' daily rhythms and emotional fluctuations, resulting in a lack of physiological support for the timing of intervention. Second, music recommendation strategies often use static matching methods, making it difficult to optimize intervention effects by incorporating historical feedback. These two shortcomings may affect the timeliness and individual adaptability of music interventions, limiting the universality and stability of emotion regulation systems in real-world application scenarios. Summary of the Invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides a personalized emotion regulation method based on music therapy to solve the problems of inaccurate intervention triggering and lack of adaptability of the recommendation mechanism.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] In a first aspect, the present invention provides a personalized emotion regulation method based on music therapy, which comprises:

[0008] Collect and pre-process the user's life rhythm parameters, calculate the rhythm deviation value, identify the user's biorhythm type, and obtain the individual's highly sensitive emotional regulation time period;

[0009] Based on the individual's high-sensitivity emotion regulation period, the user's real-time emotion indicators are collected, and the emotional perturbation index is predicted through the weighted time decay integral function, and the user's mild emotion fluctuation state is identified;

[0010] Construct an embedded feature projection model to generate a set of intervention music parameters based on the user's mild emotional fluctuation state;

[0011] Play the intervention music according to the intervention music parameter set, collect feedback signals in real time, and calculate the intervention response score by constructing the intervention response score function;

[0012] Based on the intervention data group, a historical intervention response data set is constructed, and an intervention music recommendation strategy is generated through the intervention scenario parameter mapping method.

[0013] As a preferred embodiment of the personalized emotion regulation method based on music therapy of the present invention, the life rhythm parameters include sleep-wake time series, morning heart rate average, nighttime skin conduction peak, daytime alpha wave frequency average, and daytime theta wave frequency average;

[0014] Preprocessing includes data cleaning, standardization and normalization.

[0015] As a preferred solution of the personalized emotion regulation method based on music therapy of the present invention, wherein: by calculating the rhythm deviation value, the biorhythm type of the user is identified,

[0016] A logarithmic weighted structure with dual ratio modulation was used to construct a rhythm deviation evaluation function and calculate the rhythm deviation value.

[0017] Based on the rhythm deviation value and according to the rhythm deviation type identification standard, the user's biorhythm type is identified.

[0018] As a preferred solution of the personalized emotion regulation method based on music therapy described in the present invention, wherein: obtaining the individual high-sensitivity emotion regulation time period, the steps are as follows:

[0019] Based on the historical differences in the regularity of the distribution of neural regulation ability, emotional susceptibility, and intervention response scores of people with different biorhythm types over time, a rhythm time period mapping table is defined, and a set of optimal candidate time periods for regulation is selected based on the user's biorhythm type;

[0020] Based on the set of optimal candidate time periods for regulation and the historical intervention response data set, an individual regulation time scoring function is constructed to obtain the individual highly sensitive emotion regulation time period.

[0021] As a preferred solution of the personalized emotion regulation method based on music therapy described in the present invention, the real-time emotion indicators include heart rate variability sequence, skin electrical change sequence, respiratory rhythm sequence and resting respiratory rate mean.

[0022] As a preferred solution of the personalized emotion regulation method based on music therapy described in the present invention, the steps of constructing an embedded feature projection model are as follows:

[0023] Collect the emotional perturbation index, rhythm deviation value, individual high-sensitivity emotional regulation period, previous round of intervention response score and individual type code to form a five-dimensional feature input vector;

[0024] Based on the five-dimensional feature input vector, the first layer of hidden mapping is performed through the hyperbolic tangent function to obtain the hidden representation of the first stage;

[0025] Through linear transformation, the hidden representation of the first stage is mapped to a low-dimensional embedding space to obtain a compact embedding vector of the individual state;

[0026] Perform activation mapping on the compact embedding vectors of individual states to generate high-dimensional emotion regulation feature representations;

[0027] Based on the high-dimensional emotion regulation feature representation, a set of intervention music parameters is generated through high-dimensional feature transformation to form an embedded feature projection model;

[0028] Based on the training goal of minimizing the prediction error of the intervention response score, the embedded feature projection model is trained through supervised learning, and the embedded feature projection model is updated through the loss function until the embedded feature projection model reaches the set iteration rounds.

[0029] As a preferred solution of the personalized emotion regulation method based on music therapy described in the present invention, the feedback signal includes an HRV curve and a skin electrical change rate.

[0030] As a preferred solution of the personalized emotion regulation method based on music therapy of the present invention, the steps of constructing a historical intervention response data set based on the intervention data set are as follows:

[0031] Based on the intervention time point, rhythm deviation value, individual high-sensitivity emotion regulation period, emotion perturbation index, intervention music parameter set and intervention response score, multiple intervention data groups are formed;

[0032] By statistically analyzing multiple intervention data groups in chronological order, a historical intervention response data set is constructed.

[0033] As a preferred solution of the personalized emotion regulation method based on music therapy described in the present invention, wherein: an intervention music recommendation strategy is generated by an intervention situation parameter mapping method, the steps are as follows:

[0034] Based on the historical intervention response data set, a positive intervention response threshold is set, and the intervention data group with intervention response scores higher than the positive intervention response threshold is selected as the positive feedback sample set;

[0035] The intervention scenario vector is constructed by combining the rhythm deviation value, individual high-sensitivity emotion regulation period and emotion perturbation index;

[0036] A multidimensional clustering algorithm is used to perform cluster analysis on the intervention music parameter set and intervention context vector in the positive feedback sample set, construct an intervention response category set, and calculate the cluster center of each intervention response category in the intervention response category set;

[0037] Based on each cluster center in the intervention response category set, an inverse mapping function from the intervention scenario vector to the intervention music parameter set is constructed, and an intervention music recommendation strategy is generated.

