Psychological crisis early warning intervention system and method based on behavior analysis

Abstract

The invention discloses a psychological crisis early warning intervention system and method based on behavior analysis, and relates to the technical field of psychological health management, and the method comprises the steps: collecting behavior data and emotion semantic data of a user through multiple sensors, and generating an original data set; performing time axis alignment on different data sources in the original data set to obtain a structured behavior feature and emotion feature data set; monitoring the behavior pattern and emotion expression of the user in real time through a multi-dimensional behavior mutation network, and identifying abnormal mutation points deviating from the personalized behavior baseline model; and combining the behavior mutation point, the historical behavior data of the user and the mutation amplitude of the emotion expression, and formulating a personalized intervention strategy. The behavior and emotion of the user are monitored in real time by using the multi-dimensional behavior mutation network, and the abnormal mutation points deviating from the personalized behavior baseline model are identified, so that effective intervention measures can be taken in time in the early stage of the psychological crisis, and the risk caused by the psychological crisis is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of mental health management, and in particular to a psychological crisis early warning intervention system and method based on behavior analysis. Background Art

[0002] As social pressure continues to increase, mental health issues are receiving increasing attention. Psychological crises, especially crisis events that lead to the risk of self-harm or suicide, have become social problems that people urgently need to solve. Traditional mental health assessment methods mostly rely on self-reporting or professional psychological counseling, but these methods have significant lag and subjectivity problems. In addition, with the development of intelligent technology, modern mental health monitoring has gradually turned to the use of smart devices and big data analysis to assist psychological assessment and intervention through behavioral monitoring, sentiment analysis and other means. Research in this field includes physiological data collection based on wearable devices, emotion recognition technology based on voice and expression, and behavioral pattern detection technology based on big data analysis. Although existing technologies have made many contributions to the identification and intervention of mental health problems, they still have many shortcomings, especially in real-time, personalization and multimodal data fusion.

[0003] Most of the current psychological crisis warning systems rely on a single data source, such as a single physiological indicator or emotion recognition, and lack comprehensive monitoring of user behavior and emotions. In addition, existing systems are usually based on group average models, ignoring individual differences, making it difficult to develop personalized warning and intervention plans for different individuals' behavior patterns and psychological states. This leads to insufficient sensitivity and accuracy of the system, especially for users whose psychological states fluctuate rapidly or have complex emotional expressions. Existing technologies are unable to capture subtle changes in their behavior or emotions in a timely manner, and may miss the best time for intervention. More importantly, the existing psychological crisis warning system lacks an effective fusion mechanism when processing multimodal data, resulting in the failure to fully utilize the temporal correlation between behavioral data and emotional data, affecting the overall crisis warning effect. Summary of the invention

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

[0005] Therefore, the present invention provides a psychological crisis early warning intervention method based on behavioral analysis to solve the problems of lack of effective fusion mechanism and lack of personalized early warning and intervention in the prior art.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions: In a first aspect, the present invention provides a psychological crisis early warning intervention method based on behavior analysis, which comprises: collecting user behavior data and emotional semantic data through multiple sensors to generate an original data set; Align the time axis of different data sources in the original dataset to obtain structured behavioral and emotional feature datasets; Based on the structured behavioral feature and emotional feature datasets, we build a personalized behavioral baseline model and emotional semantic model for users to capture their behavioral patterns and emotional expression characteristics. Through the multi-dimensional behavior mutation network, the user's behavior patterns and emotional expressions are monitored in real time to identify abnormal mutation points that deviate from the personalized behavior baseline model; Combine behavioral mutation points, user historical behavior data, and the mutation amplitude of emotional expression to develop personalized intervention strategies.

[0007] As a preferred solution of the psychological crisis early warning intervention method based on behavior analysis described in the present invention, wherein the user's behavior data and emotional semantic data are collected by multiple sensors to generate an original data set, the specific steps are as follows: Acquire physiological data, voice data, and facial data; Connect smart wearable devices to central data acquisition equipment and collect and transmit data through data acquisition programs; Set the sampling frequency for each sensor and unify the time interval; The sensor data stream is given a unified time stamp and context label and transmitted to the central server in the form of real-time stream; The real-time transmitted data stream is formatted and uniformly stored, and after quality monitoring and preprocessing, the original data set is generated.

[0008] As a preferred solution of the psychological crisis early warning intervention method based on behavioral analysis described in the present invention, the time axis alignment of different data sources in the original data set is performed to obtain a structured behavioral feature and emotional feature data set, and the specific steps are as follows: For the data stream with unified timestamp, set the time window and use the interpolation method to fill the low-frequency data; Extract physiological features, speech features, and facial features from the aligned data to form structured behavioral and emotional features; The extracted features are combined with context labels to obtain structured behavioral and emotional feature datasets.

