Intelligent virtuality and reality combined mental health service device based on digital elements

Abstract

The invention provides an intelligent virtuality and reality combined psychological health service device based on digital elements, and aims to improve the accuracy and individuation level of psychological health management. The device firstly obtains the physiological indexes, behavior data, environmental factor data and social economic data of an individual, and carries out multi-modal fusion to form comprehensive feature data. Based on this, a psychological health environment factor model is constructed to quantify the influence of the external environment on the individual psychological state, and the calculation weight and prediction logic of the negative emotion large model are optimized. And the optimized negative emotion large model is used for identifying an individual emotion state, analyzing factors such as social environment, economic pressure and life events in combination with the mental health environment factor model, and generating an individual mental health assessment result. And according to an evaluation result, the virtual digital doctor provides intelligent pre-inquiry and other services. Through data-driven intelligent analysis and virtual-real combined intervention means, the accessibility, accuracy and intervention effect of psychological health services are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent health services, and particularly relates to an intelligent virtual-real combined psychological health service device based on digital elements. BACKGROUND

[0002] With the increase of social pressure, mental health problems have attracted increasing attention, and mental health service models are also developing. At present, mental health intervention mainly relies on traditional offline counseling, hospital diagnosis and treatment, and part of online psychological assessment tools. Some medical institutions and health management platforms have introduced artificial intelligence technology to provide emotion recognition, psychological assessment and remote counseling services, and health prediction models based on big data have also been used for auxiliary diagnosis and risk assessment. These technologies still mainly rely on structured questionnaire surveys and limited physiological data collection.

[0003] The existing mental health service model has many limitations. First, traditional offline services are subject to the shortage of professional resources and cannot meet the large-scale personalized needs, while the accuracy of online assessment is limited and it is difficult to provide precise intervention. Second, the current mental health management methods lack comprehensive consideration of external environmental factors (such as social pressure, economic conditions, life events, etc.), resulting in insufficient accuracy of mental disorder prediction. In addition, the existing AI auxiliary system mainly relies on static data analysis and lacks dynamic feedback mechanism, making it difficult to achieve long-term health management and individualized adjustment, affecting the intervention effect.

[0004] In view of the above problems, an intelligent psychological health service device that can combine multi-source data, dynamically monitor emotional state, and comprehensively analyze the influence of individual external environment is urgently needed to improve the accuracy and intervention effect of mental health management. SUMMARY

[0005] The present application provides an intelligent virtual-real combined psychological health service device based on digital elements to improve the accuracy and individualization level of mental health management.

[0006] The present application provides an intelligent virtual-real combined psychological health service device based on digital elements, comprising: An acquisition unit is configured to acquire individual mental health related data including physiological indicators, behavior data, environmental factor data and socio-economic data, and perform multi-modal fusion to form comprehensive feature data; A construction unit is configured to construct a mental health environmental factor model based on the comprehensive feature data to quantify the influence of external environment on individual mental state and use it as an input parameter to optimize the calculation weight and prediction logic of the negative emotion large model; The identification unit is used to identify an individual's emotional state using an optimized negative emotion model, and combines it with a mental health environmental factor model to analyze the individual's social environment, economic pressure, and life events to predict the risk of mental disorders and generate individual mental health assessment results. The unit provides a personalized mental health management plan by a virtual digital doctor based on the individual's mental health assessment results, including intelligent pre-diagnosis, emotion regulation suggestions, intervention measures, and long-term health tracking services; The adjustment unit is used to combine AI emotion recognition virtual human companionship service, utilize non-contact physiological indicator collection and multimodal emotion recognition technology to monitor the user's emotional state in real time, and dynamically adjust the interactive content of AI emotion recognition virtual human based on the analysis results of the mental health environmental factor model to provide personalized psychological support.

[0007] The beneficial effects of the technical solution provided in this application include: (1) By integrating multimodal data, we comprehensively analyze an individual's physiological indicators, behavioral data, environmental factors, and socioeconomic status. Combined with the mental health environmental factor model, we optimize the negative emotion model, making emotion recognition and mental disorder risk prediction more accurate, effectively reducing the misjudgment rate, and improving the reliability of mental health assessment. (2) Relying on virtual digital doctors, we provide intelligent pre-diagnosis, emotion regulation suggestions, and intervention plans based on individual mental health assessment results. Combined with long-term health tracking services, we dynamically adjust intervention strategies to make mental health management more targeted and meet the personalized needs of different users. Attached Figure Description

[0008] Figure 1 This is a schematic diagram of an intelligent virtual-real integrated mental health service device based on digital elements provided in the first embodiment of this application. Detailed Implementation

[0009] Many specific details are set forth in the following description to provide a full understanding of this application. However, this application can be implemented in many other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of this application; therefore, this application is not limited to the specific embodiments disclosed below.

[0010] The first embodiment of this application provides an intelligent virtual-real integrated mental health service device based on digital elements. Please refer to... Figure 1 This figure is a schematic diagram of the first embodiment of this application. The following is in conjunction with... Figure 1 The first embodiment of this application provides a detailed description of an intelligent virtual-real integrated mental health service device based on digital elements.

[0011] Acquisition Unit 101: Used to acquire individual mental health-related data, including physiological indicators, behavioral data, environmental factor data, and socioeconomic data, and to perform multimodal fusion to form comprehensive feature data.

[0012] The data acquisition phase first requires acquiring individual mental health-related data, which can be obtained from sources including but not limited to physiological indicators, behavioral data, environmental factors, and socioeconomic data. To ensure data integrity and usability, data can be collected through wearable devices, mobile applications, smart sensors, electronic medical records, online questionnaires, and third-party databases.

[0013] The collection of physiological data primarily involves monitoring an individual's physiological state, including but not limited to heart rate, heart rate variability (HRV), blood oxygen saturation, electrical skin response (EDA), skin temperature, blood pressure, and electroencephalogram (EEG). This physiological data can be collected using devices such as smartwatches, chest strap heart rate monitors, finger clip pulse oximeters, and EEG sensors. Combining high-frequency sampling with low-frequency data fusion can improve data accuracy and stability. During data acquisition, noise filtering algorithms based on time-series analysis can be employed to reduce signal interference and improve data quality. Furthermore, feature extraction methods, such as Fast Fourier Transform (FFT) or wavelet transform, can be used to extract key physiological parameter features to support subsequent analysis.

[0014] Behavioral data collection involves an individual's daily activity patterns and lifestyle habits, including but not limited to sleep quality, exercise frequency, social interaction, duration of electronic device use, voice emotion characteristics, keyboard input patterns, and gait analysis. This data can be collected via smartphones, smart bracelets, fitness trackers, and social media analytics tools. Sleep data can be recorded using a combination of accelerometers and heart rate monitors, and processed using sleep staging algorithms (such as multivariate Bayesian inference methods) to identify deep sleep, REM sleep, and wakefulness. Voice emotion data can be extracted using natural language processing (NLP) techniques combined with speech spectral analysis to extract features such as tone, speech rate, and pitch variations, thereby assessing the user's emotional trends.

[0015] The collection of environmental data primarily involves an individual's physical and social environment, including but not limited to weather conditions, air quality, light intensity, noise levels, geographical location information, and office environment conditions. Weather conditions can be obtained through a meteorological API interface, including data such as temperature, humidity, air pressure, and rainfall; air quality data can be obtained from government or third-party environmental monitoring agencies, covering information such as PM2.5, PM10, carbon dioxide (CO2), and volatile organic compound (VOC) concentrations; noise levels can be measured in decibels using environmental sensors, and combined with speech recognition algorithms to distinguish between environmental noise and social noise, in order to assess the impact of the external environment on an individual's mental health.

[0016] The collection of socioeconomic data involves information on an individual's social interactions, economic stress, and life events, including but not limited to personal income levels, family economic status, occupational stress, social networks, mental health records, and major life events (such as job changes, family upheavals, etc.). This type of data can be obtained through user-authorized access to financial transaction records, social media interaction data, and online questionnaires. It can also be structured using machine learning models (such as classification models based on gradient boosting decision trees) to establish socioeconomic stress assessment indicators for users.

[0017] After data acquisition, multimodal fusion is required to form comprehensive feature data. Data fusion methods may include time series alignment, data standardization, feature selection, and dimensionality reduction. Time series alignment can employ dynamic time warping algorithms to match timestamps from different sources, improving data synchronization. Data standardization can use z-score normalization or Min-Max normalization methods to eliminate the influence between different measurement units. Feature selection can utilize principal component analysis or maximum correlation minimum redundancy (mRMR) algorithms to screen the most discriminative mental health features, reduce redundant information, and improve computational efficiency. Dimensionality reduction can employ autoencoders or t-SNE dimensionality reduction methods to compress data dimensions while retaining key features.

[0018] Ultimately, the fused and integrated feature data will be used for subsequent mental health status analysis, providing high-quality input data for the construction of a mental health environmental factor model. This step ensures the integrity, accuracy, and usability of the data, laying a solid foundation for the implementation of the entire intelligent mental health service device.

