A smart intervention and soothing system based on real-time emotion recognition

By employing multimodal emotion recognition and personalized intervention strategies, the shortcomings in user emotion recognition and intervention on social platforms have been addressed. This has enabled precise quantification and personalized intervention of users' emotional states, thereby enhancing user experience and platform stickiness.

CN123058728APending Publication Date: 2026-11-10EMOTIVERSE TECH DEVELOPMENT (HANGZHOU) CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202510598838.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2025-05-09
Publication Date
2026-11-10

AI Technical Summary

Technical Problem

Existing social platforms and emotion recognition systems are unable to take timely proactive intervention measures, lack personalized responses to changes in users' emotional states, and have insufficient multimodal emotion data fusion and adaptive adjustment capabilities, leading to further deterioration of users' emotions.

Method used

A multimodal emotion recognition engine is used to generate multidimensional emotion vectors. Combined with a risk assessor and an intelligent intervention decision engine, personalized intervention strategies are generated through user profiling and context awareness, and the system is optimized through an emotion feedback module.

Benefits of technology

It enables precise quantification and personalized intervention of users' emotional states, improves the accuracy of emotion recognition and the targeting of intervention, and enhances user experience and platform stickiness.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123058728A_ABST
    Figure CN123058728A_ABST
Patent Text Reader

Abstract

This invention relates to the fields of artificial intelligence, emotion computing, mental health, and social recommendation, and in particular to an intelligent intervention and reassurance system based on real-time emotion recognition. The system includes an emotion recognition engine, a risk assessor, an intelligent intervention decision engine, an intervention content generator, a user interface, and an emotion feedback module. It collects text, voice, and behavioral data from users to generate multi-dimensional emotion vectors, accurately quantifying the user's emotional state. The risk assessor evaluates the risk level based on the multi-dimensional emotion vectors; if a threshold is exceeded, an intervention process is triggered. The intelligent intervention decision engine generates personalized intervention strategies based on user profile data. The intervention content generator generates intervention content accordingly. The user interface presents the intervention content and collects feedback data. The emotion feedback module evaluates the intervention effect and feeds it back to the intelligent intervention decision engine, improving the accuracy and comprehensiveness of emotion recognition and improving the user's emotional state in real time.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence, emotion computing, mental health, and social recommendation, and in particular to an intelligent intervention and soothing system based on real-time emotion recognition. Background Technology

[0002] With the widespread use of the internet and social media platforms, people are spending more and more time communicating and expressing their emotions online. However, most existing social media platforms and emotion recognition systems only passively identify users' emotional states, lacking proactive response mechanisms to changes in these states. When users are experiencing negative emotions (such as loneliness, depression, or anxiety), the platform cannot intervene in a timely manner, which may lead to further deterioration of the user's emotions, or even user churn or psychological crises.

[0003] While individual technologies such as natural language processing, speech recognition, and facial expression recognition have made some progress, there is still no complete solution that can integrate multimodal emotion data to address users' psychological problems and provide emotional comfort. Most emotion recognition technologies can only perform simple emotion classification, failing to capture the complexity and multidimensionality of emotions, and lack personalized intervention mechanisms for different users and situations. Furthermore, most existing systems adopt static intervention strategies, unable to adaptively adjust and optimize based on the intervention results. Summary of the Invention

[0004] The purpose of this invention is to provide a system that triggers intelligent intervention and soothing strategies in real time based on multimodal emotion recognition results, so as to solve the problems existing in the prior art.

[0005] This invention proposes an intelligent intervention and soothing system based on real-time emotion recognition, comprising:

[0006] An emotion recognition engine is used to collect users' text, voice, and behavioral data, and generate multi-dimensional emotion vectors that represent users' emotional states.

[0007] A risk assessor, connected to the emotion recognition engine, is used to receive the multi-dimensional emotion vector, assess the risk level of the multi-dimensional emotion vector, and trigger an intervention process when the risk level exceeds a preset threshold.

[0008] The intelligent intervention decision engine, connected to the risk assessor, is used to receive the multidimensional emotion vector and risk level, and generate personalized intervention strategies based on user profile data.

[0009] An intervention content generator, connected to the intelligent intervention decision engine, is used to receive the personalized intervention strategy and generate corresponding intervention content.

[0010] The user interface is connected to the intervention content generator and is used to present the intervention content and collect user feedback data.

[0011] An emotion feedback module, connected to the user interface and the intelligent intervention decision engine, is used to receive feedback data from the user, evaluate the intervention effect, and feed the intervention effect back to the intelligent intervention decision engine.

[0012] Preferably, the emotion recognition engine includes:

[0013] The text analysis unit is used to analyze keywords, sentiment word weights, and subjective biases in user text.

[0014] The speech analysis unit is used to analyze the user's speech rate, pitch, pause frequency, and intonation fluctuations.

[0015] The behavior analysis unit is used to analyze users' login time, activity level, operation frequency, and click behavior;

[0016] The multimodal fusion unit, connected to the text analysis unit, the speech analysis unit, and the behavior analysis unit, is used to fuse the analysis results and generate a multidimensional emotion vector containing feelings of pleasure, anxiety, loneliness, and depression.

[0017] Preferably, the risk assessor includes:

[0018] Threshold setting unit, used to store risk thresholds for each emotion dimension;

[0019] A personalized calibration unit is used to adjust the risk threshold based on the user's historical emotion data;

[0020] A risk calculation unit, connected to the threshold setting unit and the personalized calibration unit, is used to compare the multidimensional emotion vector with the adjusted risk threshold and generate a risk assessment result.

[0021] A risk level classification unit, connected to the risk calculation unit, is used to classify the risk assessment results into multiple risk levels.

[0022] Preferably, the intelligent intervention decision engine includes:

[0023] A strategy library is used to store various intervention strategies and their corresponding feature vectors.

[0024] A strategy matching unit, connected to the strategy library, is used to calculate the matching degree between the multidimensional emotion vector and the feature vectors of each intervention strategy;

[0025] The user profile database is used to store users' intervention preferences and historical response data.

[0026] An intervention effect prediction unit, connected to the user profile database, is used to predict the potential effects of each intervention strategy on the current user.

[0027] The strategy optimization unit, connected to the strategy matching unit and the intervention effect prediction unit, is used to combine the matching degree and the prediction effect to determine the final personalized intervention strategy.