[0009] As a preferred solution of the psychological crisis early warning intervention method based on behavior analysis described in the present invention, the personalized behavior baseline model and emotional semantic model of the user are constructed based on the structured behavior feature and emotional feature data set to capture the user's behavior pattern and emotional expression characteristics. The specific steps are as follows: Extract heart rate variability features from structured behavioral and emotional feature datasets and embed them into a multi-dimensional emotional feature space; Based on the behavioral and emotional feature datasets, a personalized behavior baseline model is constructed through nonlinear time series modeling. At the same time, a nonlinear integral equation is introduced to capture the complex dynamic changes of user behavior. The expression is: ; in, Indicates time The behavioral baseline Indicates that the user is at time behavioral characteristics, Represents a nonlinear function of behavioral characteristics changing over time, Indicates the adjustment of acceleration term A nonlinear function of Indicates behavioral characteristics About Time The second-order derivative of represents the time window length, Represents tiny time increments in calculus; Different emotional feature sources are processed uniformly through multimodal emotional fusion, and the emotional baseline is output through the nonlinear fusion equation. The expression is: ; in, Indicates that the user is at time The emotional baseline of the moment, Indicates the user's first emotional feature Over time The nonlinear change function of Indicates the user's second emotional feature Over time The first derivative of Indicates the user's second emotional feature Over time The nonlinear function of .

[0010] As a preferred solution of the psychological crisis early warning intervention method based on behavior analysis described in the present invention, wherein the user's behavior pattern and emotional expression are monitored in real time through a multi-dimensional behavior mutation network to identify abnormal mutation points that deviate from the personalized behavior baseline model, the specific steps are as follows: Extract real-time behavior features and emotion features from the user's behavior baseline and emotion baseline as input; The behavioral features are integrated into the , acceleration term and emotional characteristics , Conduct information fusion; Based on the fused information, a bidirectional long short-term memory network is used to model the time series of behavioral and emotional features. Furthermore, the isolation forest algorithm is introduced into the bidirectional long short-term memory network to detect abnormal mutation points in real time; Compare real-time data with the user's behavioral and emotional baselines to calculate the degree of deviation between the current behavior and the baseline; Real-time behavioral characteristics and baseline behavioral mutation amplitude , the expression is: ; in, Indicates at time Real-time behavior, Indicates at time personalized behavior; Real-time emotional features and baseline emotional mutation amplitude , the expression is: ; in, Indicates at time emotional characteristics, Indicates at time emotional baseline; Through and Conduct analysis to identify breakpoints that deviate significantly from the personalized baseline; Setting the abnormal mutation point detection threshold ,if or Any one of the two exceeds the abnormal mutation point detection threshold , it is determined to be an abnormal mutation point.

[0011] As a preferred solution of the psychological crisis early warning intervention method based on behavior analysis described in the present invention, the use of a bidirectional long short-term memory network to model the time series of behavioral characteristics and emotional characteristics is specifically performed in the following steps: Through forward LSTM network and backward LSTM network processing, the historical information and future information of behavioral characteristics and emotional characteristics are captured respectively; Connect the captured historical information and future information to generate a comprehensive implicit representation of the time series; Use mean square error as the loss function and use Adam optimizer to update model parameters; After the model training is completed, the bidirectional LSTM will predict the input time series and output a continuous time series implicit representation.

[0012] As a preferred solution of the psychological crisis early warning intervention method based on behavior analysis described in the present invention, the behavior mutation point, the user's historical behavior data and the mutation amplitude of emotional expression are combined to formulate a personalized intervention strategy, and the specific steps are as follows: Change the behavior and emotional mutation amplitude Through weighted linear combination, the comprehensive amplitude is formed; Combine historical behavior and sentiment data to analyze the differences and similarities between the current mutation point and historical trends and sentiment expression patterns, and determine whether the mutation is a periodic fluctuation or an anomaly; Compare the comprehensive amplitude with historical data. When the deviation of the comprehensive amplitude from the personalized baseline is less than K, a slight intervention is taken, where K is the threshold for distinguishing normal fluctuations from mild abnormalities. When the deviation of the comprehensive amplitude from the personalized baseline is between K and P, moderate intervention is adopted, where P is the threshold for judging high-risk behavior or emotional abnormality; When the combined magnitude deviates from the personalized baseline by more than P, emergency intervention should be initiated immediately.

[0013] In a second aspect, the present invention provides a psychological crisis early warning intervention system based on behavioral analysis, including: a data acquisition module, a timeline alignment module, a model building module, a monitoring module and a strategy formulation module; The data collection module is used to collect user behavior data and emotional semantic data through multiple sensors to generate an original data set; The timeline alignment module is used to perform timeline alignment on different data sources in the original data set to obtain a structured behavioral feature and emotional feature data set; The model building module is used to build a user's personalized behavior baseline model and emotional semantic model based on the structured behavior feature and emotional feature data set to capture the user's behavior pattern and emotional expression characteristics; The monitoring module is used to monitor the user's behavior pattern and emotional expression in real time through a multi-dimensional behavior mutation network, and identify abnormal mutation points that deviate from the personalized behavior baseline model; The strategy formulation module is used to combine the behavior mutation point, the user's historical behavior data and the mutation amplitude of emotional expression to formulate a personalized intervention strategy.