[0038] In a second aspect, the present invention provides a personalized emotion regulation system based on music therapy, comprising:

[0039] The rhythm recognition module is used to collect and pre-process the user's life rhythm parameters, calculate the rhythm deviation value, identify the user's biorhythm type, and obtain the individual's highly sensitive emotional regulation time period;

[0040] The emotion recognition module is used to collect real-time user emotion indicators based on the individual's high-sensitivity emotion regulation time period, predict the emotion perturbation index through the weighted time decay integral function, and identify the user's mild emotion fluctuation state;

[0041] The intervention generation module is used to construct an embedded feature projection model and generate a set of intervention music parameters based on the user's mild emotional fluctuation state;

[0042] The intervention execution module is used to play the intervention music according to the intervention music parameter set, collect feedback signals in real time, and calculate the intervention response score by constructing the intervention response score function;

[0043] The strategy training module is used to construct a historical intervention response data set based on the intervention data group, and generate an intervention music recommendation strategy through the intervention scenario parameter mapping method.

[0044] The beneficial effects of the present invention are: generating an intervention music recommendation strategy through an intervention scenario parameter mapping method, clustering historical positive intervention samples into intervention response categories, and constructing an inverse mapping between intervention scenarios and intervention parameters based on cluster centers, thereby dynamically optimizing the recommendation strategy and improving the adaptability of subsequent interventions. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0046] Figure 1 Flowchart of a personalized approach to emotion regulation based on music therapy.

[0047] Figure 2 Schematic diagram of a personalized emotion regulation system based on music therapy.

[0048] Figure 3 Flowchart obtained for individual high-sensitivity emotion regulation time periods.

[0049] Figure 4 Flowchart generated for the recommended strategy for music intervention. DETAILED DESCRIPTION

[0050] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0051] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0052] Secondly, the term "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in various places throughout this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive of other embodiments.

[0053] Reference Figures 1 to 4 , is an embodiment of the present invention, which provides a personalized emotion regulation method based on music therapy, comprising the following steps:

[0054] S1. Collect and pre-process the user's life rhythm parameters, calculate the rhythm deviation value, identify the user's biorhythm type, and obtain the individual's highly sensitive emotion regulation time period;

[0055] Life rhythm parameters include sleep-wake time series, morning heart rate average, nighttime skin conduction peak, daytime alpha wave frequency average, and daytime theta wave frequency average;

[0056] Preprocessing includes data cleaning, standardization and normalization;

[0057] Furthermore, the sleep-wake time series was normalized and individual outliers were removed; the morning heart rate mean was Z-score standardized; the nighttime skin electrode peak was normalized according to the maximum and minimum ranges of each user; the daytime alpha wave frequency mean was extracted from the resting EEG signal to obtain the main frequency of 8-13Hz, and the data outside the physiological range was Z-score standardized and removed; the daytime theta wave frequency mean was extracted from the resting EEG signal to obtain the main frequency of 4-8Hz, and the data outside the physiological range was Z-score standardized and removed. The mean within a fixed period was calculated and normalized.

[0058] Collect user life rhythm parameters according to a fixed cycle and perform preprocessing;

[0059] It should be noted that 7 consecutive days are generally set as a fixed cycle. Setting 7 consecutive days is sufficient to cover the fluctuations of life rhythms for a full week and can filter out short-term abnormalities in the collected life rhythm parameters.

[0060] For example, users use wearable devices to collect life rhythm parameters in real time: through smart bracelet sleep monitoring, the sleep and wake-up time series is recorded once a day; the chest strap heart rate belt is used to collect the morning average heart rate within 10 minutes after waking up at a frequency of once per second; the skin electricity peak value from 0:00 to 5:00 at night is recorded through the skin electricity sensor, and the maximum value of the skin electricity peak value is selected as the nighttime skin electricity peak value; the EEG headset is used to perform a 3-minute resting EEG measurement once a day to extract the average brain wave frequency; all data are fixed for 7 consecutive days and are initialized after daily collection.

[0061] A logarithmic weighted structure with dual ratio modulation was used to construct a rhythm deviation evaluation function and calculate the rhythm deviation value.

[0062] Calculate the rhythm deviation value, the expression is:

[0063]

[0064] Among them, R is the rhythm deviation value; is the average value of the morning heart rate in a fixed period; H is the morning heart rate average; E is the peak value of the skin electrical peak; max(E) is the maximum value of the fixed period of the skin electrical peak; P α is the fixed period average value of the mean α wave frequency during the day; P θ is the fixed-period average value of the mean of theta wave frequency during the day;

[0065] It should be noted that the rhythm bias evaluation function focuses on the relationship between neuroautonomic indicators and EEG coupling, constructs the matching degree between physiological rhythm phase consistency and sympathetic nerve activity, and is used to reflect the bias of physiological rhythms and exclude interference from explicit behavioral variables. The sleep-wake time series has strong behavioral subjectivity and external environment dependence, and is an explicit behavioral variable. When calculating the rhythm bias value, the sleep-wake time series is not used.