[0028] A time-sensitive adjustment unit, connected to the strategy optimization unit, is used to adjust the personalized intervention strategy according to the current time period.

[0029] Preferably, the strategy matching unit is used for:

[0030] Calculate the similarity between the multidimensional emotion vector and the feature vectors of each intervention strategy;

[0031] A weighted coefficient is applied to the similarity to generate a preliminary matching score;

[0032] The initial matching score is adjusted based on the risk level;

[0033] Generate a sorted list of candidate strategies.

[0034] Preferably, the intervention effect prediction unit is used for:

[0035] Obtain historical response data of users to different types of interventions from the user profile database;

[0036] Identify users' intervention sensitivity characteristics;

[0037] Based on the intervention sensitivity characteristics and current emotional state, predict the expected effects of each intervention strategy;

[0038] The expected effect is adjusted based on the user's time-period response characteristics.

[0039] Preferably, the intervention content generator includes:

[0040] The emotion channel generation unit is used to generate intervention content for emotion regulation;

[0041] The cognitive channel generation unit is used to generate intervention content for cognitive restructuring;

[0042] A content template library, connected to the emotion channel generation unit and the cognitive channel generation unit, is used to provide basic expression templates;

[0043] A personalized adjustment unit, connected to the emotion channel generation unit and the cognitive channel generation unit, is used to adjust the intervention content according to user characteristics;

[0044] A security check unit, connected to the emotion channel generation unit and the cognitive channel generation unit, is used to filter content that may exacerbate negative emotions.

[0045] Preferably, the emotion feedback module includes:

[0046] The emotion change monitoring unit is used to monitor changes in emotion vectors before and after intervention;

[0047] The user interaction analysis unit is used to analyze the interaction between users and the intervention content;

[0048] An effectiveness evaluation unit, connected to the emotion change monitoring unit and the user interaction analysis unit, is used to comprehensively evaluate the intervention effect;

[0049] A strategy adjustment unit, connected to the effect evaluation unit, is used to generate strategy optimization suggestions based on the intervention effect;

[0050] The knowledge accumulation unit, connected to the effect evaluation unit, is used to store effective intervention cases.

[0051] As a preferred option, it also includes:

[0052] The nighttime protection module is connected to the intelligent intervention decision engine and is used to automatically activate the emotion protection mode during nighttime hours, adjusting the intervention strategy and content.

[0053] The social recommendation module is connected to the intelligent intervention decision engine and is used to recommend emotional companionship users or guide them into emotional healing social spaces when appropriate.

[0054] The task guidance module, connected to the intelligent intervention decision engine, is used to generate emotion regulation tasks and guide users to complete them.

[0055] Preferably, the user interface includes:

[0056] Anthropomorphic interactive units are used to present comforting characters with personality traits;

[0057] Multimedia presentation unit, used to present intervention content in the form of text, voice, images, etc.;

[0058] The privacy control unit allows users to set the scope of emotional data sharing and intervention receiving preferences;

[0059] The feedback collection unit, connected to the emotion feedback module, is used to collect user response data and subjective evaluations of the intervention content.

[0060] The beneficial effects of this invention include:

[0061] 1. By using multimodal emotion recognition and multidimensional emotion vector representation, the system can accurately quantify the user's emotional state, thereby improving the accuracy and comprehensiveness of emotion recognition;

[0062] 2. Establish a personalized emotional risk assessment mechanism to promptly identify high-risk emotional states and trigger intervention;

[0063] 3. By combining user profiles and contextual awareness, personalized intervention strategies can be selected to improve the targeting and effectiveness of interventions;

[0064] 4. By generating content through a dual-channel intervention of emotion and cognition, it comprehensively covers users' emotion regulation needs;

[0065] 5. Establish a feedback mechanism for intervention effects and a strategy optimization mechanism to enable the system to learn and evolve on its own;

[0066] 6. Ultimately, the goal is to improve users' emotional well-being, enhance user experience, and increase platform stickiness. Attached Figure Description

[0067] Figure 1 This is a diagram illustrating the overall architecture of the intelligent intervention and soothing system based on real-time emotion recognition according to the present invention.

[0068] Figure 2 This is a schematic diagram of the structure of the emotion recognition engine of the present invention;

[0069] Figure 3 This is a flowchart of the risk assessment tool of the present invention;

[0070] Figure 4 This is a diagram showing the structure and data flow of the intelligent intervention decision engine of this invention;

[0071] Figure 5 This is a schematic diagram illustrating the working principle of the content generator of this invention.

[0072] Figure 6 This is a flowchart of the emotion feedback module of the present invention;

[0073] Figure 7 This is a flowchart illustrating the overall workflow of the system of the present invention. Detailed Implementation

[0074] Please refer to the attached document. Figure 1-7 The specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. It should be noted that the present invention is not limited to the specific embodiments described below, and those skilled in the art can make various modifications and variations based on the technical solutions of the present invention without departing from the scope of protection of the present invention.

[0075] Reference Figure 1As shown, the present invention provides an intelligent intervention and soothing system based on real-time emotion recognition, including an emotion recognition engine 1, a risk assessor 2, an intelligent intervention decision engine 3, an intervention content generator 4, a user interface 5, and an emotion feedback module 6.

[0076] The emotion recognition engine 1 collects users' text, voice, and behavioral data, generating multi-dimensional emotion vectors representing users' emotional states through multimodal analysis. The risk assessor 2, connected to the emotion recognition engine 1, receives the multi-dimensional emotion vectors, assesses their risk levels, and triggers an intervention process when the risk level exceeds a preset threshold. The intelligent intervention decision engine 3, connected to the risk assessor 2, receives the multi-dimensional emotion vectors and risk levels, and generates personalized intervention strategies based on user profile data. The intervention content generator 4, connected to the intelligent intervention decision engine 3, receives the personalized intervention strategies and generates corresponding intervention content. The user interface 5, connected to the intervention content generator 4, presents the intervention content and collects user feedback data. The emotion feedback module 6, connected to the user interface 5 and the intelligent intervention decision engine 3, receives user feedback data, evaluates the intervention effect, and feeds the intervention effect back to the intelligent intervention decision engine 3, forming a closed-loop optimization.

[0077] Reference Figure 2 As shown, the emotion recognition engine 1 includes a text analysis unit 11, a speech analysis unit 12, a behavior analysis unit 13, and a multimodal fusion unit 14.