[0019] In the data acquisition unit, to accurately identify an individual's mental health status, continuous sampling of the user's physiological indicators is first performed, and key nodes of emotional changes are identified based on an abnormal fluctuation detection mechanism. The collection of physiological indicators covers multiple dimensions, including but not limited to heart rate, heart rate variability, skin conductance, blood oxygen saturation, voice characteristics, and facial muscle activity. This data can be collected through smart wearable devices, smartphone microphones, cameras, or other contactless sensors, and multimodal fusion technology is employed to ensure data accuracy and timeliness.

[0020] To improve the sensitivity of emotion change detection, the acquisition unit employs an abnormal fluctuation detection mechanism to determine whether an individual's current physiological data deviates from their normal state. In practice, the acquisition unit first establishes a physiological state baseline based on the individual's historical data and sets a reasonable fluctuation threshold. If a physiological parameter (such as heart rate) exceeds a preset range within a short period, or if multiple physiological signals (such as skin conductance and voice features) simultaneously exhibit abnormal changes, the abnormal fluctuation detection mechanism is triggered. The identification of abnormal fluctuations is based not only on absolute numerical changes but also incorporates time-series analysis methods to calculate the short-term and long-term trends of the individual's physiological state, thus eliminating the interference of occasional fluctuations. For example, if a user's heart rate temporarily increases without a significant change in skin conductance, the device can identify that this situation may be related to exercise or activity rather than emotional fluctuations, thereby avoiding false alarms.

[0021] Upon detecting potential emotional changes, the acquisition unit further employs an adaptive sampling strategy to adjust the data acquisition frequency. The core of this strategy lies in balancing sampling accuracy with device power consumption, ensuring increased data resolution when needed and reduced sampling intensity when the individual's state is stable, thereby minimizing redundant data and impacting user privacy. Specifically, when abnormal fluctuations in heart rate, skin conductance, or voice features are detected, the acquisition unit automatically increases the data sampling rate, shortening the sampling interval to acquire more refined physiological signals. For example, under normal circumstances, the acquisition unit might record the user's heart rate once per second, but after recognizing physiological signals of anxiety or excitement, it can increase the sampling frequency to five times per second to capture more detailed changes in physiological state. Similarly, during voice data acquisition, if a rapid tone, increased volume, or intensified voice tremor is detected, the acquisition unit can increase the analysis density of voice features to determine whether the change is related to emotional agitation.

[0022] On the other hand, once an individual's physiological state stabilizes, the acquisition unit will appropriately reduce the sampling frequency to minimize data redundancy, optimize the use of computing resources, and reduce the impact on user privacy. For example, when the device detects that an individual's physiological indicators gradually return to baseline and remain stable within a set time window, it automatically restores the sampling frequency to the default level, avoiding resource consumption caused by continuous high-frequency sampling. Furthermore, the device dynamically adjusts the sampling strategy based on the individual's daily behavioral patterns. For instance, the sampling frequency can be appropriately reduced at night or when the user is at rest, while the data collection frequency can be increased in advance when the user enters a social setting or faces potential stressors (such as exams or interviews) to better track the individual's emotional changes.

[0023] Construction Unit 102: Used to construct a mental health environmental factor model based on the comprehensive feature data, so as to quantify the influence of the external environment on the individual's mental state, and use it as input parameters to optimize the calculation weight and prediction logic of the negative emotion big model.

[0024] The building unit specifically includes: The processing module is used to acquire an individual's physiological indicators, behavioral data, environmental factor data, and socioeconomic data, and to normalize different types of data through adaptive multimodal fusion technology. The module is used to construct a mental health environmental factor model using time series analysis and deep learning methods. The mental health environmental factor model establishes the influence weights of individual mental state based on long-term and short-term environmental variables. Long-term environmental variables include air quality, urban noise level, and socioeconomic pressure in the individual's long-term living environment. Short-term environmental variables include temperature, light intensity, and frequency of social interaction on the day. The combined influence value of the long-term and short-term variables on the current mental state is calculated through an attention mechanism. The adjustment module is used to dynamically adjust the calculation weights of the negative emotion model based on the calculation results of the mental health environmental factors model, so that it can adapt to the individual's environment when performing individual emotion recognition. The learning module is used during the optimization process to utilize transfer learning methods, enabling the large negative emotion model to learn individual emotional patterns based on the user's historical emotional change data and dynamic adjustments to environmental factors.

[0025] The construction module requires building a mental health environmental factor model based on comprehensive feature data to quantify the impact of the external environment on individual mental states. This model is then used to optimize the computational weights and prediction logic of the large-scale negative emotion model. The entire process involves data analysis, model training, parameter optimization, and model fusion, enabling the mental health environmental factor model to capture the potential impact of external factors on individual mental health and integrate these impacts into the large-scale negative emotion model to improve the accuracy of emotion recognition and mental disorder risk prediction.

[0026] The input to mental health environmental factor models comes from comprehensive feature data, including physiological indicators, behavioral patterns, environmental factors, and socioeconomic status. Physiological indicators may include data such as heart rate, blood oxygen saturation, and skin conductance, which can be collected through wearable devices. Behavioral patterns include sleep quality, exercise habits, social activities, and electronic device usage time, reflecting an individual's daily state. Environmental factors encompass air quality, temperature, noise levels, and light intensity, used to assess the impact of the external environment on mental health. Socioeconomic status involves income level, occupational stress, social networks, and significant life events, factors that can have long-term effects on an individual's mood fluctuations. The role of mental health environmental factor models is to extract key features from this data and calculate their degree of impact on mental health. For example, regression analysis can quantify the contribution of a particular environmental factor to anxiety levels, or neural networks can be used to learn complex interactions to predict trends in an individual's mood changes in a specific environment.

[0027] The output of the mental health environmental factors model is the weights of environmental influencing factors, which are used to optimize the computational logic of the negative emotion model. The inputs to the negative emotion model include an individual's physiological and behavioral data, historical emotional records, and environmental influencing factors provided by the mental health environmental factors model. The core objective of the negative emotion model is to identify an individual's current emotional state and predict the risk of developing mental disorders. This is typically achieved using deep learning methods; for example, Long Short-Term Memory (LSTM) networks can analyze time-series data to capture the changing trends of an individual's emotions over time, while Transformer-based models can effectively combine different types of input data to improve the accuracy of emotion recognition. By introducing the environmental influencing weights provided by the mental health environmental factors model, the negative emotion model can dynamically adjust its computational logic under different environmental conditions. For example, if a person's sleep quality declines and their surroundings are noisy, the model will automatically lower the threshold for identifying anxiety states, making it easier to detect the individual's anxiety tendency and improving the sensitivity of predictions.

[0028] The output of the negative emotion model includes the current emotional state, the risk of developing a mental health disorder, and the trend of emotional change. The current emotional state can be categorized as normal, anxious, or depressed, while the risk of developing a mental health disorder is presented as a probability value, such as a user's 65% risk of depression. The trend of emotional change is based on an individual's historical data, predicting the direction of emotional development over a future period. If the mental health environmental factors model analyzes that a user has recently experienced a major negative life event and has reduced social activity, the calculation logic of the negative emotion model will be adjusted accordingly to increase the predicted value of depression risk, ensuring that the device can detect potential mental health problems earlier.

[0029] In practical implementation, a joint training method can be used, allowing the large-scale negative emotion model to automatically learn the output of the mental health environmental factors model during training. For example, an attention mechanism can be added to the neural network architecture, enabling it to dynamically consider environmental influencing factors when predicting an individual's emotional state. Another approach is to adjust the model parameters through reinforcement learning, ensuring optimal predictive ability under different environmental conditions. For instance, if a user has a high work stress index, the device can adjust the anxiety detection threshold, allowing it to identify anxiety risk at an earlier stage and provide more timely intervention suggestions.

[0030] In practical applications, the synergistic optimization of the mental health environmental factors model and the negative emotion big data model can significantly improve the accuracy of emotion recognition. For example, if a user's heart rate variability decreases, their activity level decreases, and their voice tone becomes lower, the negative emotion big data model initially assesses their risk of depression at 50%. Simultaneously, the mental health environmental factors model analyzes that the user recently experienced unemployment and reduced social interaction; these factors, after being converted into influencing factors, adjust the risk of depression to 75%. Based on this optimized prediction, the device can send intervention signals to virtual digital doctors or AI emotion recognition avatars earlier, suggesting users conduct psychological self-tests or providing relaxation training programs, thereby reducing the risk of developing mental disorders.

[0031] The entire optimization process ensures that the mental health environmental factors model is not merely an auxiliary analytical tool, but rather directly influences the computational logic of the larger negative emotion model, making emotion recognition and risk prediction more aligned with an individual's true psychological state. In this way, the device can provide more targeted and personalized mental health management solutions, improve the intelligence level of mental health services, and enable users to receive effective psychological intervention at an early stage.

