A personalized emotional intervention method, device and electronic device

Abstract

This invention provides a personalized emotion intervention method, device, and electronic device. The method includes: collecting multidimensional data from a target user; the multidimensional data includes physiological signals, behavioral performance, environmental context, and cultural background; constructing an emotion inference model based on the multidimensional data and formulating an emotion intervention strategy for the target user; executing the emotion intervention strategy and collecting feedback information from the target user; and optimizing the emotion inference model and the emotion intervention strategy based on the multidimensional data and the feedback information. This invention enables machines to understand the dynamic interaction of emotions, achieving a closed loop from emotion perception and state inference to dynamic and precise intervention, thus solving the problems of poor accuracy and adaptability in emotion inference in existing related technologies.

Description

Technical Field

[0001] This invention relates to the field of digital health technology, and in particular to a personalized emotional intervention method, device, and electronic device. Background Technology

[0002] With the continued acceleration of societal aging and the increasing demand for public mental health services, emotional companion robots and intelligent intervention systems have become a research hotspot in the intersection of artificial intelligence and digital health. Existing technological solutions typically rely on sensors such as cameras, microphones, and wearable devices to collect users' facial expressions, voice, and basic physiological data, and then use machine learning models to identify emotional states, thereby triggering preset interactive content, such as playing music, initiating conversations, or providing simple suggestions.

[0003] However, existing technologies still have the following significant limitations in achieving precise, personalized, and culturally adaptable deep emotional companionship: Emotion recognition is static and lacks dynamic extrapolation capabilities: Most existing systems classify emotions in isolation and instantaneously (such as joy, anger, sorrow, and happiness), which is essentially an "emotional snapshot" and lacks an understanding and modeling of emotions as a continuous and dynamic evolutionary process. This mechanism cannot capture subtle changes in users' emotions, let alone predict their future evolutionary trends, resulting in a serious lag in emotional intervention and making it difficult to achieve forward-looking and accurate responses.

[0004] Intervention strategies are rigid and lack adaptability and personalization: Most systems rely on preset "if-then" rule bases or fixed strategy templates for intervention logic, lacking the ability to adaptively optimize based on real-time user status and long-term feedback. Their interactive content is monotonous and cannot be deeply customized according to user characteristics, historical interaction records, and rich cultural background, leading to a rigid user experience over time and a diminishing intervention effect.

[0005] The theoretical models are simplistic and lack cultural adaptability and depth of empathy: Existing affective computing models are generally based on Western psychological theories and have failed to effectively integrate emotional cognitive patterns from the perspective of Eastern culture, especially the long-standing theories of emotions in traditional Chinese medicine (such as "emotions overcoming each other"). This makes it difficult for the system to understand the emotional connotations of Chinese users, especially the elderly, in specific cultural contexts, and thus fails to achieve a truly "heart-to-heart" and "empathetic" companionship experience.

[0006] The contradiction between data privacy and security and continuous model optimization: Sentiment data is highly sensitive and involves core user privacy. The existing centralized cloud processing model is at risk of data leakage, while relying entirely on local processing, although protecting privacy, traps the model in "data silos," preventing it from leveraging collective intelligence for continuous iteration and optimization, ultimately leading to stagnation in the system's intelligence level.

[0007] There is currently no effective solution to the problems of poor accuracy and adaptability in sentiment inference of existing related technologies. Summary of the Invention

[0008] This invention provides a personalized emotional intervention method, device, and electronic device to address the shortcomings of existing related technologies in terms of the accuracy and adaptability of emotional inference, and to achieve a closed loop from emotional perception and state inference to dynamic and precise intervention.

[0009] In a first aspect, the present invention provides a personalized emotional intervention method, comprising: Collect multidimensional data of the target user; the multidimensional data includes physiological signals, behavioral performance, environmental context, and cultural background; Based on the multidimensional data, an emotion inference model is constructed, and an emotion intervention strategy is formulated for the target users. Implement the emotional intervention strategy and collect feedback information from the target users; Based on the multidimensional data and the feedback information, the emotion inference model and the emotion intervention strategy are optimized.

[0010] According to a personalized emotion intervention method provided by the present invention, an emotion inference model is constructed based on the multidimensional data, including: Based on the multidimensional data, the emotional intensity of the target user is quantified using an emotional state calculation engine. Based on the intensity of the emotion, an emotion inference model based on the seven emotions transfer network is constructed by integrating relevant theories of traditional Chinese medicine.

[0011] According to a personalized emotional intervention method provided by the present invention, based on the multidimensional data, an emotional state calculation engine is used to quantify the emotional intensity of the target user, including: The emotional intensity components and corresponding dynamic weight coefficients of the target user in four dimensions—physiological signals, behavioral performance, environmental context, and cultural background—are determined respectively. The emotional intensity components of the target user are weighted and fused from four dimensions: physiological signals, behavioral performance, environmental context, and cultural background, to obtain a quantitative value of the target user's emotional intensity.