[0078] The text analysis unit 11 is used to analyze keywords, sentiment word weights, and subjective tendencies in user text. In one embodiment of the invention, the text analysis unit 11 employs natural language processing technology, using a sentiment dictionary and semantic analysis algorithms to extract sentiment features from the text. For example, when a user sends a message on a social platform such as "I've been feeling lonely and anxious lately, nobody understands how I feel," the system identifies key sentiment words such as "lonely," "anxious," and "nobody understands," and analyzes the context to determine a strong negative sentiment tendency. Preferably, the text analysis unit 11 also considers the text's context and the user's expression habits, improving the accuracy of sentiment analysis by establishing a personalized language model.

[0079] The speech analysis unit 12 analyzes the user's speech rate, pitch, pause frequency, and intonation fluctuations. In practical applications, such as when a user expresses emotions through voice messages, the speech analysis unit 12 extracts various acoustic feature parameters. When a user is depressed, their speech typically exhibits a significantly slower speech rate (more than 30% reduction in words per minute), a lower pitch (fundamental frequency reduced by 15-20Hz), and reduced energy (sound intensity reduced by 5-10dB). Conversely, when a user is anxious, their speech may exhibit a faster speech rate (more than 25% increase in words per minute), a higher pitch (fundamental frequency increased by 10-15Hz), and more frequent and irregular pauses (pause frequency increased by more than 40%). These changes in acoustic features provide important evidence for the system's emotion recognition.

[0080] The behavior analysis unit 13 is used to analyze users' login time, activity level, operation frequency, and click behavior. For example, the system may detect a significant increase in the user's recent nighttime (11:00 PM - 3:00 AM) login frequency (from twice a week to every night), a shift in browsing content from positive social information to topics related to loneliness and depression, and abnormal operation rhythms (such as scrolling quickly but rarely stopping to read, or spending long periods on negative content). These changes in behavioral patterns are often indirect indicators of changes in emotional state. Preferably, the behavior analysis unit 13 also incorporates time context factors, such as weekday / weekend, morning / afternoon / nighttime differences, to improve the accuracy of the analysis.

[0081] The multimodal fusion unit 14 is connected to the text analysis unit 11, the speech analysis unit 12, and the behavior analysis unit 13, and is used to fuse the analysis results of different modalities to generate a multidimensional emotion vector containing feelings of pleasure, anxiety, loneliness, and depression. In this invention, the multimodal fusion adopts a weighted fusion method, taking into account the reliability and contextual relevance of data from different modalities.

[0082] Specifically, the multidimensional emotion vector can be represented as:

[0083] V = [v pleasure ,v anxiety ,v loneliness ,v depression ],

[0084] Where V is the emotion vector; v pleasure Indicates level of pleasure, reflecting the degree of positive emotions experienced by the user; v anxiety Indicates anxiety level, reflecting the degree of tension and unease the user; v loneliness Indicates loneliness level, reflecting the degree of social isolation a user feels; v depression This indicates the level of depression, reflecting the degree of a user's depressed mood. Each component ranges from 0 to 100, where 0 represents the lowest value for that emotional dimension and 100 represents the highest value.

[0085] The process of multimodal fusion can be represented as:

[0086]

[0087] Among them, v i This represents the i-th component of the final emotion vector V; This represents the value of the i-th sentiment component obtained from text analysis; This represents the value of the i-th emotion component obtained from speech analysis; α represents the value of the i-th emotion component obtained from behavioral analysis; i β i γ i Let represent the weight coefficients of text, speech, and behavioral modalities on this emotion component, respectively, and satisfy α. i +β i +γ i =1, to ensure that the merged value is still in the range of 0-100.

[0088] In practical applications, these weighting coefficients are dynamically adjusted based on data quality and reliability. For example, in noisy environments, voice data may be less reliable, and the system will automatically reduce the beta. i Value; when users send only emojis instead of text, the weight α of text analysis. i It may decrease; when user behavior data is very rich and has obvious patterns, γ i The value will increase accordingly. Through this dynamic weight adjustment, the system can maintain a high accuracy rate in emotion recognition under different contexts.

[0089] For example, when a user sends multiple messages containing words like "lonely" and "helpless" late at night, voice analysis shows a significant slowdown in speech rate. Simultaneously, behavioral data indicates that the user repeatedly views negative content. The system may then generate an emotion vector such as [35, 65, 78, 60], representing low pleasure (35), high anxiety (65), high loneliness (78), and high depression (60). This emotion vector will trigger the system's risk assessment process.

[0090] Reference Figure 3 As shown, the risk assessor 2 includes a threshold setting unit 21, a personalized calibration unit 22, a risk calculation unit 23, and a risk level classification unit 24.

[0091] The threshold setting unit 21 is used to store risk thresholds for each emotional dimension. In a preferred embodiment of the invention, the system sets four benchmark thresholds based on a large amount of mental health data and recommendations from professional psychological counselors: the pleasure threshold is set to 30 (below this value is considered a risk signal), and the anxiety, loneliness, and depression thresholds are all set to 70 (above this value is considered a risk signal). These benchmark thresholds are determined based on research findings in emotional psychology and a large amount of user data, representing general standards for risk identification.

[0092] The personalized calibration unit 22 is used to adjust the risk threshold based on the user's historical emotional data. In practical applications, there are significant differences in the emotional baseline and fluctuation range among different users. For example, some users usually experience large emotional fluctuations, while others are relatively stable; some users are accustomed to expressing strong emotions, while others are more reserved. Therefore, this invention analyzes the user's emotional data over 30 days, calculates the mean and standard deviation of each dimension, and performs personalized threshold adjustment.

[0093] The adjusted threshold can be expressed as:

[0094]

[0095] Among them, T i adj This represents the adjusted threshold for the i-th dimension; This represents the baseline threshold for that dimension; σ i The standard deviation of a user's sentiment value in this dimension reflects the degree of fluctuation in the user's emotions; λ i This is an adjustment coefficient, typically ranging from -0.5 to 0.5, used to control the magnitude of the adjustment.

[0096] For example, for users with large mood swings (such as σ) anxiety =15), the system may adjust its anxiety threshold from the baseline of 70 to 70 + 0.3 × 15 = 74.5 to reduce false alarms; while for users who are more reserved in expressing their emotions (such as σ anxiety =5), the system may adjust the threshold to 70 - 0.3 × 5 = 68.5 to improve sensitivity. This personalized adjustment enables the system to more accurately identify each user's emotional risk state.