[0032] In this embodiment, to achieve a more accurate assessment of mental health status, it is first necessary to construct a mental health environmental factor model to quantify the impact of the external environment on an individual's mental state, and then use this model to optimize the computational weights and prediction logic of the negative emotion model. This process involves various data collection, fusion, modeling, and optimization strategies to ensure the accuracy and personalization of the mental health assessment.

[0033] During the data acquisition phase, the device collects individual physiological indicators, behavioral data, environmental factor data, and socioeconomic data. These data sources are diverse, including physiological signals recorded by wearable devices, such as heart rate, heart rate variability, skin conductance, blood oxygen levels, and body temperature; behavioral data, such as sleep quality, exercise habits, social interactions, electronic device usage time, vocal emotion characteristics, and gait patterns; environmental factor data, including air quality, urban noise levels, temperature, humidity, and light intensity; and socioeconomic data, such as income levels, occupational stress, social relationships, life events, and economic burden. To ensure that this data can be processed uniformly, the device employs adaptive multimodal fusion technology to normalize different types of data, eliminating inconsistencies caused by differences in data sources, units, and measurement methods. Normalization methods can employ z-score standardization, Min-Max normalization, or latent variable mapping based on variational autoencoders to ensure data comparability in subsequent calculations.

[0034] In the modeling process, the device utilizes time series analysis and deep learning methods to construct a model of environmental factors affecting mental health. The core of this model lies in distinguishing the impact of long-term and short-term environmental variables and establishing corresponding weights. Long-term environmental variables typically include air quality, urban noise levels, and socioeconomic stress in an individual's long-term residence. These factors have a relatively stable and far-reaching impact on an individual's mental health, thus requiring trend analysis using historical data to extract long-term impact patterns. Short-term environmental variables include daily temperature, light intensity, and frequency of social interaction. These factors have a greater impact on the immediate fluctuations in an individual's mental state, requiring the model to adapt quickly. For example, if a user is in a high-noise environment for a long time, their baseline anxiety level may be high; however, if the user experiences an abnormally high air pollution index on a particular day, their short-term anxiety may be further exacerbated. To effectively calculate the combined impact of long-term and short-term variables on an individual's mental state, the device employs an attention mechanism, calculating the importance of each variable through weighted calculations, allowing it to dynamically adjust its impact weight on the individual's mental state as the environment changes.

[0035] The adjustment module is used to dynamically adjust the calculation weights of the negative emotion model based on the calculation results of the mental health environmental factors model, so that it can adapt to the individual's environment when performing individual emotion recognition.

[0036] Based on the calculation results of the mental health environmental factors model, the adjustment module further optimizes the computational weights of the negative emotion model, enabling it to adapt to the individual's environment when recognizing emotions. Traditional emotion recognition models typically rely on an individual's physiological and behavioral data, neglecting the impact of dynamic changes in the external environment on individual emotions. This invention uses the output of the mental health environmental factors model as input parameters to guide the negative emotion model to adjust its computational logic during emotion recognition. For example, if a user is under prolonged high economic pressure, the device can lower the threshold for recognizing anxiety, making it easier to detect anxiety; conversely, if the user has a high social support index, the device can raise the threshold for judging anxiety to reduce false positives. The computational weights of the negative emotion model can be adjusted using adaptive learning methods, such as Bayesian optimization or gradient-based dynamic weight update strategies, allowing it to automatically optimize recognition rules based on environmental changes.

[0037] The learning module is used during the optimization process to utilize transfer learning methods, enabling the large negative emotion model to learn individual emotional patterns based on the user's historical emotional change data and dynamic adjustments to environmental factors.

[0038] During model optimization, the learning module employs transfer learning, enabling the large-scale negative emotion model to dynamically adjust based on users' historical emotion change data and environmental factors, achieving individualized learning and optimization. Transfer learning can be implemented through fine-tuning of a pre-trained model, allowing the large-scale negative emotion model to perform personalized emotion recognition under different individual environmental conditions. Alternatively, it can utilize meta-learning methods to enable the model to quickly adapt to new user data. For example, when a new user uses the device for the first time, the model can initially make predictions based on general emotion recognition standards. As the user accumulates historical emotion data, the device can gradually learn the user's personalized emotional response patterns and quickly adjust under new environmental conditions. Transfer learning can be implemented through parameter sharing, adversarial training, or reinforcement learning-based policy adjustments to ensure the model maintains high prediction accuracy despite changes in individual emotion patterns and environmental factors.

[0039] In this embodiment, the core objective of constructing a mental health environmental factor model is to quantify the impact of the external environment on an individual's mental state and incorporate this impact into the calculation process for emotion recognition and mental disorder risk prediction, thereby improving the accuracy and personalization of predictions. To achieve this goal, it is first necessary to acquire long-term and short-term environmental variable data and structure them according to the time dimension to form time-series input data. This data includes air quality, noise levels, socioeconomic stress, temperature, light intensity, and frequency of social interaction. Data collection can be conducted through multiple sensors, external databases, or user-reported data. For example, air quality data can be obtained through government open data interfaces or personal environmental sensors; noise levels can be collected through microphones on smartphones or wearable devices and analyzed for decibel values; and socioeconomic stress data can be modeled using information such as occupation, income, and housing stability provided by the user. All data needs to be standardized according to the time dimension before input to ensure the consistency of characteristics between long-term and short-term variables and to eliminate the influence of outliers.

[0040] After the data is input into the environmental impact factor modeling module, it is processed separately into the long-term trend analysis sub-model and the short-term impact analysis sub-model. The long-term trend analysis sub-model focuses on analyzing the chronic effects of long-term environmental variables on individual emotional states. This sub-model uses a variational autoencoder (VAE) combined with a bidirectional long short-term memory network (Bi-LSTM) to learn the emotional fluctuation patterns of individuals under long-term environmental conditions. The VAE is used to learn the latent distribution of environmental variables and generate low-dimensional representations to reduce dimensional redundancy in environmental data while maintaining the integrity of data features. Subsequently, the Bi-LSTM uses forward and backward time dependencies to extract the impact of long-term environmental factors on individual emotional states. For example, individuals who are in a high-noise environment for a long time may exhibit higher baseline anxiety values. To further improve the modeling ability of long-term emotional fluctuations, a Gaussian mixture model (GMM) is used to model the range of emotional fluctuations to calculate the probability distribution of an individual's long-term emotional state, ensuring that the model can accurately depict the trend of emotional state changes under long-term environmental influences.

[0041] The short-term impact analysis sub-model is used to analyze the immediate effects of short-term environmental variables on an individual's psychological state. This sub-model employs a self-attention mechanism combined with a temporal convolutional network (TCN) to identify the dynamic impact of environmental factors on an individual's psychological state over a short period. The self-attention mechanism captures the relationships between environmental variables and identifies which variables are more relevant to the current emotional state; for example, on a given day, light intensity may have a greater impact on mood than noise level. The TCN processes the time-series features of short-term environmental variables, extracting features along the time dimension through one-dimensional convolution operations and calculating the degree of influence of each environmental factor on the current emotional state; for example, large temperature fluctuations within a day can increase emotional susceptibility. The short-term impact analysis sub-model ultimately outputs the degree of influence of each environmental variable on the current emotional state, providing feature input for subsequent calculations.

[0042] After the long-term trend analysis sub-model and the short-term impact analysis sub-model complete their calculations, their output data is input into the environmental impact factor calculation module to further quantify the influence of various environmental factors on individual emotional states. The environmental impact factor calculation module employs reinforcement learning, using a policy gradient optimization algorithm to adjust the influence coefficients of each environmental variable and generate an environmental impact factor. This impact factor represents the degree to which different environmental factors regulate an individual's emotional state. This module first constructs a reinforcement learning framework, using environmental variables as the state space and individual emotional changes as the reward function, adjusting the weights of environmental variables on emotion recognition. For example, when an individual exhibits significant emotional fluctuations under a specific environmental variable, the device increases the weight of that environmental variable; conversely, it decreases the weight, ensuring that different individuals obtain optimal emotional state calculation parameters under different environmental conditions.

[0043] Environmental impact factors are then input into the environmental adaptation calculation module to further adjust an individual's psychological adaptability under different environmental conditions. This module employs an attention-enhanced Transformer architecture, combining an individual's historical emotional data with current environmental impact factors to calculate their adaptability under different environmental conditions and generate environmental adaptation adjustment parameters. These parameters include dynamically adjusted weights for environmental impact factors, correction factors for emotion recognition, and correction factors for mental health disorder prediction. The dynamically adjusted weights for environmental impact factors are used to adjust the intensity of environmental factors' influence at different time periods; the correction factors for emotion recognition are used to optimize the calculation weights when performing emotion recognition in a large negative emotion model; and the correction factors for mental health disorder prediction are used to improve the accuracy of individualized mental health disorder prediction. For example, for individuals under long-term high socioeconomic stress, the environmental adaptation adjustment parameters may reduce the sensitivity of emotion recognition to prevent short-term emotional fluctuations caused by external stress from being misjudged as abnormal states. Conversely, for individuals experiencing a short-term surge in stress, these parameters may increase the sensitivity of emotion recognition to detect potential mental health disorder risks earlier.