[0066] Based on the rhythm deviation value, the user's biorhythm type is identified according to the rhythm deviation type identification standard;

[0067] It should be noted that based on the statistical analysis of historical rhythm bias type data, the rhythm type threshold is defined. The rhythm type threshold includes a low rhythm type threshold (usually in the range of [0.9, 1.1]) and a high rhythm type threshold (usually in the range of [1.2, 1.4]);

[0068] The criteria for identifying rhythm deviation types are:

[0069] When R < low rhythm type threshold, the user's biorhythm type is identified as late type;

[0070] When the low rhythm type threshold ≤ R ≤ the high rhythm type threshold, the user's biorhythm type is identified as neutral;

[0071] When R>high rhythm type threshold, the user's biorhythm type is identified as early-onset type.

[0072] Based on the historical differences in the regularity of the distribution of neural regulation ability, emotional susceptibility, and intervention response scores of people with different biorhythm types over time, a rhythm time period mapping table is defined, and a set of optimal candidate time periods for regulation is selected based on the user's biorhythm type;

[0073] It should be noted that the indicators for determining neuroregulatory ability include the baseline value and fluctuation amplitude of heart rate variability and the baseline level and recovery speed of skin conductance. Specifically, the baseline value of heart rate variability is obtained by calculating the mean of the heart rate variability sequence in the same time window over multiple cycles, and the fluctuation amplitude of heart rate variability is obtained by calculating the daily standard deviation of the heart rate variability sequence. The baseline level of skin conductance is obtained by calculating the mean of the skin conductance signal over multiple time windows. The recovery speed of skin conductance is obtained by calculating the slope of the skin conductance signal from the peak level to the baseline level.

[0074] Emotional susceptibility is the degree to which an individual is sensitive to emotional perturbations within a specific time period. Emotional susceptibility is assessed by statistically analyzing the fluctuation range of the mean daytime alpha wave frequency and the mean daytime theta wave frequency in each time period.

[0075] Statistical analysis of the differences in the regularity of the distribution of intervention response scores over time, specifically comprising the following steps: dividing historical user samples into different biorhythm types according to the user rhythm bias value, extracting the intervention response scores from the historical intervention response data set, and taking one hour as a time slice, statistically analyzing the indicators of the intervention response scores of users with different biorhythm types in different time slices, including the mean and standard deviation of the intervention response scores in different time slices, and statistically analyzing the mean of the intervention response scores of users with different biorhythm types throughout the day in all time slices; generating a response deviation sequence by comparing and analyzing the degree of deviation of the intervention response scores of users with different rhythm types in each time slice from the mean of the intervention response scores throughout the day; performing a significance analysis on the response deviation sequence, and using a sliding window averaging method, identifying the time periods within a day when the response ability of the user's biorhythm type is significantly higher than the average level as the optimal candidate time periods for adjustment, and statistically analyzing all the optimal candidate time periods for adjustment to form a set of optimal candidate time periods for adjustment;

[0076] It should be noted that the mean of the intervention response scores of different time slices is used as a reference benchmark to measure the degree of deviation of each time slice and to construct a response deviation sequence; the standard deviation of the intervention response scores of different time slices is used to measure the fluctuation range of the intervention response scores throughout the day and to identify when the deviation of the mean of the intervention response scores of different time slices reaches a significant level;

[0077] Rhythm time period mapping table, example:

[0078] When the user's biorhythm type is identified as late, the optimal candidate time periods are adjusted to [11:30–13:30] and [16:00–18:30].

[0079] When the user's biorhythm type is identified as neutral, the optimal candidate time period set is adjusted to [10:00–12:00], [14:30–16:00];

[0080] When the user's biorhythm type is identified as being early morning, the set of preferred candidate time periods is adjusted to be [08:30–11:00] and [13:00–14:30].

[0081] Based on the set of optimal candidate time periods for regulation and the historical intervention response data set, an individual regulation time scoring function is constructed to obtain the individual highly sensitive emotion regulation time period;

[0082] It should be noted that the individual adjustment time scoring function is expressed as:

[0083]

[0084] Where G(t) is the individual adjustment time score at time t; i is the i-th time period; n is the number of adjustment preferred candidate time periods in the set of adjustment preferred candidate time periods; w i is the normalized weight of the historical intervention score, which is calculated by calculating the proportion of the intervention response score S1 in the i-th time period of the previous fixed cycle in the total of all historical intervention scores, w i The value range is [0,1]; t represents the current time; is the time membership function, only at t∈ΔT i 1 when ΔT is set, otherwise 0; i To adjust the i-th time period in the set of preferred candidate time periods, μ i is the average occurrence time of the high intervention response score obtained by the user in the i-th time period in history; σ is the individual rhythm response expansion parameter, which is set by analyzing the distribution width of the individual historical intervention response score in each time period and taking the half-height width of the average response peak, with a value range of [0.5, 3.0]; S1 is the historical intervention response score; j is each time period in the set of preferred candidate time periods for adjustment; ∑ j S j The sum of all historical intervention response scores of the user;

[0085] It should be noted that the individual adjustment time scoring function is the individual adjustment time score at time t, and based on the individual adjustment time scores at all times, a time matching function curve is formed in the adjustment preferred candidate time period;

[0086] Furthermore, in the time matching function curve, by performing a local maximum extraction operation on the individual adjustment time score, the peak point on the time matching function curve is selected, and filtering is performed according to the high-sensitivity time period threshold to obtain the time period in which the individual adjustment time score is higher than the high-sensitivity time period threshold, and the individual high-sensitivity emotion adjustment time period is obtained.