[0097] The risk calculation unit 23 is connected to the threshold setting unit 21 and the personalized calibration unit 22, and is used to compare the multidimensional emotion vector with the adjusted risk threshold to generate a risk assessment result. Specifically, the risk calculation unit 23 calculates the degree of deviation between the emotion vector and the threshold to generate a risk score.

[0098] The risk score is calculated using the following formula:

[0099]

[0100] Where R represents the overall risk score; v i T represents the i-th component of the emotion vector; i adj Indicates the corresponding adjusted threshold; w i This represents the weight coefficient of this dimension, reflecting the degree to which different emotion dimensions contribute to risk; the max(0,x) function ensures that only dimensions exceeding the threshold are included in the risk score; n is the dimension of the emotion vector (4 in this invention). It should be noted that the calculation method for pleasure is slightly different, as follows: Because the lower the level of pleasure, the greater the risk.

[0101] For example, when a user's emotion vector is [35, 65, 78, 60], the adjusted threshold is [30, 72, 68, 70], and the weighting coefficient is [0.25, 0.25, 0.25, 0.25], the system will calculate:

[0102] Pleasure level risk component: 0.25×max(0,(30-35) / 30×100)=0 (not exceeding the risk threshold)

[0103] Anxiety risk component: 0.25×max(0,(65-72) / 72×100)=0 (not exceeding the risk threshold)

[0104] Loneliness risk component: 0.25 × max(0, (78-68) / 68 × 100) = 0.25 × 14.71 = 3.68

[0105] Depression risk component: 0.25×max(0,(60-70) / 70×100)=0 (not exceeding the risk threshold)

[0106] Overall risk score: 0 + 0 + 3.68 + 0 = 3.68

[0107] The risk level classification unit 24 is connected to the risk calculation unit 23 and is used to classify the risk assessment results into multiple risk levels. In this invention, the risk levels are divided into 5 levels: Level 0 (normal), Level 1 (mild fluctuation), Level 2 (critical), Level 3 (severely negative), and Level 4 (emotional breakdown warning). The specific classification criteria are as follows:

[0108] A risk score R < 20: Level 0 (normal) indicates that the user's emotional state is good and no intervention is needed.

[0109] 20≤R<40: Level 1 (Mild Fluctuation), indicating the user is experiencing slight emotional fluctuations; the system continues to observe. 40≤R<60: Level 2 (Critical), indicating the user's emotional state is approaching the risk boundary; mild intervention is initiated.

[0110] 60≤R<80: Level 3 (Severe Negative), indicating that the user is in a state of significant negative emotion and requires active intervention.

[0111] R≥80: Level 4 (Emotional Breakdown Warning), indicating that the user may be in an emotional crisis and requires immediate intervention and consideration of professional help.

[0112] When the risk level reaches level 2 or above, the system will trigger the corresponding intervention process. For example, in the previous example, the calculated risk score was 3.68, corresponding to a level 0 (normal) risk level. The system will continue to monitor but will not initiate intervention for the time being. However, if the user's loneliness level continues to rise to 85, exceeding the threshold by a significant margin, the risk score may rise above 40, triggering a level 2 risk alert and intervention process.

[0113] Reference Figure 4 As shown, the intelligent intervention decision engine 3 includes a strategy library 31, a strategy matching unit 32, a user profile database 33, an intervention effect prediction unit 34, a strategy optimization unit 35, and a time-sensitive adjustment unit 36.

[0114] The strategy library 31 stores various intervention strategies and their corresponding feature vectors. In this invention, the strategy library contains multiple types of intervention strategies, such as textual reassurance, verbal comfort, task guidance, and social recommendation. Each strategy defines a feature vector describing its intervention effectiveness on various emotional dimensions.

[0115] The policy feature vector can be represented as:

[0116] S = [s pleasure ,s anxiety ,s loneliness ,s depression ],

[0117] Where S is the policy feature vector; s pleasure This indicates the effectiveness of the strategy in enhancing pleasure; s anxiety This indicates the effectiveness of the strategy in alleviating anxiety; s loneliness This indicates the effectiveness of the strategy in alleviating loneliness; depression This indicates the effectiveness of the strategy in alleviating depression. Each component ranges from 0 to 1, with larger values ​​indicating stronger intervention effectiveness in the corresponding dimension.

[0118] For example, the feature vector of the "light social companionship interaction" strategy might be [0.6, 0.8, 0.4, 0.3], indicating that this strategy is effective in improving pleasure (0.6) and alleviating anxiety (0.8), but relatively weak in alleviating loneliness (0.4) and depression (0.3). Conversely, the feature vector of the "deep companionship social" strategy might be [0.5, 0.3, 0.9, 0.6], indicating that it is particularly effective in alleviating loneliness (0.9), and also has some effect on depression (0.6) and improving pleasure (0.5), but its effect on alleviating anxiety (0.3) is relatively weak.

[0119] The strategy matching unit 32 is connected to the strategy library 31 and is used to calculate the matching degree between the multidimensional emotion vector and the feature vectors of each intervention strategy. Specifically, the strategy matching unit 32 performs the following steps: first, it calculates the similarity between the emotion vector and the feature vectors of each strategy; then, it applies a weighting coefficient to the similarity to generate a preliminary matching score; next, it adjusts the preliminary matching score in combination with the risk level; and finally, it generates a sorted list of candidate strategies.

[0120] The matching degree is calculated using a modified cosine similarity formula:

[0121]

[0122] Where M(V,S) represents the matching degree between the emotion vector V and the policy feature vector S; v i s represents the i-th component of the emotion vector; i w represents the i-th component of the policy feature vector; i This represents the weight coefficient for this dimension, which is dynamically adjusted according to the risk level; n is the vector dimension (4 in this invention). Weight coefficient w i The principle for setting the criteria is: the higher the risk level of the emotional dimension, the greater its weight, so as to ensure that the intervention strategy prioritizes the emotional issues that require the most attention.

[0123] For example, when the user's emotion vector is [35,65,78,60], the system identifies loneliness (78) as the most prominent issue. In this case, if there are two strategies: "light social companionship interaction" [0.6,0.8,0.4,0.3] and "deep companionship social interaction" [0.5,0.3,0.9,0.6], the system will assign a higher weight to the loneliness dimension (e.g., w = [0.2,0.2,0.4,0.2]), and the system will calculate that "deep companionship social interaction" has a higher matching degree because it has a stronger intervention effect on loneliness.