[0044] Ultimately, the environmental adaptation adjustment parameter is used as the final output of the mental health environmental factors model and input into the negative emotion model. The negative emotion model dynamically adjusts the computational parameters for emotion recognition, the sensitivity of mental disorder prediction, and the criteria for judging abnormal mood fluctuations based on this parameter. For example, if an individual's environmental factors over the past week indicate a high-pressure state, and the emotion recognition results show abnormal mood fluctuations, the device can lower the threshold for judging anxiety and increase the probability of predicting mental disorders, ensuring that the individual receives intervention earlier in high-risk situations. By dynamically adjusting the computational logic of emotion recognition and mental disorder prediction, the device can adapt to changes in the mental state of different individuals under different environmental conditions, improving the accuracy of emotion recognition and enhancing the effectiveness of personalized mental health interventions.

[0045] The following is the core reference implementation code for the mental health environmental factors model and the negative emotion big model: import torch import torch.nn as nn import torch.optim as optim from sklearn.mixture import GaussianMixture import numpy as np # 1. Variational Autoencoder (VAE) + LSTM for processing long-term sentiment trends class VariationalAutoencoder(nn.Module): def __init__(self, input_dim, hidden_dim, latent_dim): """ Variational autoencoders (VAEs) are used for modeling long-term environmental variables. :param input_dim: The number of long-term environment variables :param hidden_dim: Hidden layer dimension :param latent_dim: VAE latent variable dimension """ super(VariationalAutoencoder, self).__init__() self.encoder = nn.Sequential( nn.Linear(input_dim, hidden_dim), nn.ReLU(), nn.Linear(hidden_dim, latent_dim * 2) # Outputs the mean and variance ) self.decoder = nn.Sequential( nn.Linear(latent_dim, hidden_dim), nn.ReLU(), nn.Linear(hidden_dim, input_dim), nn.Sigmoid() ) def reparameterize(self, mu, logvar): Variational Reparameterization Techniques std = torch.exp(0.5 * logvar) eps = torch.randn_like(std) return mu + eps * std def forward(self, x): Calculating long-term sentiment trends x = self.encoder(x) mu, logvar = x.chunk(2, dim=-1) # Split the mean and variance z = self.reparameterize(mu, logvar) recon_x = self.decoder(z) # Reconstruct the input data return z, mu, logvar, recon_x # 2. Gaussian Mixture Model (GMM) for Calculating Long-Term Sentiment Fluctuations class GaussianMixtureModel: def __init__(self, num_components=3): """ Gaussian mixture models (GMMs) calculate the probability distribution of an individual's long-term mood fluctuations. :param num_components: The number of mixtures in the Gaussian distribution """ self.gmm = GaussianMixture(n_components=num_components,covariance_type='full') def fit(self, data): Training a GMM by inputting long-term sentiment variables. self.gmm.fit(data) def predict_proba(self, data): Calculate the probability density of the input data. return self.gmm.predict_proba(data) #3. TCN + Self-Attention Processing of Short-Term Emotional Impact class ShortTermImpactModel(nn.Module): def __init__(self, input_dim, hidden_dim, output_dim): """ The impact of short-term environmental factors on current emotional state :param input_dim: Number of short-term environment variables :param hidden_dim: Hidden layer dimension :param output_dim: Impact factor dimension """ super(ShortTermImpactModel, self).__init__() # Temporal Convolutional Networks (TCN) self.tcn = nn.Conv1d(input_dim, hidden_dim, kernel_size=3,padding=1) # Self-attention mechanism self.attention = nn.MultiheadAttention(embed_dim=hidden_dim,num_heads=4) # Output layer self.fc = nn.Linear(hidden_dim, output_dim) def forward(self, x): Calculate the impact factors of short-term environmental factors. x = x.permute(0, 2, 1) # Adjust input dimensions to adapt to TCN tcn_out = self.tcn(x).permute(2, 0, 1) # Adapt to attention mechanism attn_out, _ = self.attention(tcn_out, tcn_out, tcn_out) impact_factor = self.fc(attn_out[-1]) # Get the output of the last time step return impact_factor # 4. Calculate the final environmental impact factors class MentalHealthEnvironmentModel(nn.Module): def __init__(self, long_term_dim, short_term_dim, hidden_dim,latent_dim, output_dim): """ Calculate the combined impact of environmental factors on individual emotions. :param long_term_dim: Number of long-term environment variables :param short_term_dim: Number of short-term environment variables :param hidden_dim: Hidden layer dimension :param latent_dim: VAE latent variable dimension :param output_dim: Final environmental impact factor dimension """ super(MentalHealthEnvironmentModel, self).__init__() self.vae = VariationalAutoencoder(long_term_dim, hidden_dim,latent_dim) self.gmm = GaussianMixtureModel(num_components=3) # Calculate long-term mood fluctuations self.short_term_model = ShortTermImpactModel(short_term_dim,hidden_dim, output_dim) self.fc = nn.Linear(latent_dim + output_dim, output_dim) def forward(self, long_term_data, short_term_data): "Taking into account both long-term and short-term environmental factors, calculate the final environmental impact factor." long_term_z, mu, logvar, _ = self.vae(long_term_data) short_term_impact = self.short_term_model(short_term_data) # Calculate the range of long-term mood fluctuations long_term_prob = torch.tensor(self.gmm.predict_proba(long_term_z.detach().numpy()), dtype=torch.float32) # Combining long-term trends and short-term impacts combined = torch.cat([long_term_prob, short_term_impact], dim=-1) impact_factor = self.fc(combined) return impact_factor, mu, logvar #5. The Transformer Model of Negative Emotions class NegativeEmotionModel(nn.Module): def __init__(self, input_dim, hidden_dim, output_dim): """ Large Model of Negative Emotions :param input_dim: Input dimension of environmental impact factors :param hidden_dim: Hidden layer dimension :param output_dim: Predicts emotional state (anxiety, depression, etc.) """ super(NegativeEmotionModel, self).__init__() self.fc_input = nn.Linear(input_dim, hidden_dim) self.transformer = nn.TransformerEncoder( nn.TransformerEncoderLayer(d_model=hidden_dim, nhead=4), num_layers=2 ) self.fc_output = nn.Linear(hidden_dim, output_dim) def forward(self, env_factors): Calculate the prediction results for negative emotions. x = torch.relu(self.fc_input(env_factors)) x = self.transformer(x.unsqueeze(0)).squeeze(0) output = torch.softmax(self.fc_output(x), dim=-1) return output Furthermore, in the environmental impact factor calculation module, the step of adjusting the impact coefficients of each environmental variable and generating environmental impact factors based on reinforcement learning methods and policy gradient optimization algorithms includes: Based on individual long-term and short-term environmental variable data, a reinforcement learning state space is constructed, where state variables... From time step Environment variable vector Composition, in which Indicates the first Environmental variables, including but not limited to air quality, noise level, socioeconomic stress, temperature, light intensity, and frequency of social interaction, are all normalized to ensure consistent dimensions. Time smoothing is also considered, and the variables are processed using the exponential moving average filter provided in Formula 1 below: in This is a smoothing coefficient used to reduce the impact of short-term fluctuations in environmental variables; a recommended value is 0.5. Indicates time After smoothing, the first One environment variable value; Indicates the previous time After smoothing, the first One environment variable value; Indicates the first Environmental variables at time The original value; In reinforcement learning models, the reward function is defined according to the following formula 2. To optimize the adjustment of environmental impact factors, this function is used to maximize the stability of emotion recognition and minimize drastic fluctuations in environmental impact factors. The calculation method is as follows: in For the first Emotional state in time The recognition probability, calculated by the negative emotion big model, represents the likelihood that an individual will be identified as having that emotional state (e.g., anxiety, depression, calmness, etc.) at that moment. For the first Emotional state in time The probability of recognition; This represents the importance weight of emotional states. This weight is used to adjust the contribution of different emotional states to reward calculation. The recommended value range is... Larger values ​​are suitable for more important emotional states, such as anxiety and depression, while smaller values ​​are suitable for relatively minor emotional states, such as ordinary stress.

[0046] Represents the total number of all emotional states; To smooth out the control factor, a value of 0.1 is recommended. Larger values ​​are used to suppress drastic changes in environmental impact factors, while smaller values ​​allow environmental impact factors to adapt to environmental changes more quickly, thereby improving the device's response to sudden environmental changes.

[0047] For time Environment variables Influence factors; For time Environment variables Influence factors; The number of environment variables; According to Formula 3 below, the environmental impact factor is updated based on the policy gradient optimization algorithm: in, The updated environmental impact factors; Environmental impact factors at the current moment; For the learning rate, a value of 0.01 is recommended. A smaller value... A value that can improve the stability of the optimization but results in a slower convergence speed, while a larger value... Values ​​can accelerate the optimization process, but may cause drastic fluctuations in environmental impact factors.