[0087] It should be noted that the high-sensitivity time period threshold is obtained by statistically analyzing the matching function value distribution of the high-response interval of the historical intervention response score. Usually, the upper quartile is in the interval [0.5, 0.8]. Therefore, the value range of the high-sensitivity time period threshold is [0.5, 0.8].

[0088] S2. Based on the individual's high-sensitivity emotion regulation time period, collect the user's real-time emotion index, predict the emotion perturbation index through the weighted time decay integral function, and identify the user's mild emotion fluctuation state;

[0089] Real-time emotion indicators include heart rate variability series, skin electrical change series, respiratory rhythm series, and resting respiratory rate average;

[0090] Based on the individual's high-sensitivity emotion regulation period, a continuous monitoring window is set to collect the user's real-time emotion indicators according to the physiological signal sampling frequency;

[0091] For example, if the individual rhythm bias value is late, the set of preferred candidate time periods includes the time interval of 15:00–17:00 in the afternoon, and 16:00 is the individual's highly sensitive emotional regulation time period. 16:00 is selected as the current data collection time point, the collection device is worn and the continuous detection window is started. The continuous monitoring window time is set to 5 minutes, and the physiological signal sampling frequency is 10 times per second. The physiological signals of the heart rate variability sequence, the skin electrode change sequence, and the respiratory rhythm sequence are collected in real time, and the average respiratory rate in the resting state is recorded at the same time;

[0092] It should be noted that the setting of the physiological signal sampling frequency needs to take into account the balance between signal accuracy and energy consumption of wearable devices.

[0093] Through the weighted time decay integral function, the emotional perturbation index is predicted and the user's mild emotional fluctuation state is identified;

[0094] It should be noted that the weighted time decay integral function is expressed as:

[0095]

[0096] in, is the emotional perturbation index, the unit is the dimensionless perturbation score; t represents the current moment; h(t) is the heart rate variability sequence at moment t; e(t) is the skin electrode change sequence at moment t; r(t) is the respiratory rhythm sequence at moment t; is the mean of resting respiratory rate; λ is the exponential decay factor; A is the length of a single detection window;

[0097] It should be noted that before using the weighted time decay integral function, h(t), e(t), and r(t) need to be normalized; the derivative of h(t) reflects the degree of heart rate instability; the derivative of e(t) reflects the instantaneous response strength of the sweat glands; the exponential decay factor is set according to the sensitivity of the physiological signal response to time mutations. The value of the exponential decay factor is to prevent the historical signal from being too small and causing excessive influence, while preventing the instantaneous fluctuation from being amplified and introducing noise due to being too large. The value range of the exponential decay factor is generally set to [0.01, 0.1]; the single detection window length is set according to the detectability of the emotional state change in a short time interval. The value of the single detection window length is to cover the shortest effective intervention feedback cycle and avoid the intervention response being diluted by the background signal. The value range of the single detection window length is generally set to: [60, 600]; the square term of the derivative of h(t) Indicates the intensity of heart rate fluctuations. The larger the value, the more unstable the user's mental state is; the square term of the derivative of the skin electrical change sequence It is an important indicator of stress response, and sudden changes in slope indicate potential mood swings; is the square term of respiratory interval deviation, which indicates the difference between the actual respiratory interval and the resting state, reflecting the rhythm change caused by anxiety or excitement; -λt The time decay weight is the data closer to the current moment, the higher the weight is. The total skin electrical energy term represents the overall skin electrical response intensity of an individual. It can correct the difference in sweating ability among different individuals and avoid individuals with high skin electrical activity being misjudged as having persistent emotional fluctuations. By adding "1" and then taking the square root, we can prevent the zero division error and provide a normalization suppression term to prevent the molecular result from changing dramatically with the deviation of the skin electrode baseline. The range of is [0,+∞);

[0098] Furthermore, when the emotional perturbation index at a certain time point in the individual's high-sensitivity emotional regulation period is greater than the perturbation trigger threshold, it is identified that the user is in a state of mild emotional fluctuation;

[0099] It should be noted that the disturbance trigger threshold is set based on the correlation distribution between the emotional perturbation index and the intervention effect in the historical intervention response data. The value of the disturbance trigger threshold is set to prevent it from being too low, resulting in frequent false triggering of intervention, and at the same time, it is set to prevent it from being too high, resulting in missing the individual's highly sensitive emotional regulation period that is weak but has regulatory value. The value range of the disturbance trigger threshold is generally set to [1.2, 2.0].