[0124] User profile database 33 is used to store users' intervention preferences and historical response data. In this invention, the user profile includes various information, such as the user's sensitivity indicators to different intervention types, historical intervention effect records, and time pattern characteristics. For example, the system records the user's historical responses to different intervention methods such as text reassurance, voice comfort, task guidance, and social recommendations, forming a personalized intervention sensitivity profile.

[0125] The intervention effect prediction unit 34 is connected to the user profile database 33 to predict the potential effects of each intervention strategy on the current user. Specifically, the intervention effect prediction unit 34 performs the following steps: first, it obtains historical response data of users to different intervention types from the user profile database; then, it identifies the user's intervention sensitivity characteristics; next, based on the intervention sensitivity characteristics and the current emotional state, it predicts the expected effects of each intervention strategy; finally, it adjusts the expected effects by combining the user's time-period response characteristics.

[0126] The formula for calculating the expected results is as follows:

[0127] E(S,U)=M(V,S)·P(U,S)·T(t),

[0128] Where E(S,U) represents the expected effect of strategy S on user U; M(V,S) represents the matching degree between emotion vector V and strategy S; P(U,S) represents user U's historical response score to strategy S, reflecting the user's sensitivity to this type of intervention; T(t) represents the time adjustment factor, reflecting the difference in intervention effect at different times; t is the current hour (0-23). ​​For example, suppose a user's historical response score to the "deep companionship social" strategy is P(U,S) = 0.85 (indicating a very good effect), the current time is 10 pm (t = 22), and the time adjustment factor T(22) = 1.2 (indicating a more positive response to social intervention in the evening), then the expected effect of this strategy may be E(S,U) = 0.78 × 0.85 × 1.2 = 0.80, which is a relatively high expected effect score.

[0129] The strategy optimization unit 35 is connected to the strategy matching unit 32 and the intervention effect prediction unit 34, and is used to comprehensively consider the matching degree and the predicted effect to determine the final personalized intervention strategy. In this invention, the strategy optimization unit 35 considers not only the effect of a single strategy, but also the synergistic effect of strategy combinations to achieve comprehensive intervention.

[0130] For example, when a user is identified to have both high loneliness and moderate anxiety, the system may combine two strategies—"deep companionship-based social interaction" (primarily targeting loneliness) and "breathing relaxation guidance" (primarily targeting anxiety)—to form a comprehensive intervention plan. The strategy optimization unit 35 also considers the order in which the strategies are implemented, such as first alleviating anxiety through relaxation guidance and then introducing social interaction to alleviate loneliness, in order to achieve a better overall effect.

[0131] The time-sensitive adjustment unit 36 ​​is connected to the strategy optimization unit 35 and is used to adjust personalized intervention strategies according to the current time period. In this invention, the system recognizes significant differences in users' emotional states and response patterns at different times, thus introducing a time-sensitive adjustment mechanism. For example, users may respond better to encouraging and activating interventions in the morning (7:00-9:00), while they may respond better to soothing and relaxing interventions at night (22:00-2:00). Especially during the late night period (22:30-2:00), the system automatically adopts a gentler and more soothing intervention method to avoid overstimulation.

[0132] The formula for calculating the time adjustment factor is:

[0133]

[0134] Where T(t) represents the time adjustment factor; t is the current hour (0-23); The function is a periodic function, causing the adjustment factor to change regularly throughout the day. It reaches its maximum value of 1.5 at noon (t-12) and its minimum value of 0.5 at midnight (t0 or t=24), reflecting the user's response characteristics at different times of day.

[0135] In practical applications, for example, when a user experiences emotional problems late at night (t=1), with a time adjustment factor T(1)≈0.55, the system will automatically reduce the weight of activation strategies such as social recommendations and increase the weight of soothing strategies such as relaxation guidance to avoid overstimulating the user. This time-sensitive adjustment makes the system's intervention more aligned with the user's actual needs and state.

[0136] Reference Figure 5 As shown, the intervention content generator 4 includes an emotion channel generation unit 41, a cognitive channel generation unit 42, a content template library 43, a personalized adjustment unit 44, and a safety check unit 45.

[0137] The emotion channel generation unit 41 is used to generate intervention content for emotion regulation. In this invention, the emotion channel mainly focuses on the user's emotional catharsis, empathic support, and emotional stability, adopting a warm and understanding language style. When the user exhibits obvious negative emotions, the emotion channel will first provide understanding and support, such as "I can feel how you feel right now. This feeling is natural and common," helping the user accept their emotional state and reduce emotional stress.

[0138] For example, when the system detects that a user sent a message late at night expressing strong feelings of loneliness, the emotion channel might generate content like: "I notice you may be feeling a little lonely right now. This feeling can indeed become more intense late at night. Your feelings are perfectly normal and valid; I'm here to keep you company during this time." This kind of content directly addresses the user's emotional state, providing emotional support and a sense of companionship.

[0139] The cognitive channel generation unit 42 is used to generate intervention content for cognitive restructuring. In this invention, the cognitive channel mainly focuses on the user's thinking adjustment, perspective shift, and solution provision, employing a rational and constructive language style. After the emotional channel helps the user stabilize their emotions, the cognitive channel guides the user to view the problem from different angles, breaking negative thinking patterns.

[0140] For example, for the same user expressing loneliness, the cognitive channel might generate content like this: "Sometimes, we focus so much on the feeling of loneliness that we neglect the people and things that still connect us. Try recalling any small, pleasant moments or interactions you've had recently. Or, you could try doing something gentle, like listening to a favorite song or writing down your thoughts. These little things can sometimes help us reconnect with ourselves and the world." This type of content aims to guide users to shift their perspective, discover positive factors, and provide specific coping suggestions.

[0141] The content template library 43 is connected to the emotion channel generation unit 41 and the cognitive channel generation unit 42, and is used to provide basic expression templates. In this invention, the content template library contains a large number of expression templates reviewed by professional psychological counselors, categorized according to different emotional states and intervention strategies. Each template is marked with the applicable emotional range, intervention type, and language style.

[0142] For example, an emotional channel template for loneliness might include:

[0143] "I understand the feelings of loneliness / isolation / misunderstood, especially at this time, when those feelings may be even stronger."