[0048] This indicates gradient computation based on policy gradient optimization, which is based on the current state. and current action Calculate the reward function Partial derivatives with respect to environmental impact factors; In reinforcement learning models, the current state Representing individuals in time A collection of environmental information and mental health-related data at any given moment. This state includes not only time step... Environment variables under ,in This represents the smoothed, normalized values ​​of various environmental factors (such as air quality, noise levels, socioeconomic stress, temperature, light intensity, and frequency of social interaction), and also includes the individual's historical emotional state distribution, current mental health assessment results, and recent environmental change trends. For example, if an individual has recently been in a high-noise environment, accompanied by high frequency of social interaction and low sleep quality, then the combination of these variables constitutes their current state. In addition, state variables can also include an individual's physiological signals (such as heart rate and skin conductance) as well as historical emotional change patterns to enhance the model's comprehensive understanding of an individual's psychological state.

[0049] Current action Representing the reinforcement learning agent in time step The adopted strategy adjustment scheme, with its action space encompassing a range of possible adjustment strategies, aims to optimize the computational logic of environmental influencing factors, enabling the model to more accurately predict changes in an individual's mental health. Specifically, the action... This mainly includes fine-tuning environmental impact factors, redistributing weights, dynamically adjusting the learning rate, and detecting and handling abnormal states. For example, one possible action is to adjust the impact factors of a specific environmental variable. Adjustments will be made to ensure its progress in the future. The contribution of reinforcement learning models to mental health predictions may increase or decrease. Furthermore, reinforcement learning models may adjust the relative weights of different environmental factors based on historical data, allowing more important variables to dominate emotion predictions under certain environmental conditions, while the influence of secondary variables is appropriately weakened.

[0050] Policy gradient optimization calculation term Reflects the current state Next, take the current action. After that, the reward function The model adjusts the influence factors of environmental variables relative to their changing trends. By calculating this gradient, the model can adjust the influence factors of environmental variables, making emotion predictions more accurate over future time steps while reducing abnormal fluctuations in mental states. For example, if the current optimization direction leads to increased instability in emotional states, the policy gradient optimization calculation may weaken the adjustment in that direction; conversely, if adjusting the influence factor of a certain environmental variable helps reduce fluctuations in emotion recognition, the weight of that variable will be further strengthened. The entire optimization process enables the reinforcement learning agent to dynamically adapt to individual environmental changes, providing individuals with more personalized and accurate mental health predictions and intervention plans.

[0051] This is a second-order fine-tuning factor, with a recommended value of 0.01, used to control the second-order gradient correction term of the environmental impact factor.

[0052] This is a second-order gradient correction term for environmental impact factors, used to enhance stability; The updated environmental impact factor is used as the output of the reinforcement learning training results and input into the environment adaptation calculation module.

[0053] In this embodiment, the computational weights of the negative emotion model need to be dynamically adjusted based on the individual's environmental factors to ensure the accuracy and adaptability of emotion recognition. First, the individual's environmental adaptability parameters are determined using the calculation results of the mental health environmental factors model. These parameters are a quantitative representation of the combined effects of long-term and short-term environmental variables, including factors such as socioeconomic stress levels, long-term living environment quality, noise exposure, social frequency, and the impact of sudden life events. These parameters are obtained based on long-term monitored behavioral data, environmental data, and the individual's self-reported psychological state information. Time series analysis and attention mechanisms are used to calculate the individual's adaptability under different environmental conditions to assess the impact of environmental factors on emotional stability.

[0054] Based on environmental adaptability parameters, the emotion recognition threshold of the large-scale negative emotion model is differentially adjusted, enabling the model to more accurately identify an individual's emotional state and avoid misjudgments due to environmental differences. For individuals who have long been under high socioeconomic pressure or harsh environmental conditions, such as those facing long-term economic burdens, weak social support systems, excessive occupational stress, or residing in areas with high noise or severe air pollution, their anxiety and depression may have already begun to develop due to the long-term influence of external negative factors, but may not have reached the recognition threshold in general emotion recognition devices, leading to a failure to provide timely warnings. Therefore, in cases where environmental adaptability parameters are high, the device will lower the recognition threshold for anxiety and depression, allowing the model to identify even individual emotional fluctuations that have not yet reached the general standards for anxiety or depression, thus providing the possibility of early intervention. For example, if the recognition threshold for anxiety is generally set to a specific range, the device will adjust this range for such individuals, allowing the model to recognize even milder anxiety, thereby improving the sensitivity of individualized recognition.

[0055] Conversely, for individuals experiencing low environmental stress but significant short-term mood swings—such as those with stable daily environments and low socioeconomic pressure, but whose mood changes drastically in a short period due to unexpected events (e.g., increased temporary work pressure, exam anxiety, or social conflict)—the tolerance threshold for negative emotion recognition needs to be increased to avoid misjudgments. In such cases, although an individual's short-term anxiety level may rise significantly, they are usually able to recover quickly due to their strong long-term environmental adaptability. Therefore, it is inappropriate to directly diagnose short-term mood fluctuations as anxiety or depression. To this end, the device will appropriately increase the recognition threshold for anxiety and depression, ensuring that slight, short-term fluctuations do not trigger negative emotion alarms, thereby reducing false alarm rates and improving the accuracy of emotion recognition. For example, if a user experiences high-intensity work pressure in a short period, their heart rate variability decreases, and their vocal emotion characteristics indicate a tense state, but their long-term mental health data shows that they have a high level of psychological adaptability, then the device will appropriately increase the anxiety recognition threshold during this period, ensuring that short-term anxiety signals do not directly trigger anxiety warnings and avoiding over-intervention.

[0056] The entire adjustment process relies on real-time calculation of environmental adaptability parameters and is optimized in conjunction with the individual's historical emotional patterns. This ensures that the model can both identify the risk of emotional disorders in individuals under long-term high stress at an early stage and avoid misjudgments caused by short-term fluctuations. Through this method, the large-scale negative emotion model can more accurately adapt to the individual's environmental conditions, enabling emotion recognition to be based not only on physiological and behavioral data but also to fully consider the individual's long-term and short-term environmental context. This improves the model's personalized emotion recognition capabilities and optimizes the level of intelligence in mental health services.

[0057] In this embodiment, to enable the large-scale negative emotion model to more accurately identify individual emotional states and adapt to different individual emotional patterns, transfer learning is employed to optimize it based on users' historical emotional change data and dynamic adjustments to environmental factors. Because individual emotional responses are highly personalized, traditional emotion recognition models often rely on large-scale data training. However, for new individuals, data collected in a short period may be insufficient, making it difficult for the model to accurately adapt. Therefore, transfer learning is used to enable the model to extract stable features from historical emotional trajectories and environmental influencing factors, and to quickly adapt using existing emotional patterns, thereby improving its ability to predict individual emotional states.

[0058] First, an individual's emotional trend representation is constructed. This representation is modeled based on the individual's historical emotional data, environmental change patterns, and physiological signal data to capture the individual's emotional fluctuation trends under different environments. During data collection, the device continuously acquires the individual's physiological indicators (such as heart rate variability and skin conductance), behavioral patterns (such as social activities and sleep quality), and environmental variables (such as temperature, noise, and air quality). Combined with negative emotion recognition results, a time-series emotional trajectory is formed. This emotional trajectory can be modeled using a time-series neural network (such as LSTM) to identify the individual's long-term emotional fluctuation patterns. For example, an individual may consistently exhibit anxiety in a high-noise environment but tend to be more stable in a quiet environment. Furthermore, an environmental influencing factor regulation network is constructed. This network learns the influence weights of different environmental variables on individual emotions, enabling the large-scale negative emotion model to predict emotional states using an individual's historical emotional trajectory and environmental characteristics even when short-term data is insufficient. For example, if an individual's emotions have historically fluctuated significantly in hot weather, the model can automatically increase the influence weight of temperature on emotion recognition when the temperature rises again, allowing for a reasonable prediction of the individual's emotional state even without complete data for the current moment.

[0059] Building upon this foundation, a reinforcement representation learning framework is employed to further extract individual emotion change patterns under different environmental conditions. This framework learns the emotional distribution of individuals in different environmental states through contrastive learning, enabling the model to better understand the transfer relationships of emotional states. For example, in groups of individuals experiencing significant socioeconomic stress, the model can learn common emotional responses among different individuals in similar environments and adjust weights using historical data from similar situations when transferring to new individuals. To improve individualized adaptability, an adaptive weight mapping layer is constructed. This layer adjusts the model's computational parameters, allowing it to quickly adapt to the emotional response characteristics of different individuals when transferring to new individuals. Specifically, this mapping layer dynamically adjusts the internal computational path of the large negative emotion model based on the individual's initial emotional state distribution, ensuring high emotion recognition accuracy even with limited individual data. For instance, when the device detects that an individual's historical emotional state is highly similar to the emotional pattern of a certain group in the training data, the model can automatically use the group's parameter weights as initial reference values ​​and fine-tune them based on the individual's current data to ensure accurate emotion recognition.