[0100] S3. Construct an embedded feature projection model to generate a set of intervention music parameters based on the user's mild emotional fluctuation state;

[0101] Collect the emotional perturbation index, rhythm deviation value, individual high-sensitivity emotional regulation period, previous round of intervention response score and individual type code to form a five-dimensional feature input vector;

[0102] It should be noted that the individual high-sensitivity emotion regulation period includes all time periods in the individual high-sensitivity emotion regulation period and the duration of each time period; the previous round of intervention response score is extracted from the historical intervention response data set;

[0103] For example, if the individual's high-sensitivity emotion regulation time period is 10:00-11:30, the duration of the time period is 90 minutes;

[0104] The individual type code is calculated based on the rhythm code and the high-sensitivity dimension code. The specific acquisition process is as follows:

[0105] The user's biorhythm type is distinguished based on the rhythm bias value. When the user's biorhythm type is identified as late, the rhythm code is 0; when the user's biorhythm type is identified as neutral, the rhythm code is 1; when the user's biorhythm type is identified as early, the rhythm code is 2;

[0106] Based on the user's mild emotional fluctuation state, the dominant frequency of the emotional perturbation index greater than the disturbance trigger threshold within a fixed period is counted, and the frequency of the dominant frequency affected by the real-time emotional index is analyzed to classify it. Specifically, when the user is in a mild emotional fluctuation state, when it is judged that the user is most affected by the heart rate variability sequence, the high-sensitivity dimension is coded as 0; when it is judged that the user is most affected by the skin electrode change sequence, the high-sensitivity dimension is coded as 1; when it is judged that the user is most affected by the respiratory interval deviation, the high-sensitivity dimension is coded as 2;

[0107] Based on the five-dimensional feature input vector, the first layer of hidden mapping is performed through the hyperbolic tangent function to obtain the hidden representation of the first stage;

[0108] Through linear transformation, the hidden representation of the first stage is mapped to a low-dimensional embedding space to obtain a compact embedding vector of the individual state;

[0109] Perform activation mapping on the compact embedding vectors of individual states to generate high-dimensional emotion regulation feature representations;

[0110] Based on the high-dimensional emotion regulation feature representation, a set of intervention music parameters is generated through high-dimensional feature transformation to form an embedded feature projection model;

[0111] Furthermore, a five-dimensional feature input vector is collected, the five-dimensional feature input vector is combined in a fixed order and standardized, and the standardized five-dimensional feature input vector is input into the first-layer hidden structure of the embedded feature projection model at one time. The first-layer hidden structure consists of a set of weight matrices and bias vectors, and the hyperbolic tangent function is used as a nonlinear activation function; the standardized five-dimensional feature input vector is weighted and activated to obtain the hidden representation of the first stage, which is used to capture the nonlinear relationship between the five-dimensional feature input vectors; the hidden representation of the first stage is input into the second set of weight structures, and the hidden representation of the first stage is compressed through linear mapping to obtain a compact embedding vector of the individual state; the compact embedding vector of the individual state is nonlinearly activated and input into the output structure, and a high-dimensional transformation is performed through the output weight matrix to obtain a set of intervention music parameters, including rhythm speed, mode tendency, timbre brightness and darkness, and intervention music duration;

[0112] Based on the training goal of minimizing the prediction error of the intervention response score, the embedded feature projection model is trained through supervised learning, and the embedded feature projection model is updated through the loss function until the embedded feature projection model reaches the set iteration round;

[0113] Furthermore, the training goal is to optimize the prediction error of the intervention response score; each set of training samples includes a five-dimensional feature input vector and an intervention response score of the five-dimensional feature input vector. After predicting the output, the embedded feature projection model calculates the error between the predicted value and the true value of the intervention response score, and adds a parameter regularization term; all weight matrices and bias vectors in the embedded feature projection model are extracted to form a parameter set to be optimized; after the embedded feature projection model initializes all parameters in the parameter set to be optimized, supervised learning is used for training; each round of training includes: generating an intervention music parameter set through forward propagation and feeding back the predicted value of the intervention response score; calculating and back-propagating the error through the loss function; finally, updating the parameters according to the optimization algorithm; the training process continues to iterate and gradually improves the prediction accuracy; when the prediction error on the validation set does not decrease significantly in multiple consecutive iterations, the training process is terminated, and all parameters in the model parameter set to be optimized are finalized;

[0114] It should be noted that the predicted value of the intervention response score is the estimated value of the intervention response score predicted and output by the embedded feature projection model based on the "five-dimensional feature input vector", and is the value generated by the embedded feature projection model through forward propagation during the training process; the true value of the intervention response score is the intervention response score in the historical intervention response data set; the validation set is sample data divided and extracted from the historical intervention response data set, which is used to evaluate the generalization ability and prediction stability of the model during the training process of the embedded feature projection model, prevent the embedded feature projection model from overfitting, and determine the termination time of training. It does not participate in the direct update of the parameters to be optimized in the embedded feature projection model;

[0115] It should be noted that the loss function is expressed as:

[0116]

[0117] Among them, L(Θ) is the loss function; Θ is the set of parameters to be optimized in the embedded feature projection model, including W f , b f , W e , b e , W Z , b z , W o ; N is the number of samples; S2 is the actual intervention response score of the i-th sample under the intervention music parameter set; The intervention response score predicted by the embedded feature projection model; M i is the individual type code of the i-th sample; τ is the regularization factor, which is used to control the complexity of the model and avoid overfitting; The L2 norm squared is used to prevent overfitting of the embedded feature projection model and to avoid excessive parameter weights leading to decreased generalization ability. It is a standard structural regularization.