[0144] Feeling lonely is a very natural human emotion, and many people experience moments like this.

[0145] "Your feelings are completely valid, and I'm here with you through this moment."

[0146] Cognitive pathway templates for loneliness may include:

[0147] Sometimes when we focus on loneliness, we may overlook the positive connections and small joys that still exist.

[0148] Try to recall if there have been any small things recently that made you feel [warm / connected / understood]?

[0149] "Perhaps you could try [specific activity suggestions], which can sometimes help reconnect with yourself and others."

[0150] The personalized adjustment unit 44 is connected to the emotion channel generation unit 41 and the cognitive channel generation unit 42, and is used to adjust the intervention content according to user characteristics. In this invention, personalized adjustment takes into account the user's language habits, expression style, and historical interaction records, so that the generated content better meets the user's expectations and needs. For example, for users who prefer concise expression, the system will generate more refined content; for users who like detailed explanations, the system will provide richer explanations; for users familiar with psychological concepts, the system may use more professional terminology.

[0151] The safety check unit 45 is connected to the emotion channel generation unit 41 and the cognitive channel generation unit 42, and is used to filter content that may exacerbate negative emotions. In this invention, the safety check is an important step in ensuring the quality of intervention. For example, the system will avoid using words or expressions that may trigger stronger negative emotions; for high-risk users (such as level 3-4 risk), the system will pay special attention to avoiding expressions that may be misunderstood as downplaying their suffering; the system will also ensure that the suggested activities or methods are safe and feasible and will not put additional pressure on the user.

[0152] The content generation process is as follows: First, the system selects a suitable basic template from the content template library based on the user's emotional state (e.g., high loneliness, low pleasure) and the selected intervention strategy (e.g., "deep companionship-based social interaction"). Then, based on the user's personal characteristics and historical interaction data, the language style and content details of the template are adjusted. Next, content is generated for both the emotional channel (providing emotional support) and the cognitive channel (providing thought guidance). Finally, after a security check to ensure that the content will not have any unexpected negative impact, the final intervention content is output.

[0153] Reference Figure 6 As shown, the emotion feedback module 6 includes an emotion change monitoring unit 61, a user interaction analysis unit 62, an effect evaluation unit 63, a strategy adjustment unit 64, and a knowledge accumulation unit 65.

[0154] The emotion change monitoring unit 61 is used to monitor changes in emotion vectors before and after the intervention. In this invention, the system continuously monitors the user's emotional state after the intervention and evaluates the intervention effect through changes in emotion vectors. This monitoring is not intrusive, but rather analyzes the user's natural interaction data (such as text expression, voice features, and behavioral patterns).

[0155] Formula for calculating the rate of change in mood:

[0156]

[0157] Where ΔV represents the rate of change in mood; V before V represents the emotional vector before intervention; after This represents the emotional vector after intervention; the molecule (V) after -V before ) represents the vector difference. When calculating it, we need to consider the expected change direction of different dimensions: the expected level of pleasure increases, while the expected level of anxiety, loneliness, and depression decreases.

[0158] For example, if a user's emotion vector before intervention is [35, 65, 78, 60], and after intervention the emotion vector becomes [45, 55, 65, 55], then the rates of change are: happiness increased by 28.6%, anxiety decreased by 15.4%, loneliness decreased by 16.7%, and depression decreased by 8.3%. This indicates that the intervention is most effective in increasing happiness and reducing loneliness.

[0159] The user interaction analysis unit 62 is used to analyze the interaction between users and the intervention content. In this invention, interaction metrics include response time (how long it takes for a user to respond after receiving the intervention content), message count (the number of messages the user responds with), and completion rate (the user's completion rate of the recommended task). For example, if a user responds quickly after receiving the intervention content and sends multiple positive messages, this usually indicates a good intervention effect; while if a user does not respond for a long time or responds briefly and negatively, it may indicate a poor intervention effect.

[0160] The effectiveness evaluation unit 63 is connected to the emotion change monitoring unit 61 and the user interaction analysis unit 62 to comprehensively evaluate the intervention effect. In this invention, the effectiveness evaluation adopts multi-dimensional indicators, considering not only emotion changes, but also interaction quality and user subjective evaluation.

[0161] Formula for comprehensive evaluation of intervention effectiveness:

[0162] E = α·ΔV + β·I + γ·C,

[0163] Where E represents the overall intervention effectiveness score; ΔV represents the rate of change in mood, reflecting the direct impact of the intervention on emotional state; I represents the interaction index, reflecting the user's response and participation in the intervention; C represents the task completion rate (if task recommendations are provided); α, β, and γ are weighting coefficients, satisfying α + β + γ = 1, used to balance the importance of each factor. In practical applications, these weights may be set to α = 0.5, β = 0.3, and γ = 0.2, indicating that mood change is the most important evaluation indicator.

[0164] For example, if the average user's mood change rate is 17.3%, the interaction index is 0.75 (indicating good interaction), and the task completion rate is 0.8 (indicating that most of the recommended task was completed), then the overall score is E = 0.5 × 17.3 + 0.3 × 75 + 0.2 × 80 = 8.65 + 22.5 + 16 = 47.15. This score can be compared with historical intervention effects to assess the relative effectiveness of the current intervention.

[0165] The strategy adjustment unit 64 is connected to the effect evaluation unit 63 and is used to generate strategy optimization suggestions based on the intervention effect. In this invention, the strategy adjustment adopts a reinforcement learning method to adjust the strategy selection probability according to the intervention effect.

[0166] Strategy weight adjustment formula:

[0167]

[0168] in, This indicates the adjusted strategy weights; Indicates the strategy weights before adjustment; E represents the intervention effectiveness score. This represents the historical average performance score. This represents the difference between the effect of this intervention and the average level; η is the learning rate, usually around 0.1, to control the magnitude of the adjustment.

[0169] For example, if a strategy currently has a weight of 0.25, the current intervention effect score is 65, the historical average effect score is 50, and the learning rate is 0.1, then the adjusted weight is 0.25 + 0.1 × (65 - 50) = 0.25 + 0.1 × 15 = 0.25 + 1.5 = 0.265. This means the system will slightly increase the probability of selecting this strategy because it has shown above-average performance.