[0012] According to a personalized emotional intervention method provided by the present invention, based on the multidimensional data, the dynamic weight coefficients are dynamically adjusted according to the data quality and contextual relevance of the corresponding dimension. When the data quality of a certain emotional intensity component decreases, the corresponding dynamic weight coefficient decreases, and the decreased portion is proportionally allocated to other emotional intensity components. In specific situations, the dynamic weight coefficients of dimensions related to the current situation can be temporarily increased.

[0013] According to a personalized emotional intervention method provided by the present invention, based on the multidimensional data, an emotional intervention strategy is formulated for the target user, including: The emotional state of the target user is identified by the emotional inference model, and the emotional evolution trend of the target user is predicted to obtain the prediction result. A personalized emotional profile and a dynamic emotional graph of the target user are generated by combining a dynamic knowledge graph. Based on reinforcement learning algorithms, an emotional intervention strategy for the target user is generated according to the emotional intensity, the prediction result, and the dynamic emotional graph.

[0014] According to a personalized emotional intervention method provided by the present invention, based on the multidimensional data, a personalized emotional profile and a dynamic emotional graph of the target user are generated by combining dynamic knowledge graphs, including: The physiological signals, emotional state, and current events of the target user are integrated into triplet nodes to construct a personalized emotional profile for the target user and generate a dynamic emotional graph.

[0015] According to a personalized emotion intervention method provided by the present invention, the dynamic emotion map, based on the multidimensional data, includes: Event nodes are used to record the time, type, and environment of interactions; Emotion nodes are used to store the intensity, emotion tag, and duration of emotions output by the emotional state calculation engine; Physiological nodes are used to store multidimensional data collected from the target user.

[0016] According to a personalized emotional intervention method provided by the present invention, based on the multidimensional data, the emotional intervention strategy for the target user includes one or more of the following: Music with the property of mutual restraint was selected based on relevant theories of traditional Chinese medicine. Generate nostalgic narrative content that aligns with the cultural background of the target users; Guide the target users to perform mindfulness breathing and relaxation training.

[0017] Secondly, the present invention also provides a personalized emotional intervention device, comprising a smart terminal device, an edge computing node, and a cloud management platform, specifically including: The perception module is used to collect multidimensional data of the target user; the multidimensional data includes physiological signals, behavioral performance, environmental context, and cultural background. The decision-making module is used to construct an emotion inference model based on the multidimensional data and formulate emotion intervention strategies for the target users. An execution module is used to execute the emotional intervention strategy and collect feedback information from the target user; An optimization module is used to optimize the emotion inference model and the emotion intervention strategy based on the multidimensional data and the feedback information.

[0018] Thirdly, the present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the personalized emotional intervention method as described in the first aspect above.

[0019] Fourthly, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the personalized emotion intervention method as described in the first aspect above.

[0020] Fifthly, the present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the personalized emotional intervention method as described in the first aspect above.

[0021] Compared with the prior art, the present invention has the following beneficial effects: This invention provides a personalized emotional intervention method that deeply integrates the theory of mutual restraint of emotions in Traditional Chinese Medicine with modern emotional computing. It constructs an emotional state transfer network, enabling machines to understand the dynamic interaction of emotions and achieving a theoretical leap from "static perception" to "dynamic deduction and prediction," providing a theoretical foundation for proactive intervention. Furthermore, this method solves the challenges of personalization and adaptability in intervention strategies through a closed-loop technology of "multi-dimensional fusion perception + dynamic emotional deduction + reinforcement learning decision-making." It continuously learns and adapts to the user's unique responses, providing precise emotional support tailored to each individual, much like a true "TCM confidant." This achieves a closed loop from emotional perception and state deduction to dynamic and precise intervention, addressing the issues of poor accuracy and adaptability in existing related technologies. It is suitable for providing in-depth emotional companionship and mental health support to elderly individuals with a deep Chinese cultural background. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0023] Figure 1 This is a flowchart of the personalized emotional intervention method provided by the present invention; Figure 2 This is a schematic diagram illustrating data acquisition and fusion processing in an embodiment of the present invention; Figure 3 This is a schematic diagram of the emotional state transition network structure and state inference principle in an embodiment of the present invention; Figure 4 This is a schematic diagram of the node relationships and construction process of the dynamic knowledge graph in an embodiment of the present invention; Figure 5 This is a structural block diagram of the personalized emotional intervention device provided by the present invention; Figure 6 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0025] This invention provides a personalized emotional intervention method. Figure 1 This is a flowchart of the personalized emotional intervention method provided by the present invention, such as... Figure 1 As shown, the method includes the following steps: Step S101: Collect multidimensional data of the target user; multidimensional data includes physiological signals, behavioral performance, environmental context, and cultural background. Step S102: Based on multidimensional data, construct an emotion inference model and formulate an emotion intervention strategy for the target users; Step S103: Implement emotional intervention strategies and collect feedback information from target users; Step S104: Optimize the emotion inference model and emotion intervention strategy based on multidimensional data and feedback information.