[0118] It should be noted that before using the loss function, S2 and Perform normalization processing, For the i-th sample in type M i Under the condition of , the intervention response score predicted by the embedded feature projection model is one of the terminal output results of the embedded feature projection model structure, which is obtained by fitting the actual intervention response score S2. The specific process is as follows: after inputting the five-dimensional feature vector into the embedded feature projection model, the intervention music parameter set is obtained through the forward reasoning of the embedded feature projection model; based on the intervention music parameter set and the current individual type code, the intervention response score predicted by the embedded feature projection model is calculated through the intervention response score prediction function. The intervention response score prediction function is obtained by supervised learning of the real S2 label in the training data;

[0119] The intervention response score prediction function is used to predict the possible intervention response score of this round of intervention based on the given intervention music parameter set and individual type code. In the training stage, known input-output pairs are first extracted from the historical intervention response data set: the input of the input-output pair is the intervention music parameter set and the individual type code, and the output of the input-output pair is the actual collected intervention response score; the input-output pair is used as the basic sample for supervised learning, and for each set of inputs, the mapping relationship between the fitting and the output results is determined. Specifically, a set of initial mapping relationships initialized based on the weighting function form is selected; by minimizing the error between the predicted value and the true score, the parameter weights of the weighting function are gradually corrected, and the mapping relationship is continuously updated to achieve better prediction accuracy on the entire training data set; when the error between the predicted value and the true score converges, the mapping relationship function is the intervention response score prediction function;

[0120] W f is the weight matrix of the first layer hidden mapping, which is used to map the five-dimensional feature input vector to the hidden representation of the first stage through linear transformation and nonlinear activation; b f is the first layer bias vector; W e is the weight matrix that maps the hidden representation from the first stage to the low-dimensional embedding space, which is used to learn the transformation from the nonlinear latent feature space to the compact expression of the individual state; b e is the bias vector that maps the hidden representation from the first stage to the low-dimensional embedding space; W Z is the weight matrix of high-dimensional feature transformation, which is used to transform the low-dimensional embedding vector into a high-dimensional emotion regulation feature representation through linear transformation; W o To map the high-dimensional emotion regulation feature representation into an output weight matrix of the set of intervention music parameters;

[0121] The set of intervention music parameters includes: tempo, tonality, timbre, and duration;

[0122] Among them, tempo is the rhythm speed, which is used to control the rhythm of music and affect heart rate synchronization;

[0123] Tonality is a modal tendency, which is used to map the seven natural modes in traditional Western music and control the emotional tendency of music;

[0124] Timbre is the brightness of the timbre, which is used to control the bright and soft listening experience;

[0125] Duration is the duration of the intervention music, which is used to control the intensity and duration of the intervention.

[0126] S4. Play the intervention music according to the intervention music parameter set, collect feedback signals in real time, and calculate the intervention response score by constructing an intervention response score function.

[0127] Feedback signals include HRV curve and skin electrical change rate.

[0128] Furthermore, based on the tempo value, intervention music tracks are screened to ensure that the intervention music's rhythm is coordinated with the user's current heart rate fluctuations. Based on the tonality value, through seven-tone mapping, intervention music categories that match the user's emotional tonality are selected, and debugging clips are selected. Based on the timbre value, intervention music with the closest brightness and softness characteristics in the tone spectrum is matched to optimize the user's subjective listening experience. Based on the duration value, the duration of the intervention music is selected to ensure that the intervention intensity matches the duration of the intervention music.

[0129] It should be noted that HRV stands for heart rate variability. The pulse interval method is used to obtain an HRV curve composed of heart rate variability. The HRV curve reflects the recovery dynamics of the sympathetic and parasympathetic nervous systems after music intervention and is used to measure the degree of heart rate stability and parasympathetic activation trends. The skin conductance change rate is obtained by real-time acquisition of skin conductance signals and derivative processing. The skin conductance change rate reflects the rate of decline in autonomic nervous system activity after music stimulation and serves as a sensitive indicator of relaxation response.

[0130] The intervention response score function is constructed by performing supervised fitting between the five-dimensional feature input vector and the actual intervention response score. The expression is:

[0131]

[0132] Where S is the intervention response score, which is used to measure the regulatory effect of this round of intervention; X is the total music playback time, F(t) is the heart rate variability index at time t, and K(t) is the skin electrode change rate at time t. The higher the K(t) value, the more significant the user's relaxation trend.

[0133] It should be noted that F(t) and K(t) were normalized before using the intervention response score function;

[0134] The larger the value of S is, the better the effect of this round of intervention music on regulating the individual's emotional state is.

[0135] S5. Based on the intervention data group, a historical intervention response data set is constructed, and an intervention music recommendation strategy is generated through the intervention scenario parameter mapping method.

[0136] Based on the intervention time point, rhythm deviation value, individual high-sensitivity emotion regulation period, emotion perturbation index, intervention music parameter set and intervention response score, multiple intervention data groups are formed;

[0137] By statistically analyzing multiple intervention data groups in chronological order, a historical intervention response data set is constructed;

[0138] Furthermore, during the execution of multiple independent interventions, the key parameters of each intervention were recorded in sequence to generate multiple intervention data sets. Each intervention data set includes the current intervention time point, rhythm deviation value, individual high-sensitivity emotion regulation period, real-time emotional perturbation index, intervention music parameter set used in this round, and intervention response score. The collected intervention data sets were preliminarily sorted according to the chronological order of the intervention to ensure the organizational continuity of the data on the time axis. Each intervention data set was screened and processed to exclude data sets with missing key parameters, including failure to successfully collect response scores and missing intervention music parameters. Data sets with collection anomalies and too short monitoring periods were marked as invalid samples and eliminated. Collection anomalies included severe HRV noise pollution and loss of skin electrodermal signals. The retained intervention data sets were further subjected to feature standardization processing, and all parameters used for subsequent calculations were uniformly converted to a comparable numerical scale range to eliminate the offset effect of parameters from different sources on the analysis results.