[0060] This embodiment utilizes transfer learning to enable the large negative emotion model to fully leverage historical emotional trajectories and environmental influencing factors when faced with insufficient short-term data. This improves the stability and personalized adaptability of emotion recognition, allowing even individuals using the device for the first time to obtain accurate emotional state recognition results in a shorter time and enhancing the effectiveness of mental health interventions.

[0061] Identification Unit 103: Used to identify an individual's emotional state using an optimized negative emotion model, and combined with a mental health environmental factor model to analyze the individual's social environment, economic pressure, and life events, predict the risk of mental disorders, and generate individual mental health assessment results.

[0062] Once the negative emotion model receives optimized input, it first classifies and identifies the individual's current emotional state. For example, if a user's physiological data indicates a faster heart rate, behavioral data indicates decreased sleep quality, and language data indicates faster speech and a higher pitch, the model may identify the user as being in an anxious state. Simultaneously, by combining historical data, the model can determine whether the current emotional change is a short-term fluctuation or a long-term trend. For instance, if the model finds that the user's emotional identification results show an increasing trend of anxiety over the past two weeks, it may tend to judge the user as being in a state of persistent anxiety rather than a temporary emotional fluctuation.

[0063] After completing emotion recognition, the device further integrates the analysis results of a mental health environmental factors model to comprehensively assess the individual's environmental stress and life events. For example, if a user has a low work environment rating, decreased social interaction frequency, and high economic pressure, these factors may exacerbate the likelihood of anxiety or depression. When calculating risk prediction, the model predicts the probability of mental disorders occurring in the future based on the combined influence of individual historical data and environmental factors.

[0064] Specifically, risk models for predicting mental health disorders can employ time-series-based statistical methods or deep learning methods, such as calculating state transition probabilities based on Markov chains or time-series modeling methods based on the Transformer architecture. The device can calculate the frequency and amplitude of short-term mood fluctuations, as well as the trends in environmental factors. For example, if a user's anxiety score has been consistently rising over the past seven days while their support index has decreased, the device may predict a greater than 60% probability that their anxiety will develop into moderate or severe anxiety in the coming week. Furthermore, the risk prediction model can also incorporate an individual's emotional stability over different time periods. By calculating the mean and standard deviation of short-term and long-term mood fluctuations, it can assess the persistence and relapse risk of the user's current emotional state. For instance, if an individual experiences significant mood fluctuations, and the mental health environmental factors model indicates that the individual is under high environmental stress, the model may conclude that their mood is more likely to worsen further within the next three weeks, thus increasing the intervention priority.

[0065] Ultimately, based on emotion recognition and mental health risk prediction, the device generates a personalized mental health assessment report, including current emotional state, risk assessment of future mental health issues, potential influencing factors, and emotional change trends. This assessment not only provides data support for subsequent intelligent interventions but can also be used for long-term health management, making mental health services more precise, personalized, and intelligent.

[0066] In this embodiment, the identification unit 103 is specifically used to implement the following steps: To enable large-scale negative emotion models to more accurately identify individual emotional states, it is necessary to model long-term emotional trends and construct personalized emotion recognition thresholds based on these trends to adapt to the fluctuating psychological states of different individuals. Since the accuracy of emotion recognition largely depends on the stability of an individual's baseline emotional state, traditional fixed-threshold methods may lead to misjudgments of emotional fluctuations in some individuals. Therefore, by introducing personalized emotion recognition thresholds, the model can dynamically adjust the detection criteria for anxiety and depression based on an individual's historical emotional change patterns.

[0067] First, historical emotional data of individuals is analyzed to extract long-term emotional fluctuation patterns. This data comes from physiological indicators (such as heart rate variability and skin conductance), behavioral data (such as social frequency and activity level), environmental factors (such as temperature and noise), and individual-reported psychological states. Time series analysis is used to calculate stability indicators of individual emotional states, including parameters such as the amplitude of emotional fluctuations, the rate of change, and the duration of emotional states. For example, individuals with large long-term emotional fluctuations are typically observed to have a higher frequency of anxiety states, a wider range of emotional changes, and shorter periods of emotional stability, while individuals with more stable emotions usually exhibit smaller fluctuation amplitudes and longer periods of stability.

[0068] During emotion recognition, the detection thresholds for anxiety and depression are adjusted based on an individual's long-term emotional fluctuation characteristics. If an individual experiences significant emotional fluctuations, it means they may experience multiple transitions between anxiety and depression within a short period. In this case, lowering the detection thresholds for anxiety and depression allows the model to identify abnormal emotional states earlier, improving early risk warning capabilities. For example, if an individual has a high baseline for emotional fluctuations, traditional models might not issue an alert when their anxiety index rises slightly. However, this device appropriately lowers the triggering conditions for anxiety recognition, ensuring that even minor emotional fluctuations are promptly detected and appropriate psychological intervention suggestions are provided. Furthermore, for individuals with severe emotional fluctuations, the device pays closer attention to the duration of the emotional state to ensure that short-term anxiety is not misdiagnosed as long-term anxiety disorder.

[0069] Conversely, for individuals with relatively stable emotions, their emotional state remains within a low fluctuation range over a long period, with fewer sudden changes in anxiety or depression. In this case, raising the emotion recognition threshold reduces false alarms caused by occasional emotional fluctuations. For example, if an individual's emotional state remains stable for a long time, and they occasionally experience brief periods of anxiety due to unexpected events, traditional models might misjudge this as an anxiety state. However, this device, by raising the recognition threshold, allows it to filter out these brief fluctuations, triggering an alarm only when the emotional state remains consistently abnormal. This not only reduces the false alarm rate but also improves the accuracy of individual emotion recognition, making the device more closely reflect the individual's actual psychological state.

[0070] Unit 104: Based on the individual's mental health assessment results, a virtual digital doctor provides a personalized mental health management plan, including intelligent pre-consultation, emotion regulation suggestions, intervention measures, and long-term health tracking services.

[0071] In Unit 104, based on individual mental health assessment results, a virtual digital doctor provides personalized mental health management plans to meet individual psychological needs and improve the accuracy of interventions and the effectiveness of long-term management. This plan encompasses intelligent pre-consultation, emotion regulation suggestions, intervention measures, and long-term health tracking services, and is dynamically adjusted according to changes in the individual's emotional state to ensure the scientific and continuous nature of mental health management.

[0072] Once the device receives the individual's mental health assessment results, the virtual digital doctor first conducts an intelligent pre-consultation to assess the user's current mental state and guide the user to provide necessary supplementary information. The intelligent pre-consultation process is based on Natural Language Processing (NLP) technology, combining the user's self-reported information with the emotion recognition results analyzed by the device to generate a targeted consultation path. For example, if a user is identified as having anxiety tendencies, the device may ask questions such as whether the user has recently experienced significant stressful events, whether their sleep quality has declined, and whether they are experiencing excessive worry. The device can further analyze the answers using an emotion recognition model to determine whether the user's described emotional state matches the emotion score calculated by the device, and adjust the subsequent consultation process accordingly.

[0073] After the intelligent pre-diagnosis is completed, the device generates emotion regulation suggestions based on the user's mental health assessment results and consultation feedback. These suggestions cover a variety of psychological adjustment methods, including breathing exercises, mindfulness meditation, progressive muscle relaxation training, and music therapy. The device recommends the most suitable method based on the user's specific situation and provides guidance. For example, for users experiencing high anxiety and poor sleep quality, the device may recommend breathing exercises and guide them into a relaxed state; while for users with chronic depression, the device may recommend behavioral activation training, such as encouraging them to create a daily activity plan and increase social interaction. The device's recommendations can be presented via text, voice, or video to ensure users can clearly understand and effectively implement them.

[0074] In addition to providing emotion regulation suggestions, the device also develops personalized intervention measures based on individual mental health risk predictions. The content of the intervention measures depends on the user's risk level. For low-risk users, the device may only provide daily psychological adjustment suggestions, while for medium- and high-risk users, the device may recommend further psychological counseling, online psychological guidance, or suggest that the user contact a professional doctor. The device can establish a linkage mechanism with mental health institutions, automatically recommending appropriate mental health service resources when the user's emotional state reaches the intervention threshold. For example, if the user's anxiety risk exceeds 80%, the device can suggest that the user make an appointment with a psychologist and provide appropriate online or offline resources. In addition, the device can adjust the intervention plan based on the user's lifestyle and environmental factors. For example, if the user's high stress comes from the workplace, the device may recommend workplace psychological management skills, while if it comes from interpersonal relationships, the device may provide social anxiety relief strategies.