[0026] This method first collects multidimensional data on target users, including physiological signals, behavioral performance, environmental context, and cultural background. Then, based on this collected multidimensional data, it integrates the traditional Chinese medicine theory of "emotional interaction" to construct an emotion prediction model. This model enables dynamic identification, evolution prediction, and trend analysis of the target users' emotional states, allowing for real-time formulation and adjustment of emotion intervention strategies. The aforementioned emotion intervention strategies are then implemented, and feedback from the target users is collected. Finally, through a federated learning mechanism, the emotion prediction model is continuously optimized while protecting user data privacy. This process deeply integrates the traditional Chinese medicine theory of emotional interaction with modern affective computing, constructing an emotion state transition network that enables machines to understand the dynamic interaction of emotions. This represents a theoretical leap from "static perception" to "dynamic prediction and forecasting," providing a theoretical foundation for proactive intervention. Moreover, this method solves the problem of personalized and adaptive intervention strategies through a technical closed loop of "multi-dimensional fusion perception + dynamic emotional inference + reinforcement learning decision-making". It continuously learns and adapts to the unique reactions of users, and can provide precise emotional support "one policy for one person" like a true "traditional Chinese medicine confidant". It realizes a closed loop from emotional perception and state inference to dynamic and precise intervention, and solves the problems of poor accuracy and adaptability of emotional inference in existing related technologies. It is suitable for providing in-depth emotional companionship and mental health support to the elderly with a deep Chinese cultural background.

[0027] In step S101, the collected multidimensional data includes: Physiological signals: such as heart rate, skin conductance, heart rate variability (HRV), etc. Behavioral characteristics: such as facial micro-expressions, speech acoustic features, body movements, etc. Contextual factors: such as time, location, and current activity; Cultural background data: such as dialect usage, cultural content preferences, personal life experiences, etc.

[0028] By integrating multi-dimensional data, we can provide decision-makers with a comprehensive and continuous understanding of user status.

[0029] For example, as the data input source, this method employs a multi-sensor fusion technique, such as... Figure 2 As shown, Figure 2This is a schematic diagram illustrating data acquisition and fusion processing in an embodiment of the present invention. The physiological signal acquisition subsystem integrates medical-grade sensors, capable of receiving real-time sensor data of multidimensional physiological indicators such as heart rate, skin conductance, and body temperature via Bluetooth, and can calculate the time-domain (Standard Deviation of NN intervals, SDNN) and frequency-domain (LF / HF) indices of heart rate variability (HRV). These sensors, through optimized signal processing algorithms, effectively filter out environmental noise and motion artifacts, ensuring the accuracy and reliability of the acquired data.

[0030] The specific hardware configuration is as follows: ECG monitoring uses the MAX86150 biosensor module, which supports simultaneous acquisition of PPG and ECG; skin conductance response uses the Grove-GSR sensor, with a measurement range of 0.1-100μS and an accuracy of ±5%; body temperature monitoring uses the MLX90614 non-contact infrared temperature sensor, with an accuracy of ±0.2℃.

[0031] Signal processing flow: 1. Raw signal acquisition: Python def acquire_physio_signals(): ecg_data = max86150.read_ecg(sample_rate=250) gsr_data = grove_gsr.read_conductance() temp_data = mlx90614.read_temperature() return preprocess_signals(ecg_data, gsr_data, temp_data) 2. Signal Preprocessing: ECG signals were processed using a 0.5-40Hz bandpass filter to remove power line interference and baseline drift; AE signals were processed using a Butterworth low-pass filter with a cutoff frequency of 2Hz; data standardization was achieved using z-score normalization. Data was transmitted to the main processor via Bluetooth 5.0, where an FIR filter was used for signal denoising.

[0032] For target user behavior analysis, the system combines computer vision and speech processing technologies to achieve real-time analysis of facial expressions, body movements, and speech features. This part of the system employs a deep learning model enhanced with an attention mechanism to ensure accurate recognition of specific emotional expressions within the Chinese cultural context. The visual analysis hardware uses a Logitech C920 camera with 1080P resolution and a 30fps frame rate; it uses the OpenPose skeleton recognition algorithm to extract 21 key body joints; facial expression analysis uses a ResNet-18 model trained on the FER2013 dataset; micro-expression recognition uses the CASME II dataset, achieving an accuracy of 85.6%. The speech analysis hardware uses a ReSpeaker 6-microphone circular array with a 16kHz sampling rate; acoustic feature extraction includes 88 dimensions such as fundamental frequency, energy, and spectral centroid; emotion recognition uses an LSTM network trained on the RAVDESS dataset; real-time processing involves frame-by-frame processing of the audio stream, with a frame length of 25ms and a frame shift of 10ms.