[0170] The knowledge accumulation unit 65 is connected to the effect evaluation unit 63 and is used to store effective intervention cases. In this invention, the system records successful intervention cases, including information such as the user's emotional state, the selected intervention strategy, the intervention content, and the effect evaluation, forming a knowledge base. These cases can provide reference and guidance for interventions in similar situations later. For example, the system may find that the combination strategy of "gentle guidance + creative expression task" is particularly effective for young users who feel lonely at night; while the combination of "breathing relaxation + mind restructuring" is more suitable for workplace users who experience anxiety during the workday. This accumulated knowledge continuously enriches the system's intervention capabilities.

[0171] The user interface 5 includes a human-like interaction unit 51, a multimedia presentation unit 52, a privacy control unit 53, and a feedback collection unit 54.

[0172] The anthropomorphic interaction unit 51 is used to present a comforting role with personality traits. In this invention, the system can interact with users through virtual characters, enhancing emotional connection and trust. For example, the system may present a character with warm qualities such as a "guardian star" or an "emotional partner." These characters can have different personality types, such as "quiet" (gentle and soothing communication style), "encouraging" (positive and motivating communication style), and "companion" (empathetic and supportive communication style), to suit different user preferences. The anthropomorphic design makes the intervention process more natural and intimate, reduces users' psychological defenses, and increases the acceptance of the intervention.

[0173] The multimedia presentation unit 52 is used to present intervention content in the form of text, voice, images, etc. In this invention, the system supports multiple presentation methods, selecting the most suitable form based on user preferences and context. For example, in noisy environments or when privacy is required, the system will prioritize text; when the user is relaxing or engaged in other activities, it may choose voice; for scenarios requiring guidance in meditation or relaxation exercises, the system may combine voice with background music. Multimedia presentation enhances the immersiveness and effectiveness of the intervention, adapting to the needs of different usage scenarios.

[0174] The privacy control unit 53 allows users to set the scope of emotional data sharing and intervention receiving preferences. In this invention, users can finely control their privacy settings, such as: which types of emotional data can be collected and analyzed by the system; whether to allow the system to proactively intervene in specific situations; and whether to accept specific types of intervention content (such as social recommendations). This control is crucial for enhancing user trust and system acceptance, and it also aligns with the principles of data protection and user autonomy.

[0175] The feedback collection unit 54 is connected to the emotion feedback module 6 and is used to collect user response data and subjective evaluations of the intervention content. In this invention, feedback collection emphasizes naturalness and non-intrusiveness, primarily collecting feedback through the user's natural interaction behavior (such as response content and interaction methods). The system may also, at appropriate times, lightly inquire about the user's feelings, such as "Was this interaction helpful to you?" or provide simple rating options. This feedback data is crucial for evaluating the intervention's effectiveness and optimizing the system.

[0176] The system of the present invention also includes a night protection module, a social recommendation module, and a task guidance module.

[0177] The late-night protection module connects to the intelligent intervention decision engine 3 to automatically activate the emotion protection mode during nighttime hours (10:30 PM - 2:00 AM), adjusting intervention strategies and content. In this invention, the system recognizes that nighttime is a critical period of emotional vulnerability, especially for users with sleep problems. The late-night protection mode adopts a gentler, more soothing intervention approach, avoiding overstimulation or triggering strong emotional fluctuations. For example, the system avoids recommending tasks requiring high concentration or social interaction, instead offering quiet companionship, soft music, or simple relaxation exercises. This feature specifically addresses the user's psychological safety and health during the late-night period.

[0178] The social recommendation module connects to the intelligent intervention decision engine 3 to recommend users seeking emotional support or guide them into emotional healing social spaces when appropriate. In this invention, the system understands the importance of social connection in improving emotional states, especially for users exhibiting high levels of loneliness. When a user's need for social support is identified, the system may recommend "emotional support officers" with similar emotional states but better recovery for interaction, or invite the user to join a themed emotional support group (such as a "warm late-night chat room" or a "relaxing stress-relief space"). This social recommendation takes into account the user's personality traits, social preferences, and current emotional state, ensuring that the recommended social interactions are comfortable and beneficial.

[0179] The task guidance module connects to the intelligent intervention decision engine 3 to generate emotion regulation tasks and guide users in completing them. In this invention, the system designs various regulation tasks targeting different emotional issues. For example, for loneliness, the system might recommend "writing a letter to your future self" or "emotional message in a bottle" (anonymously sharing feelings); for anxiety, it might recommend "three-minute breathing space" or "worry list and solutions"; for low mood, it might recommend "today's gratitude list" or "recording small joys," etc. These tasks are carefully designed, based on psychological theories, and can help users shift their attention, reconstruct cognition, and express emotions, thereby improving their emotional state. The system adjusts the difficulty and type of tasks based on the user's response and completion status, ensuring that the tasks are challenging yet within the user's capabilities.

[0180] Reference Figure 7 As shown, the overall workflow of the system of the present invention is as follows:

[0181] 1. Users interact with the system through text, voice, and other means, generating multimodal data;

[0182] 2. The emotion recognition engine 1 analyzes multimodal data and generates emotion vectors;

[0183] 3. Risk assessor 2 evaluates the risk level of the emotion vector;

[0184] 4. If the risk level exceeds the preset threshold (≥2 levels), the intervention process will be triggered;

[0185] 5. The intelligent intervention decision engine 3 combines user profiles to generate personalized intervention strategies;

[0186] 6. Intervention Content Generator 4 generates intervention content for both emotional and cognitive channels based on the intervention strategy;

[0187] 7. The user interface 5 presents the intervention content in an appropriate format;

[0188] 8. Users interact with the intervention content, generating feedback data;

[0189] 9. Emotional Feedback Module 6 analyzes feedback data and evaluates the intervention effect;

[0190] 10. Feedback results are used to adjust intervention strategies and optimize the system, forming a closed loop.

[0191] For example, a user repeatedly browses negative content late at night and sends messages expressing loneliness and helplessness. The system, through its emotion recognition engine, generates an emotion vector [30, 60, 85, 65], indicating low happiness, moderate anxiety, high loneliness, and high depression. The risk assessor calculates a risk score of 58, corresponding to Level 2 (critical) risk, triggering the intervention process. The intelligent intervention decision engine analyzes the user profile and finds that the user responds well to written expression and creative tasks. Considering the characteristics of late-night hours, a combination strategy of "gentle guidance + creative expression task" is selected. The intervention content generator generates emotional channel content (providing understanding and companionship) and cognitive channel content (guiding a shift in thinking), and recommends the "Write to Your Tomorrow's Self" task. The user interface presents this content in a gentle, anthropomorphic way. The user interacts with the system, completes the recommended task, and the emotion feedback module detects an improvement in the user's emotion vector to [45, 50, 70, 55], assessing a good intervention effect. The system records this successful case and adjusts the strategy weights accordingly to increase the probability of selecting the "gentle guidance + creative expression task" strategy in similar situations.