[0139] Based on the historical intervention response data set, a positive intervention response threshold is set, and the intervention data group with intervention response scores higher than the positive intervention response threshold is selected as the positive feedback sample set;

[0140] Furthermore, a positive intervention response threshold is set, the historical intervention response data set is traversed, and the intervention response score in each intervention data group is compared with the positive intervention response threshold one by one; for data groups with intervention response scores higher than the positive intervention response threshold, they are judged as positive feedback samples, and all positive feedback samples are counted to form a positive feedback sample set.

[0141] It should be noted that the positive intervention response threshold is set based on the upper quartile of the statistical characteristics of the distribution of historical intervention response scores and is a real number ranging from 1 to 2;

[0142] The intervention scenario vector is constructed by combining the rhythm deviation value, individual high-sensitivity emotion regulation period and emotion perturbation index;

[0143] A multidimensional clustering algorithm is used to perform cluster analysis on the intervention music parameter set and intervention context vector in the positive feedback sample set, construct an intervention response category set, and calculate the cluster center of each intervention response category in the intervention response category set;

[0144] It should be noted that the intervention response category set is constructed and the cluster center of each intervention response category in the intervention response category set is calculated. The specific steps are:

[0145] Intervention data groups with intervention response scores greater than the positive intervention response threshold are screened out from the historical intervention response data set as the positive feedback sample set; the intervention music parameter set and intervention scenario vector are extracted from the positive feedback sample set to form an intervention parameter scenario pair set, and feature normalization is performed to unify the different feature dimensions in the clustering process; a multidimensional clustering algorithm is used to obtain the intervention response category set by clustering the intervention parameter scenario pair set; the mean of the samples in each intervention response category is calculated to obtain the cluster center of the intervention music parameter set and the cluster center of the intervention scenario vector.

[0146] Based on each cluster center in the intervention response category set, an inverse mapping function from the intervention scenario vector to the intervention music parameter set is constructed, and an intervention music recommendation strategy is generated.

[0147] Furthermore, the intervention scenario vector cluster center of each cluster category and the intervention music parameter set cluster center are formed into input-output pairs, and the input-output pairs are used as discrete sample points for function mapping to construct a parameter mapping relationship table. Each row of the parameter mapping relationship table corresponds to an intervention response category, and the intervention scenario vector cluster center of the current intervention response category is recorded as input and the intervention music parameter set cluster center is recorded as output. A multidimensional interpolation algorithm is used to numerically approximate the discrete sample points, and a set of continuous mapping functions from the intervention scenario vector space to the intervention music parameter space is fitted. The obtained continuous mapping function is defined as the intervention music recommendation strategy, which is used in subsequent interventions to generate an intervention music parameter set that matches the current scenario vector in real time.

[0148] This embodiment also provides a personalized emotion regulation system based on music therapy, comprising: a rhythm recognition module, an emotion recognition module, an intervention generation module, an intervention execution module, and a strategy training module;

[0149] The rhythm recognition module is used to collect and pre-process the user's life rhythm parameters, calculate the rhythm deviation value, identify the user's biorhythm type, and obtain the individual's highly sensitive emotional regulation time period;

[0150] The emotion recognition module is used to collect real-time user emotion indicators based on the individual's high-sensitivity emotion regulation time period, predict the emotion perturbation index through the weighted time decay integral function, and identify the user's mild emotion fluctuation state;

[0151] The intervention generation module is used to construct an embedded feature projection model and generate a set of intervention music parameters based on the user's mild emotional fluctuation state;

[0152] The intervention execution module is used to play the intervention music according to the intervention music parameter set, collect feedback signals in real time, and calculate the intervention response score by constructing the intervention response score function;

[0153] The strategy training module is used to construct a historical intervention response data set based on the intervention data group, and generate an intervention music recommendation strategy through the intervention scenario parameter mapping method.

[0154] This embodiment also provides a computer device suitable for the personalized emotion regulation method based on music therapy, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute computer-executable instructions to implement the personalized emotion regulation method based on music therapy proposed in the above embodiment.

[0155] The computer device may be a terminal, comprising a processor, a memory, a communication interface, a display screen and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device comprises a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner may be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display or an electronic ink display screen, and the input device of the computer device may be a touch layer covering the display screen, or a button, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse.

[0156] This embodiment also provides a storage medium having a computer program stored thereon, which, when executed by a processor, implements the personalized emotion regulation method based on music therapy as proposed in the above embodiment; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0157] In summary, the present invention generates an intervention music recommendation strategy through: an intervention scenario parameter mapping method, clusters historical positive intervention samples into intervention response categories, and constructs an inverse mapping between intervention scenarios and intervention parameters based on cluster centers, thereby dynamically optimizing the recommendation strategy and improving the adaptability of subsequent interventions.

[0158] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A personalized emotion regulation method based on music therapy, characterized by: include, Collect and pre-process the user's life rhythm parameters, calculate the rhythm deviation value, identify the user's biorhythm type, and obtain the individual's highly sensitive emotional regulation time period; Based on the individual's high-sensitivity emotion regulation period, the user's real-time emotion indicators are collected, and the emotional perturbation index is predicted through the weighted time decay integral function, and the user's mild emotional fluctuation state is identified; Construct an embedded feature projection model to generate a set of intervention music parameters based on the user's mild emotional fluctuation state; Play the intervention music according to the intervention music parameter set, collect feedback signals in real time, and calculate the intervention response score by constructing the intervention response score function; Based on the intervention data group, a historical intervention response data set is constructed, and an intervention music recommendation strategy is generated through the intervention scenario parameter mapping method.