[0033] For environmental context awareness of the target user, a multi-source sensor network continuously collects information about the user's physical environment, including parameters such as temperature, humidity, and light intensity, and combines this with location information for contextual understanding. Hardware configuration: The environmental sensor uses a BME680 multi-functional sensor (temperature, humidity, air pressure, VOC); time information is obtained through the system clock, distinguishing between weekdays / weekends and mornings / noon / evenings; location information is obtained using a U-blox NEO-M8N GPS module for positioning, combined with Wi-Fi fingerprint-assisted positioning; activity recognition uses an MPU9250 nine-axis motion sensor, accelerometer, and gyroscope data, based on an LSTM network to identify the user's current activity status.

[0034] The software code is as follows: python class EnvironmentMonitor: def __init__(self): self.sensors = initialize_sensors() self.activity_model = load_activity_model() def monitor_environment(self): env_data = { 'temperature': self.sensors.read_temperature(), 'humidity': self.sensors.read_humidity(), 'location': self.sensors.get_gps_coordinates(), 'activity': self.predict_activity() } return env_data For the collection of cultural background data from target users, we analyze their media consumption habits, the frequency of dialect vocabulary use in conversations (e.g., detecting Wu dialect characteristic words appearing more than 5 times / minute), and their historical viewing records (e.g., a preference for Huangmei Opera reaching 85%) to construct a user cultural profile that includes cultural preferences, language habits, and personal experiences. This provides important contextual information for subsequent emotional intervention. Data collection examples: Media consumption analysis uses API interfaces to obtain users' music and video viewing history to collect user cultural preferences; Language feature analysis uses a dialect recognition model based on DeepSpeech; Personal experience construction adopts a progressive question-and-answer approach to avoid information overload at once.

[0035] In some embodiments, step S102, based on multidimensional data, constructs an emotion inference model, including: based on multidimensional data, using an emotional state calculation engine to quantify the emotional intensity of the target user; based on the emotional intensity, integrating relevant theories of traditional Chinese medicine to construct an emotion inference model based on the seven emotions transfer network, thereby realizing dynamic identification, evolution prediction and trend analysis of the target user's emotional state.

[0036] Specifically, based on multidimensional data, the emotional intensity of the target user is quantified using an emotional state calculation engine. This includes: determining the emotional intensity components and corresponding dynamic weight coefficients of the target user in four dimensions: physiological signals, behavioral performance, environmental context, and cultural background; and weighting and fusing the emotional intensity components of the target user in the four dimensions to obtain a quantitative value of the target user's emotional intensity.

[0037] For example, the formula used by the emotion state calculation engine to quantify emotion intensity is:

[0038] in, Indicates the overall emotional intensity. , , and These represent the emotional intensity components across four dimensions: physiological signals, behavioral performance, environmental context, and cultural background. , , and Let each represent the dynamic weight coefficient corresponding to each dimension, and satisfy the following conditions: .

[0039] physiological signal components The calculation formula is:

[0040] in, Indicates the first i Physiological signals This represents the feature extraction and normalization function being used. n Indicates the number of physiological signal categories. Indicates the first i Normalized weights for physiological signals, and .

[0041] Behavioral performance components The calculation formula is:

[0042] in, Indicates the first j Confidence scores for class-specific behavioral features Indicates the first j Dynamic weights of class behavioral features m Indicates the number of behavioral feature categories.

[0043] Contextual components The calculation formula is:

[0044] in, Indicates the time factor. Represents spatial location factor, Representing historical context factors, A function that quantifies the intensity of situational factors on the induction of a specific emotion.

[0045] Cultural background weight The calculation formula is:

[0046] in, This indicates dialects and language habits. Indicates cultural content preferences, Representing life experiences and memories, This represents a cultural mapping function that maps cultural characteristics to emotional intensity.

[0047] Furthermore, the dynamic weighting coefficients are dynamically adjusted based on the data quality and contextual relevance of the corresponding dimensions. When the data quality of an emotional intensity component in a certain dimension decreases, the corresponding dynamic weighting coefficient decreases, and the decreased portion is proportionally allocated to other emotional intensity components. In specific contexts, the dynamic weighting coefficients of dimensions relevant to the current context are temporarily increased. For example, when insufficient ambient light leads to a decrease in the reliability of visual data, the weighting ratio of speech and physiological signals is automatically increased (e.g., from the baseline). Adjusted to ).

[0048] The dynamic weight adjustment algorithm is as follows: Python def calculate_dynamic_weights(data_quality): base_weights = {'physio': 0.3, 'behavior': 0.3, 'context': 0.2, 'culture': 0.2} # Adjust weights based on data quality quality_factors = calculate_quality_factors(data_quality) adjusted_weights = {} for modality in base_weights: adjustment = quality_factors[modality] adjusted_weights[modality] = base_weights[modality]* adjustment # Normalization total = sum(adjusted_weights.values()) return {k: v / total for k, v in adjusted_weights.items()} Based on this, step S102 involves formulating an emotional intervention strategy for the target user, including: identifying the target user's emotional state through an emotional inference model and predicting the emotional evolution trend of the target user to obtain the prediction results; generating a personalized emotional profile and dynamic emotional graph for the target user by combining a dynamic knowledge graph to form a deep understanding of the target user's emotional world; and generating an emotional intervention strategy for the target user based on a reinforcement learning algorithm, according to the emotional intensity, prediction results, and dynamic emotional graph.