[0075] Long-term health tracking services are a crucial component of the overall mental health management solution. The device continuously collects physiological and behavioral data throughout the user's daily life and regularly updates mental health assessment results to determine trends in the user's mood. The device can be set up for regular mental state monitoring, such as providing a weekly mood self-assessment questionnaire, and adjusting personalized management plans based on physiological and behavioral data. Furthermore, the device can perform pattern analysis based on the user's historical data to identify cycles of mood fluctuations and provide early warnings and interventions at specific times. For example, if the device detects that a user typically experiences low mood at the beginning of the month, it can send psychological adjustment suggestions at the beginning of the following month to help the user prepare mentally and reduce the impact of mood fluctuations.

[0076] The providing unit 104 is specifically used to implement the following steps: When providing personalized mental health management plans, virtual digital doctors first analyze an individual's adaptability to different environmental factors based on their mental health assessment results. The assessment of environmental adaptability comprehensively considers an individual's historical emotional change patterns, behavioral data, frequency of social interaction, and the calculation results of a mental health environmental factor model to determine the individual's emotional fluctuation characteristics under different stressors or environmental conditions. If an individual can maintain a relatively stable emotional state when facing sudden events or long-term stress, it indicates strong environmental adaptability; conversely, if an individual experiences significant emotional fluctuations under environmental changes or social pressure, it indicates weak adaptability.

[0077] Building upon this foundation, the virtual digital doctor employs a dynamic adjustment mechanism to provide corresponding mental health intervention programs tailored to individuals with varying levels of adaptability. For individuals with higher adaptability, the device prioritizes self-help emotion regulation programs, such as meditation, exercise, or music therapy, to help users regulate their emotions autonomously in daily life and reduce psychological burden. The device can also incorporate individual lifestyle habits, such as sending exercise suggestions during the user's usual workout times or providing suitable meditation audio during their rest periods, thereby enhancing the effectiveness of the intervention.

[0078] For individuals with lower adaptability, the device tends to recommend professional psychological counseling or remote intervention services to provide more direct support. When pushing intervention plans, the device considers the user's daily behavioral patterns, such as analyzing their weekday and weekend routines, social activity, and sleep quality, to select the time when the user is most receptive to psychological intervention, preventing the intervention content from being ignored during periods of poor individual well-being or busyness. In this way, the virtual digital doctor can provide more precise and intelligent mental health management plans tailored to the needs of different individuals, improving the effectiveness of psychological interventions.

[0079] Unit 105: Used to combine AI emotion recognition virtual human companionship service, utilize non-contact physiological indicator collection and multimodal emotion recognition technology to monitor the user's emotional state in real time, and dynamically adjust the interactive content of AI emotion recognition virtual human based on the analysis results of the mental health environmental factor model to provide personalized psychological support.

[0080] In adjustment unit 105, combined with AI emotion recognition virtual human companionship service, non-contact physiological indicator collection and multimodal emotion recognition technology are used to monitor the user's emotional state in real time. Based on the analysis results of the mental health environmental factor model, the interactive content of the AI ​​emotion recognition virtual human is dynamically adjusted to provide personalized psychological support. The core of this step is to enable users to obtain long-term, personalized psychological companionship services through continuous emotion data collection and dynamic interaction, thereby improving the effectiveness of mental health intervention.

[0081] The device first continuously monitors the user using non-contact physiological indicator acquisition technology. Data sources can include cameras, microphones, infrared sensors, ambient light sensors, and other devices to capture the user's facial expressions, voice features, posture changes, and physiological signals. For example, the camera can analyze the user's micro-expression changes, such as raised eyebrows and downturned corners of the mouth, using computer vision technology. These subtle facial movements can serve as reliable indicators of emotional state. The microphone can identify the user's tone, speech rate, and pause patterns through voice emotion analysis to determine whether the user is in an excited, anxious, or depressed state. In addition, the infrared sensor can detect changes in skin temperature, and the ambient light sensor can analyze the user's daily routines in conjunction with time and brightness to determine whether their psychological state is affected by the external environment.

[0082] After acquiring this physiological and behavioral data, the device performs multimodal emotion recognition, fusing information from different data sources using a deep learning model to improve the accuracy of emotion recognition. The model can employ a combination of convolutional neural networks (CNNs) and long short-term memory networks (LSTMs) to extract facial features from static images and combine this with time-series analysis of speech data to detect emotional fluctuations. For example, if a user's facial expression is tense, and their speech contains numerous pauses and is faster than usual, the device may identify that the user is in an anxious state. To further improve recognition accuracy, the device also incorporates the results of a mental health environmental factor model into its calculations. For instance, if a user is in a high-stress environment, the model can appropriately lower the threshold for anxiety recognition to more sensitively detect emotional abnormalities.

[0083] Once the device recognizes the user's emotional state, it automatically adjusts the interactive content of the AI ​​emotion-recognition virtual human to better suit the user's emotional needs. If the device determines that the user is anxious, the virtual human may adjust its tone to make the conversation more soothing and offer deep breathing exercises or emotional stabilization techniques. If the user exhibits loneliness or low mood, the virtual human may enhance interactivity, guiding the user to engage in more voice or text communication, or even recommending suitable mindfulness training or meditation courses. Furthermore, the virtual human can dynamically adjust its interaction methods based on the user's interests and habits; for example, it might recommend soothing background music to users who enjoy music, or provide comfortable virtual environment animations to users who prefer visual stimulation, to enhance the relaxation effect.

[0084] The device's dynamic adjustment capabilities ensure personalized psychological support, avoiding the mechanical feeling or inefficiency that static interaction modes may bring. In addition to adjusting the interactive content in real time, the device can also optimize based on the user's long-term emotional trends. For example, if it detects that a user experiences significant mood swings at night, the virtual avatar may provide additional psychological support in the evening, such as lighthearted conversation or guiding the user into a relaxing state before sleep. If a user's emotional stability significantly declines over a period of time, the device may increase the frequency of the virtual avatar's proactive interactions and recommend additional interventions based on a mental health environmental factors model, such as adjusting the user's daily routine or guiding the user to participate in more social activities to reduce psychological burden.

[0085] The adjustment unit 105 is specifically used to implement the following steps: In this embodiment, the AI ​​emotion recognition virtual human analyzes an individual's historical emotional fluctuation patterns and dynamically adjusts its voice tone, facial expressions, and interactive content to provide psychological support that better meets individual needs. Based on an individual's long-term emotional data, the device identifies the stability and fluctuation trends of emotional states and their responses to external environmental stimuli. Combined with calculations from a mental health environmental factor model, it infers the individual's current psychological state and emotional needs, thereby optimizing the virtual human's interaction methods.

[0086] For users who are chronically anxious, the virtual avatar adopts a gentle, slow tone of voice, avoiding overly abrupt or forceful expressions to create a sense of security. Furthermore, during interaction, the virtual avatar provides positive emotional feedback, such as conveying encouragement, comfort, and recognition through voice, facial expressions, or body language, allowing users to feel emotionally supported and reducing anxiety levels. The virtual avatar can also appropriately reduce the input of complex information to avoid exacerbating the user's psychological burden due to information overload.

[0087] For users experiencing short-term mood swings, the device identifies a recent downward trend in their emotional state, but indicates relatively high long-term emotional stability. This suggests the user may be experiencing a temporary mood fluctuation rather than a persistent psychological issue. In this case, the virtual human increases its initiative in voice interaction to reduce the user's sense of social isolation. For example, the virtual human initiates conversations more frequently, guiding the user into lighthearted topics or providing interest-related information to encourage the user to express their emotions. Furthermore, the virtual human's language style becomes more social, employing a friendly and approachable tone to simulate a natural communication context, allowing the user to feel accompanied and cared for during the interaction, thereby promoting self-regulation of emotions.

[0088] The entire interaction process relies on the real-time calculations of the AI ​​emotion recognition device and is personalized by combining the individual's historical emotion patterns. This enables the virtual human to provide more accurate and contextualized psychological support when interacting with users, improves the level of intelligence in psychological intervention, and helps users regulate their emotions more effectively.

[0089] Although this application discloses preferred embodiments as described above, it is not intended to limit this application. Any person skilled in the art can make possible changes and modifications without departing from the spirit and scope of this application. Therefore, the scope of protection of this application should be determined by the scope defined in the claims of this application.

Claims

1. A digital-element-based intelligent virtual-reality mental health service device, characterized in that, The method comprises the following steps: An acquisition unit is configured to acquire individual mental health-related data, including physiological indicators, behavior data, environmental factor data, and socio-economic data, and perform multi-modal fusion to form comprehensive feature data; A construction unit is configured to construct a mental health environmental factor model based on the comprehensive feature data, to quantify the influence of external environment on individual mental state, and use the same as an input parameter to optimize the calculation weight and prediction logic of the negative emotion large model; An identification unit is configured to identify the individual emotional state using the optimized negative emotion large model, and analyze the social environment, economic pressure, and life event factors of the individual based on the mental health environmental factor model to predict the risk of mental disorder and generate an individual mental health assessment result; A provision unit is configured to provide personalized mental health management solutions from a virtual digital doctor based on the individual mental health assessment result; An adjustment unit is configured to combine AI emotion recognition virtual human companion services, use non-contact physiological indicator acquisition and multi-modal emotion recognition technology to monitor the user's emotional state in real time, and dynamically adjust the interaction content of the AI emotion recognition virtual human based on the analysis result of the mental health environmental factor model to provide personalized psychological support.