2. The personalized emotion regulation method based on music therapy according to claim 1, characterized in that: The life rhythm parameters include sleep-wake time series, morning heart rate average, nighttime skin electrical peak, daytime alpha wave frequency average, and daytime theta wave frequency average; The preprocessing includes data cleaning, standardization and normalization.

3. The personalized emotion regulation method based on music therapy according to claim 2, characterized in that: The steps of calculating the rhythm deviation value and identifying the user's biorhythm type are as follows: A logarithmic weighted structure with dual ratio modulation was used to construct a rhythm deviation evaluation function and calculate the rhythm deviation value. Based on the rhythm deviation value and according to the rhythm deviation type identification standard, the user's biorhythm type is identified.

4. The personalized emotion regulation method based on music therapy according to claim 3, characterized in that: The steps for obtaining the individual's high-sensitivity emotion regulation time period are as follows: Based on the historical differences in the regularity of the distribution of neural regulation ability, emotional susceptibility, and intervention response scores of people with different biorhythm types over time, a rhythm time period mapping table is defined, and a set of optimal candidate time periods for regulation is selected based on the user's biorhythm type; Based on the set of optimal candidate time periods for regulation and the historical intervention response data set, an individual regulation time scoring function is constructed to obtain the individual highly sensitive emotion regulation time period.

5. The personalized emotion regulation method based on music therapy according to claim 4, characterized in that: The real-time emotion indicators include heart rate variability sequence, skin electrical change sequence, respiratory rhythm sequence and resting respiratory rate mean.

6. The personalized emotion regulation method based on music therapy according to claim 5, characterized in that: The steps of constructing the embedded feature projection model are as follows: Collect the emotional perturbation index, rhythm deviation value, individual high-sensitivity emotional regulation period, previous round of intervention response score and individual type code to form a five-dimensional feature input vector; Based on the five-dimensional feature input vector, the first layer of hidden mapping is performed through the hyperbolic tangent function to obtain the hidden representation of the first stage; Through linear transformation, the hidden representation of the first stage is mapped to a low-dimensional embedding space to obtain a compact embedding vector of the individual state; Perform activation mapping on the compact embedding vectors of individual states to generate high-dimensional emotion regulation feature representations; Based on the high-dimensional emotion regulation feature representation, a set of intervention music parameters is generated through high-dimensional feature transformation to form an embedded feature projection model; Based on the training goal of minimizing the prediction error of the intervention response score, the embedded feature projection model is trained through supervised learning, and the embedded feature projection model is updated through the loss function until the embedded feature projection model reaches the set iteration rounds.

7. The personalized emotion regulation method based on music therapy according to claim 6, characterized in that: The feedback signal includes an HRV curve and a skin electrical change rate.

8. The personalized emotion regulation method based on music therapy according to claim 7, characterized in that: The steps of constructing a historical intervention response data set based on the intervention data set are as follows: Based on the intervention time point, rhythm deviation value, individual high-sensitivity emotion regulation period, emotion perturbation index, intervention music parameter set and intervention response score, multiple intervention data groups are formed; By statistically analyzing multiple intervention data groups in chronological order, a historical intervention response data set is constructed.

9. The personalized emotion regulation method based on music therapy according to claim 8, characterized in that: The intervention music recommendation strategy is generated by the intervention scenario parameter mapping method, and the steps are as follows: Based on the historical intervention response data set, a positive intervention response threshold is set, and the intervention data group with intervention response scores higher than the positive intervention response threshold is selected as the positive feedback sample set; The intervention scenario vector is constructed by combining the rhythm deviation value, individual high-sensitivity emotion regulation period and emotion perturbation index; A multidimensional clustering algorithm is used to perform cluster analysis on the intervention music parameter set and intervention context vector in the positive feedback sample set, construct an intervention response category set, and calculate the cluster center of each intervention response category in the intervention response category set; Based on each cluster center in the intervention response category set, an inverse mapping function from the intervention scenario vector to the intervention music parameter set is constructed, and an intervention music recommendation strategy is generated.

10. A personalized emotion regulation system based on music therapy, based on the personalized emotion regulation method based on music therapy according to any one of claims 1 to 9, characterized in that: include, The rhythm recognition module is used to collect and pre-process the user's life rhythm parameters, calculate the rhythm deviation value, identify the user's biorhythm type, and obtain the individual's highly sensitive emotional regulation time period; The emotion recognition module is used to collect real-time user emotion indicators based on the individual's high-sensitivity emotion regulation time period, predict the emotion perturbation index through the weighted time decay integral function, and identify the user's mild emotion fluctuation state; The intervention generation module is used to construct an embedded feature projection model and generate a set of intervention music parameters based on the user's mild emotional fluctuation state; The intervention execution module is used to play the intervention music according to the intervention music parameter set, collect feedback signals in real time, and calculate the intervention response score by constructing the intervention response score function; The strategy training module is used to construct a historical intervention response data set based on the intervention data group, and generate an intervention music recommendation strategy through the intervention scenario parameter mapping method.

Images