[0049] Specifically, the process combines dynamic knowledge graphs to generate personalized emotional profiles and dynamic emotional graphs for target users. This includes integrating the target user's physiological signals, emotional state, and current events into triple nodes to construct a personalized emotional profile and generate a dynamic emotional graph.

[0050] In this embodiment, as Figure 3 and Figure 4 As shown, Figure 3 This is a schematic diagram illustrating the structure and state deduction principle of the emotional state transition network in this embodiment of the invention. Figure 4 This is a schematic diagram illustrating the node relationships and construction process of the dynamic knowledge graph in this embodiment of the invention. Based on the collected multidimensional data, an emotional state calculation engine is used to quantify the intensity of emotions, and the theory of "emotional mutual restraint" in traditional Chinese medicine is integrated to construct an emotional inference model based on the seven-emotion state transfer network, realizing the dynamic identification, evolution prediction, and trend analysis of the target user's emotional state; combined with the dynamic knowledge graph, personalized emotional profiles and dynamic emotional graphs of the target user are generated; and personalized emotional intervention strategies are formulated and adjusted in real time through a reinforcement learning-driven strategy generator.

[0051] For example, the network structure includes 7 core emotional state nodes (joy, anger, worry, thought, grief, fear, and shock); the transition probability is calculated based on historical data and traditional Chinese medicine theory, such as the transition probability of "anger → grief" being 0.3; the inference algorithm uses a hidden Markov model to predict the evolution of emotions.

[0052] The code for calculating the transition probability is as follows: Python def calculate_transition_probability(current_state, context): base_prob = state_transition_matrix[current_state] # Adjustments based on environmental factors context_factor = calculate_context_factor(context) # Considering personal history adjustments history_factor = calculate_history_factor(current_state) adjusted_prob = base_prob * context_factor * history_factor return normalized(adjusted_prob) More specifically, the dynamic sentiment graph includes: event nodes, which record the time, type, and environment of the interaction; emotion nodes, which store the emotion intensity, emotion label, and duration output by the emotional state calculation engine; and physiological nodes, which store multidimensional data collected from the target user.

[0053] Furthermore, event nodes and emotion nodes are connected via event-emotion edges, whose weights... The calculation formula is:

[0054] in, This indicates the change in the intensity of emotion. It represents the change over time.

[0055] Emotional nodes and physiological nodes are connected by an emotion-physiological edge, the weight of which is... Physiological signals and emotional intensity Correlation coefficient between .

[0056] In practical applications, Neo4j graph database is used to store user emotional data. All user data is encrypted and stored locally, and the graph structure is updated daily using an incremental learning algorithm. When the probability of a user experiencing "worry" on a weekday evening exceeds a threshold (65%), a "work stress-worry-insomnia" association path is automatically established in the graph.

[0057] The reinforcement learning-driven policy generator uses the emotional state projection results as the state space, intervention methods based on the "emotional counteraction" principle as the action space, and the improvement of the target user's emotional state as the reward signal. It generates personalized intervention strategies through continuous interactive learning. The engine's output is a structured emotional state report, including the current dominant emotion, its intensity, and the most likely next emotion to transform into. Based on the Deep Q-Network reinforcement learning algorithm, the reward function design comprehensively considers both short-term feedback (such as the degree of HRV improvement within 5 minutes after intervention) and long-term effects (such as the trend of emotional intensity changes in the same situation over a consecutive week).

[0058] The code for the state space design is as follows: Python state_dimension = { 'emotion_intensity': 7,# The intensity of the seven emotions 'emotion_trend': 7,# Emotion trend 'physiological_state': 12,# Physiological indicators 'environment_context': 8,# Environment characteristics 'user_profile': 15# User Profile } The code for the reward function is as follows: Python def calculate_reward(previous_state, current_state, intervention): immediate_reward = calculate_immediate_improvement(previous_state,current_state) long_term_reward = calculate_long_term_benefit(intervention) user_feedback = get_user_feedback() total_reward = (0.6 * immediate_reward + 0.3 * long_term_reward + 0.1 * user_feedback) return total_reward Emotional intervention strategies for target users include one or more of the following: selecting music with mutually restraining attributes based on relevant theories of traditional Chinese medicine; generating nostalgic narrative content that conforms to the cultural background of target users; and guiding target users to conduct mindfulness breathing and relaxation training.

[0059] Specifically, its execution methods include, but are not limited to: 1. To drive bionic robots to express empathic expressions and postures; 2. Play music selected based on the principle of "emotional balance"; 3. Generate and tell nostalgic narratives that align with the target users' cultural background and personal experiences; 4. Guide target users to engage in interactive activities such as mindfulness breathing and relaxation training.