2. The digital-element-based intelligent virtual-real combined mental health service device according to claim 1, wherein, The construction unit comprises: A processing module is configured to acquire physiological indicators, behavior data, environmental factor data, and socio-economic data of an individual, and perform normalization processing on different types of data through adaptive multi-modal fusion technology; A construction module is configured to construct a mental health environmental factor model using time series analysis and deep learning methods, wherein the mental health environmental factor model establishes the influence weight of individual mental state based on long-term and short-term environmental variables, the long-term environmental variables include air quality, urban noise level, and socio-economic pressure of the individual's long-term living environment, the short-term environmental variables include temperature, light intensity, and social interaction frequency of the day, and the long-term and short-term variables calculate their comprehensive influence value on the current mental state through an attention mechanism; An adjustment module is configured to dynamically adjust the calculation weight of the negative emotion large model based on the calculation result of the mental health environmental factor model, so that the individual emotional state can be adapted to the environment in which the individual is located; A learning module is configured to use a transfer learning method to enable the negative emotion large model to learn the individual emotional pattern based on the historical emotional change data of the user and the dynamic adjustment of the environmental factors during the optimization process. 3.The digital-element-based intelligent virtual-real combined mental health service device according to claim 2, wherein, The construction module is specifically configured to: Acquire long-term and short-term environmental variable data of an individual, including air quality, noise level, socio-economic pressure, temperature, light intensity, and social interaction frequency, and perform structured processing according to the time dimension to form time series input data; inputting the time series input data into an environmental impact factor modeling module, the environmental impact factor modeling module comprising a long-term trend analysis submodel and a short-term impact analysis submodel, wherein the long-term trend analysis submodel extracts the chronic impact of environmental factors on the emotional state of an individual based on long-term environmental variables by combining a variational autoencoder with a bidirectional long short-term memory network, and calculates the probability distribution of the long-term emotional fluctuation range of the individual using a Gaussian mixture model; the short-term impact analysis submodel identifies the dynamic impact of environmental factors on the psychological state of an individual in a short time based on short-term environmental variables by combining a self-attention mechanism with a time convolution network, and calculates the impact degree of each environmental factor on the current emotional state; inputting the outputs of the long-term trend analysis submodel and the short-term impact analysis submodel into an environmental impact factor calculation module, the environmental impact factor calculation module adjusting the impact coefficients of each environmental variable based on a reinforcement learning method through a policy gradient optimization algorithm, and generating environmental impact factors representing the regulation degree of different environmental factors on the emotional state of an individual; inputting the environmental impact factors into an environmental adaptation calculation module, the environmental adaptation calculation module calculating the psychological state adaptation ability of an individual under different environmental conditions based on an attention-enhanced Transformer architecture by combining the historical emotional data of the individual with the current environmental impact factors, and generating environmental adaptability adjustment parameters, wherein the environmental adaptability adjustment parameters include dynamic adjustment weights of the environmental impact factors, correction factors for emotional recognition, and correction factors for psychological disorder prediction; inputting the environmental adaptability adjustment parameters as the final output of the mental health environmental factor model into the negative emotion large model, so that the negative emotion large model dynamically adjusts the calculation parameters of emotional recognition, the sensitivity of psychological disorder prediction, and the judgment standard of emotional fluctuation abnormalities according to the environmental adaptability of the individual, to enhance the accuracy of emotional recognition and improve the effectiveness of personalized mental health intervention.

4. The digital-element-based intelligent virtual-real combined mental health service device according to claim 3, characterized in that, The environmental impact factor calculation module is specifically configured to: Based on individual long-term and short-term environmental variable data, a reinforcement learning state space is constructed, where state variables... From time step Environment variable vector Composition, in which Indicates the first Environmental variables, including air quality, noise level, socioeconomic stress, temperature, light intensity, and frequency of social interaction, were all normalized to ensure consistent dimensions. Time smoothing was also considered, and the variables were processed using the exponential moving average filter provided in Formula 1 below: wherein is a smoothing coefficient, used to reduce the impact of short-term fluctuations in the environmental variable; represents the original value of the environmental variable at time after smoothing; the environmental variable value at time represents the original value of the environmental variable at time after smoothing; the environmental variable value at time represents the original value of the environmental variable at time after smoothing; the environmental variable value at time In the reinforcement learning model, the reward function is defined as follows in Equation 2 with the adjustment of the optimization environmental impact factor: in For the first Emotional state in time The recognition probability, calculated by the negative emotion big model, represents the likelihood that an individual will be identified as having that emotional state at that moment. For the first Emotional state in time The probability of recognition; Weighting of the importance of emotional state; Represents the total number of all emotional states; As a smoothing control factor; For time Environment variables Influence factors; For time Environment variables Influence factors; The number of environment variables; update the environmental impact factors based on the policy gradient optimization algorithm according to Formula 3 as follows: wherein, is the updated environmental impact factor; is the current time's environmental impact factor; is the learning rate; denotes the gradient computation based on policy gradient optimization, is the gradient based on the current state and the current action computes the reward function is the partial derivative of the environmental impact factor; is the second-order fine-tuning factor; is the second-order gradient correction term of the environmental impact factor, used to enhance stability; The updated environmental impact factors are output as the reinforcement learning training results and are input into the environmental adaptation calculation module.

5. The digital-element-based intelligent virtual-real combined mental health service device of claim 2, wherein, The adjustment module is specifically configured to: differentially adjust the recognition thresholds of different emotional states based on the environmental adaptability parameters of the individual to improve the personalized accuracy of emotional recognition, specifically including: for individuals who are under high social and economic pressure or in harsh environmental conditions for a long time, reducing the recognition thresholds of anxiety and depression states so that the negative emotion large model can detect potential psychological disorders earlier; for individuals who are under low environmental pressure but have large short-term emotional fluctuations, increasing the tolerance range of negative emotion recognition to avoid misjudgment.

6. The digital-element-based intelligent virtual-real combined mental health service device of claim 2, wherein, The learning module is specifically configured to: construct an individual emotional trend representation, and through the environmental impact factor regulation network, enable the negative emotion large model to predict the emotional state of an individual under the current environment using the historical emotional trajectory in the case of insufficient short-term data; Based on the reinforcement representation learning framework, the emotional change pattern of the individual under different environmental conditions is extracted, and an adaptive weight mapping layer is constructed, so that the negative emotion large model can quickly adjust the calculation parameters when migrating to a new individual to adapt to the individualized emotional response characteristics.

7. The digital-element-based intelligent virtual-real combined mental health service device of claim 1, wherein, The acquisition unit is specifically used for: Continuously sampling the physiological indicators of the user, and identifying the key nodes of emotional changes based on an abnormal fluctuation detection mechanism; Through an adaptive sampling strategy, when abnormal fluctuations in the individual's heart rate, skin electrical response or voice features are detected, the data sampling frequency is increased to obtain more detailed physiological signals, thereby improving the accuracy of subsequent emotion recognition, while avoiding the impact of long-term high-frequency sampling on user privacy and device energy consumption.

8. The digital-element-based intelligent virtual-real combined mental health service device of claim 1, wherein, The recognition unit is specifically used for: Based on the individual's long-term emotional trend, a personalized emotion recognition threshold is constructed, for individuals with large emotional fluctuations, the detection threshold of anxiety and depression state is reduced to improve the early risk identification ability, and for individuals with relatively stable emotions, the recognition threshold is increased to reduce the false positive rate and improve the accuracy of personalized emotion analysis.

9. The digital-element-based intelligent virtual-real combined mental health service device of claim 1, wherein, The providing unit is specifically used for: Combined with the user's environmental adaptability, a dynamic adjustment mechanism is provided for the psychological intervention scheme, specifically including: For individuals with strong adaptability, self-help emotion regulation schemes are preferentially recommended, including meditation, exercise or music therapy; for individuals with weak adaptability, professional psychological counseling or remote intervention services are preferentially recommended, and the push time of the intervention scheme is optimized in combination with the user's daily behavior pattern.

10. The digital-element-based intelligent virtual-real combined mental health service device of claim 1, wherein, The adjustment unit is specifically used for: According to the historical emotional fluctuation pattern of the individual, the voice tone, expression change and interactive content of the virtual person are dynamically adjusted, wherein for the user in long-term anxiety state, the virtual person adopts soft tone and increases positive emotional feedback, and for the user in short-term emotional depression, the initiative of voice interaction is increased and more social conversation content is provided.

Images