[0192] In summary, this invention provides an intelligent intervention and reassurance system based on real-time emotion recognition. Through multimodal emotion recognition, personalized risk assessment, intelligent intervention decision-making, and effect feedback optimization, it achieves timely identification and effective intervention of users' negative emotions. This system not only improves users' emotional health but also enhances the platform's user experience and user stickiness, possessing broad application prospects and commercial value.

[0193] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. An intelligent intervention and soothing system based on real-time emotion recognition, characterized in that, include: An emotion recognition engine is used to collect users' text, voice, and behavioral data, and generate multi-dimensional emotion vectors that represent users' emotional states. A risk assessor, connected to the emotion recognition engine, is used to receive the multi-dimensional emotion vector, assess the risk level of the multi-dimensional emotion vector, and trigger an intervention process when the risk level exceeds a preset threshold. The intelligent intervention decision engine, connected to the risk assessor, is used to receive the multidimensional emotion vector and risk level, and generate personalized intervention strategies based on user profile data. An intervention content generator, connected to the intelligent intervention decision engine, is used to receive the personalized intervention strategy and generate corresponding intervention content. The user interface is connected to the intervention content generator and is used to present the intervention content and collect user feedback data. An emotion feedback module, connected to the user interface and the intelligent intervention decision engine, is used to receive feedback data from the user, evaluate the intervention effect, and feed the intervention effect back to the intelligent intervention decision engine.

2. The system according to claim 1, characterized in that, The emotion recognition engine includes: The text analysis unit is used to analyze keywords, sentiment word weights, and subjective biases in user text. The speech analysis unit is used to analyze the user's speech rate, pitch, pause frequency, and intonation fluctuations. The behavior analysis unit is used to analyze users' login time, activity level, operation frequency, and click behavior; The multimodal fusion unit, connected to the text analysis unit, the speech analysis unit, and the behavior analysis unit, is used to fuse the analysis results and generate a multidimensional emotion vector containing feelings of pleasure, anxiety, loneliness, and depression.

3. The system according to claim 1, characterized in that, The risk assessor includes: Threshold setting unit, used to store risk thresholds for each emotion dimension; A personalized calibration unit is used to adjust the risk threshold based on the user's historical emotion data; A risk calculation unit, connected to the threshold setting unit and the personalized calibration unit, is used to compare the multidimensional emotion vector with the adjusted risk threshold and generate a risk assessment result. A risk level classification unit, connected to the risk calculation unit, is used to classify the risk assessment results into multiple risk levels.

4. The system according to claim 1, characterized in that, The intelligent intervention decision engine includes: A strategy library is used to store various intervention strategies and their corresponding feature vectors. A strategy matching unit, connected to the strategy library, is used to calculate the matching degree between the multidimensional emotion vector and the feature vectors of each intervention strategy; The user profile database is used to store users' intervention preferences and historical response data. An intervention effect prediction unit, connected to the user profile database, is used to predict the potential effects of each intervention strategy on the current user. The strategy optimization unit, connected to the strategy matching unit and the intervention effect prediction unit, is used to combine the matching degree and the prediction effect to determine the final personalized intervention strategy. A time-sensitive adjustment unit, connected to the strategy optimization unit, is used to adjust the personalized intervention strategy according to the current time period.

5. The system according to claim 4, characterized in that, The strategy matching unit is used for: Calculate the similarity between the multidimensional emotion vector and the feature vectors of each intervention strategy; A weighted coefficient is applied to the similarity to generate a preliminary matching score; The initial matching score is adjusted based on the risk level; Generate a sorted list of candidate strategies.

6. The system according to claim 4, characterized in that, The intervention effect prediction unit is used for: Obtain historical response data of users to different types of interventions from the user profile database; Identify users' intervention sensitivity characteristics; Based on the intervention sensitivity characteristics and current emotional state, predict the expected effects of each intervention strategy; The expected effect is adjusted based on the user's time-period response characteristics.

7. The system according to claim 1, characterized in that, The intervention content generator includes: The emotion channel generation unit is used to generate intervention content for emotion regulation; The cognitive channel generation unit is used to generate intervention content for cognitive restructuring; A content template library, connected to the emotion channel generation unit and the cognitive channel generation unit, is used to provide basic expression templates; A personalized adjustment unit, connected to the emotion channel generation unit and the cognitive channel generation unit, is used to adjust the intervention content according to user characteristics; A security check unit, connected to the emotion channel generation unit and the cognitive channel generation unit, is used to filter content that may exacerbate negative emotions.

8. The system according to claim 1, characterized in that, The emotion feedback module includes: The emotion change monitoring unit is used to monitor changes in emotion vectors before and after intervention; The user interaction analysis unit is used to analyze the interaction between users and the intervention content; An effectiveness evaluation unit, connected to the emotion change monitoring unit and the user interaction analysis unit, is used to comprehensively evaluate the intervention effect; A strategy adjustment unit, connected to the effect evaluation unit, is used to generate strategy optimization suggestions based on the intervention effect; The knowledge accumulation unit, connected to the effect evaluation unit, is used to store effective intervention cases.

9. The system according to claim 1, characterized in that, Also includes: The nighttime protection module is connected to the intelligent intervention decision engine and is used to automatically activate the emotion protection mode during nighttime hours, adjusting the intervention strategy and content. The social recommendation module is connected to the intelligent intervention decision engine and is used to recommend emotional companionship users or guide them into emotional healing social spaces when appropriate. The task guidance module, connected to the intelligent intervention decision engine, is used to generate emotion regulation tasks and guide users to complete them.

10. The system according to claim 1, characterized in that, The user interface includes: Anthropomorphic interactive units are used to present comforting characters with personality traits; Multimedia presentation unit, used to present intervention content in the form of text, voice, images, etc.; The privacy control unit allows users to set the scope of emotional data sharing and intervention receiving preferences; The feedback collection unit, connected to the emotion feedback module, is used to collect user response data and subjective evaluations of the intervention content.