[0060] By using various intervention methods (such as music, cultural nostalgia, breathing training, etc.) as "actions" and improving emotional state as "rewards", the intervention strategy most suitable for the current target user is dynamically generated and optimized through continuous trial and error and learning, while collecting feedback information from the target user after the intervention.

[0061] For example, the generated abstract strategy can be transformed into concrete, perceptible multimodal interaction actions as follows: Robotic Facial Expression Control: Through a sophisticated servo control system, the bionic robot platform can reproduce rich and nuanced facial expressions and body language, making emotional interaction more natural and realistic. For example, it uses 42 SMA actuators to control facial expressions, with expression parameters adjusted in real time according to emotional intensity; posture control: dual-degree-of-freedom neck movement simulates the human listening posture.

[0062] Music Intervention: The music intervention subsystem establishes a music library based on the Five Tones theory of Traditional Chinese Medicine, and pre-stores all music resources in local memory to ensure basic services are still available during network outages. Simultaneously, it personalizes the music using psychoacoustic algorithms, automatically matching music attributes according to the principle of emotional balance to ensure effective regulation of the target emotion. For example, for a state of "sorrow," the system selects the Zheng mode music based on the principle of "joy overcoming sorrow"; for excessive "thinking," it selects the Jiao mode music based on the principle of "anger overcoming thinking."

[0063] The music intervention code is as follows: Python MusicIntervention: def __init__(self): self.music_library = load_music_library() self.mapping_rules = load_tcm_mapping_rules() def select_music(self, emotion_state): target_emotion = self.mapping_rules.get_target_emotion(emotion_state) music_list = self.music_library.get_by_emotion(target_emotion) return self.personalize_selection(music_list) Narrative Generation: The narrative generation system employs advanced large language models, such as GPT-based models, to generate personalized stories. It combines the target user's cultural background and personal experiences to create personalized story content with emotional resonance. Furthermore, speech synthesis utilizes emotional speech synthesis technology. For example, relevant opera excerpts can be played for target users who enjoy Peking Opera, evoking positive emotional memories. In addition, this embodiment provides various interactive intervention methods such as mindfulness breathing and relaxation training. For instance, breathing rhythm animations can be displayed on a robot screen, accompanied by soothing background music, allowing target users to choose the appropriate method according to their personal preferences.

[0064] After intervention, the user's emotional state is reassessed. If the intensity of "worry" decreases significantly, the agent is given a positive reward; otherwise, a negative reward is given. Through numerous such interactions, the agent eventually learns the most effective combination of intervention strategies for users in different states.

[0065] Based on the above embodiments, step S104, which optimizes the emotion inference model and emotion intervention strategy according to multidimensional data and feedback information, is as follows: All raw user data is stored on the local terminal, and model training is performed only locally. Only the encrypted model parameters are uploaded to the server for secure aggregation; The server generates an optimized global model and distributes it to each terminal for updates.

[0066] This mechanism enables continuous synergistic optimization of the emotional projection model and intervention strategies while strictly protecting user privacy.

[0067] This embodiment employs a federated learning mechanism to achieve continuous model evolution while protecting user privacy. Specifically, a layered model update strategy is adopted: at the terminal device level, incremental learning is performed using local user data to train the emotion inference model and emotion intervention strategy, maintaining sensitivity to user-specific characteristics. Local model parameters are encrypted and uploaded to the cloud server every 24 hours. At the edge node level, collaborative optimization of user models within the region is achieved. At the cloud level, the server integrates global knowledge through a secure aggregation algorithm, generates an optimized global model, and distributes it to each terminal for updates. The average data size for each update does not exceed 2MB, ensuring the entire process is completed within 10 minutes in a typical home network environment. The model update cycle is set to 24 hours. After each update, the system evaluates the performance of the global model on the validation set. When the emotion recognition accuracy improves by more than 2% or the intervention effect satisfaction improves by more than 5%, the model distribution process is triggered. To prevent model degradation, the historically best model version is retained, and automatic rollback occurs when the performance of the new model deteriorates. To ensure privacy and security, differential privacy technology is used, adding carefully calibrated noise before uploading model parameters to effectively prevent the leakage of personal information. At the same time, a comprehensive model version management and quality monitoring mechanism has been established to ensure the stability and reliability of the update process.

[0068] The local training process code is as follows: Python def local_training(local_data, global_model): local_model = global_model.copy() for epoch in range(LOCAL_EPOCHS): for batch in local_data: loss = compute_loss(local_model, batch) gradients = compute_gradients(loss) # Add differential privacy noise noisy_gradients = add_dp_noise(gradients) local_model.update(noisy_gradients) return compute_model_update(global_model, local_model) In terms of performance optimization, various techniques such as load balancing, data caching strategies, and communication protocol optimization ensure the efficient operation of this method in resource-constrained environments. Especially in emotional interaction scenarios with high real-time requirements, edge computing and lightweight model technologies keep response latency within an acceptable range for users. Regarding reliability assurance, a multi-layered fault-tolerant design is adopted, including sensor redundancy, data verification, and automatic fault recovery. A security protection system permeates all levels, from hardware-level security modules to software-level access control, providing comprehensive protection for user data.

[0069] Through the aforementioned collaborative hardware and software design, real-time emotion perception and intervention with a latency of less than 500ms was achieved in a typical home environment, providing users with a smooth and natural emotional support experience. The local large-capacity storage system not only ensures user privacy and security but also guarantees the system's continuous and reliable operation in unstable network environments, realizing a complete technical loop from "multi-dimensional perception" to "dynamic emotion inference," and then to "personalized intervention execution" and "continuous model optimization." Furthermore, this method, through an intervention content library based on cultural background awareness, enables the system to understand and respond to the unique cultural psychology and emotional expression of Chinese users, enhancing the affinity and depth of empathy in the support. The adoption of a federated learning framework fundamentally eliminates the risk of leakage of original sensitive data, establishing a new paradigm of "data not leaving the domain, model can evolve," clearing obstacles for the compliant application of affective computing technology in sensitive fields.

[0070] This invention provides a personalized emotional intervention device. The personalized emotional intervention device provided by this invention is described below. The personalized emotional intervention device described below can be referred to in correspondence with the personalized emotional intervention method described above. Figure 5 This is a structural block diagram of the personalized emotional intervention device provided by the present invention, as shown below. Figure 5 As shown, this device consists of three core components: intelligent terminal equipment, edge computing nodes, and cloud management platform.

[0071] In practical deployments, smart terminal devices typically take the form of customized home companion robots. Their hardware configuration must meet the demands of real-time multimodal data processing, including high-performance ARM architecture processors, ample memory and storage, and a custom Linux-based operating system. Edge computing nodes are deployed in the user's local network, using Intel Xeon Silver 4210 processors, equipped with 32GB of memory and 1TB of SSD storage, responsible for handling real-time computing tasks such as emotion recognition and instant interactive responses. The cloud platform is based on the Kubernetes container orchestration system, providing elastic computing resources and centralized management functions, supporting large-scale concurrent user access and model training tasks. All system components exchange data via encrypted communication protocols to ensure the security and integrity of data transmission.

[0072] The software implementation adopts a layered architecture design, specifically including: The perception module is used to collect multidimensional data of the target user; the multidimensional data includes physiological signals, behavioral performance, environmental context, and cultural background. The decision-making module is used to build an emotion inference model based on multidimensional data and formulate emotion intervention strategies for target users. The execution module is used to implement emotional intervention strategies and collect feedback information from target users; The optimization module is used to optimize the emotion inference model and emotion intervention strategy based on multidimensional data and feedback information.

[0073] In use, this device first collects multi-dimensional data on the target user's physiological signals, behavioral performance, environmental context, and cultural background. Then, based on this collected multi-dimensional data, the decision-making module integrates the traditional Chinese medicine theory of "emotional interaction" to construct an emotion inference model. This model enables dynamic identification, prediction, and trend analysis of the target user's emotional state, allowing for real-time formulation and adjustment of emotional intervention strategies. The execution module then implements these intervention strategies and collects feedback from the target user. Finally, the optimization module continuously optimizes the emotion inference model through a federated learning mechanism, while protecting user data privacy. This process deeply integrates the traditional Chinese medicine theory of emotional interaction with modern affective computing, constructing an emotional state transition network that enables machines to understand the dynamic dynamics of emotions. This represents a theoretical leap from "static perception" to "dynamic inference and prediction," providing a theoretical foundation for proactive intervention. Moreover, this device solves the problem of personalized and adaptive intervention strategies through a technical closed loop of "multi-dimensional fusion perception + dynamic emotional inference + reinforcement learning decision-making". It continuously learns and adapts to the user's unique response, and can provide precise emotional support "one policy for one person" like a true "traditional Chinese medicine confidant". It realizes a closed loop from emotional perception and state inference to dynamic and precise intervention, and solves the problems of poor accuracy and adaptability of emotional inference in existing related technologies. It is suitable for providing in-depth emotional companionship and mental health support for the elderly with a deep Chinese cultural background.

[0074] Figure 6 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 6 As shown, the electronic device may include: a processor 601, a communications interface 602, a memory 603, and a communication bus 604, wherein the processor 601, the communications interface 602, and the memory 603 communicate with each other via the communication bus 604. The processor 601 can call logical instructions in the memory 603 to execute a personalized emotional intervention method, which includes: Collect multidimensional data from target users; multidimensional data includes physiological signals, behavioral performance, environmental context, and cultural background; Based on multidimensional data, an emotion inference model is constructed, and an emotion intervention strategy is formulated for the target users. Implement emotional intervention strategies and collect feedback from target users; Based on multidimensional data and feedback information, optimize the emotion inference model and emotion intervention strategy.

[0075] Furthermore, the logical instructions in the aforementioned memory 603 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0076] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer is able to perform the personalized emotional intervention methods provided by the above methods, the method including: Collect multidimensional data from target users; multidimensional data includes physiological signals, behavioral performance, environmental context, and cultural background; Based on multidimensional data, an emotion inference model is constructed, and an emotion intervention strategy is formulated for the target users. Implement emotional intervention strategies and collect feedback from target users; Based on multidimensional data and feedback information, optimize the emotion inference model and emotion intervention strategy.

[0077] In another aspect, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to perform the personalized emotion intervention methods provided by the methods described above, the method comprising: Collect multidimensional data from target users; multidimensional data includes physiological signals, behavioral performance, environmental context, and cultural background; Based on multidimensional data, an emotion inference model is constructed, and an emotion intervention strategy is formulated for the target users. Implement emotional intervention strategies and collect feedback from target users; Based on multidimensional data and feedback information, optimize the emotion inference model and emotion intervention strategy.

[0078] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0079] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0080] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A personalized emotional intervention method, characterized in that, include: Collect multidimensional data from target users; The multidimensional data includes physiological signals, behavioral performance, environmental context, and cultural background; Based on the multidimensional data, an emotion inference model is constructed, and an emotion intervention strategy is formulated for the target users. Implement the emotional intervention strategy and collect feedback information from the target users; Based on the multidimensional data and the feedback information, the emotion inference model and the emotion intervention strategy are optimized.

2. The personalized emotional intervention method according to claim 1, characterized in that, Based on the aforementioned multidimensional data, an emotion inference model is constructed, including: Based on the multidimensional data, the emotional intensity of the target user is quantified using an emotional state calculation engine. Based on the intensity of the emotion, an emotion inference model based on the seven emotions transfer network is constructed by integrating relevant theories of traditional Chinese medicine.

3. The personalized emotional intervention method according to claim 2, characterized in that, Based on the aforementioned multidimensional data, the emotional intensity of the target user is quantified using an emotional state calculation engine, including: The emotional intensity components and corresponding dynamic weight coefficients of the target user in four dimensions—physiological signals, behavioral performance, environmental context, and cultural background—are determined respectively. The emotional intensity components of the target user are weighted and fused from four dimensions: physiological signals, behavioral performance, environmental context, and cultural background, to obtain a quantitative value of the target user's emotional intensity.

4. The personalized emotional intervention method according to claim 3, characterized in that, The dynamic weighting coefficients are dynamically adjusted based on the data quality and contextual relevance of the corresponding dimension; When the data quality of a certain emotional intensity component decreases, the corresponding dynamic weight coefficient decreases, and the decreased portion is proportionally allocated to other emotional intensity components. In specific situations, the dynamic weight coefficients of dimensions related to the current situation can be temporarily increased.

5. The personalized emotional intervention method according to claim 2, characterized in that, Develop emotional intervention strategies for the target users, including: The emotional state of the target user is identified by the emotional inference model, and the emotional evolution trend of the target user is predicted to obtain the prediction result. A personalized emotional profile and a dynamic emotional graph of the target user are generated by combining a dynamic knowledge graph. Based on reinforcement learning algorithms, an emotional intervention strategy for the target user is generated according to the emotional intensity, the prediction result, and the dynamic emotional graph.

6. The personalized emotional intervention method according to claim 5, characterized in that, The personalized sentiment profile and dynamic sentiment graph of the target user are generated by combining a dynamic knowledge graph, including: The physiological signals, emotional state, and current events of the target user are integrated into triplet nodes to construct a personalized emotional profile for the target user and generate a dynamic emotional graph.

7. The personalized emotional intervention method according to claim 5, characterized in that, The dynamic sentiment graph includes: Event nodes are used to record the time, type, and environment of interactions; Emotion nodes are used to store the intensity, emotion tag, and duration of emotions output by the emotional state calculation engine; Physiological nodes are used to store multidimensional data collected from the target user.

8. The personalized emotional intervention method according to claim 1, characterized in that, Emotional intervention strategies for the target users include one or more of the following: Music with the property of mutual restraint was selected based on relevant theories of traditional Chinese medicine. Generate nostalgic narrative content that aligns with the cultural background of the target users; Guide the target users to perform mindfulness breathing and relaxation training.

9. A personalized emotional intervention device, comprising a smart terminal device, an edge computing node, and a cloud management platform, characterized in that, Specifically, it includes: The perception module is used to collect multidimensional data from the target user; The multidimensional data includes physiological signals, behavioral performance, environmental context, and cultural background; The decision-making module is used to construct an emotion inference model based on the multidimensional data and formulate emotion intervention strategies for the target users. An execution module is used to execute the emotional intervention strategy and collect feedback information from the target user; An optimization module is used to optimize the emotion inference model and the emotion intervention strategy based on the multidimensional data and the feedback information.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the personalized emotional intervention method as described in any one of claims 1 to 8.

Images