[0014] In a third aspect, the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein when the computer program is executed by the processor, any step of the psychological crisis warning intervention method based on behavioral analysis as described in the first aspect of the present invention is implemented.

[0015] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements any step of the psychological crisis early warning intervention method based on behavioral analysis as described in the first aspect of the present invention.

[0016] The beneficial effects of the present invention are as follows: the present invention collects the user's behavior data and emotional semantic data through multiple sensors, generates an original data set, and aligns the time axes of different data sources to obtain structured behavior feature and emotional feature data sets, constructs the user's personalized behavior baseline model and emotional semantic model, and can effectively capture the user's behavior pattern and emotional expression characteristics; uses a multidimensional behavior mutation network to monitor the user's behavior and emotion in real time, identifies abnormal mutation points that deviate from the personalized behavior baseline model, and thus can take effective intervention measures in time at the early stage of a psychological crisis to reduce the risks brought about by the psychological crisis. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0018] Figure 1 This is a flow chart of the psychological crisis early warning intervention method based on behavioral analysis in Example 1.

[0019] Figure 2 This is a module diagram of the psychological crisis early warning intervention system based on behavior analysis in Example 1. DETAILED DESCRIPTION

[0020] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0021] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0022] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0023] Example 1, reference Figure 1 and Figure 2 , which is the first embodiment of the present invention, and provides a psychological crisis early warning intervention method based on behavior analysis, comprising the following steps: S1, collects user behavior data and emotional semantic data through multiple sensors to generate original data sets.

[0024] Furthermore, physiological data (such as heart rate, exercise data, etc.), voice data and facial data can be obtained through smart wearable devices (such as smart watches and smart bracelets); It should be noted that when acquiring physiological data, it should be ensured that the smart wearable device supports multiple sensors (heart rate, acceleration, microphone, camera, etc.) and that these devices are compatible with data acquisition devices (such as mobile phones or other central devices).

[0025] It should be noted that "behavioral data" refers to data related to the user's physiological reactions, voice activities, and facial expressions collected in real time through a variety of sensors (such as smart wearable devices, voice recognition devices, and facial expression analysis technology). Specifically, physiological data includes physiological indicators such as the user's heart rate, body temperature, and respiratory rate, voice data includes the user's voice content, tone, and speaking speed, and facial data includes changes in the user's facial expressions, eye activities, etc. The data collected by these sensors can provide a basis for further analysis of the user's behavior.

[0026] Furthermore, "emotional semantic data" is to extract information related to the user's emotional state through further analysis of the above behavioral data. These data are usually processed by machine learning and sentiment analysis algorithms, with the aim of identifying the user's current emotional state, such as joy, tension, sadness, and other emotions. For example, the tone and speed of speech can reflect the user's emotional fluctuations, facial expression analysis can help determine the user's emotional response, and changes in physiological data (such as heart rate, skin temperature, etc.) often correspond to emotional fluctuations. By analyzing these emotional semantic data, we can have a deeper understanding of the user's emotional state and behavioral patterns.

[0027] Connect the smart wearable device to the central data collection device (such as a smartphone or IoT gateway) via Bluetooth, Wi-Fi, etc., and collect and transmit data through the data collection program; Among them, the data acquisition program should support data access from multiple sensors and provide a stable data transmission protocol (such as MQTT or HTTP) to ensure that data is not lost or delayed; at the same time, device initialization is also required to ensure the normal operation of each sensor.

[0028] Set the sampling frequency for each sensor and unify the time interval; For example, the heart rate sensor is set to update 5 times per second, and the GPS positioning is updated every 5 seconds. Ensure that the time intervals of all sensors are consistent so that subsequent data can be synchronized and aligned. The sensor data stream is marked with a unified timestamp and context label and transmitted to the central server in the form of a real-time stream; Preferably, a unified sampling frequency can not only avoid data redundancy, but also ensure the comparability of data between different sensors (for example, heart rate, motion data, voice data, etc. must be collected at the same time point).

[0029] The sensor data stream is marked with a unified timestamp and context label (such as walking, stillness, and talking) and transmitted to the central server in the form of real-time stream; Preferably, the point of unified timestamp is to ensure that the data from each sensor can be compared and analyzed under the same time reference, even if they have different sampling frequencies, so that the timestamp can be used to synchronize the subsequent processing, alignment and analysis of multiple data sources across sensors.

[0030] It should also be noted that in addition to adding a unified timestamp, the NTP protocol or the device's internal clock can be used for time synchronization to ensure that multi-sensor data is aligned at the same time point. At the same time, context labels can be automatically generated based on changes in sensor data, and speech sentiment analysis models can also add sentiment labels to speech data.

[0031] The real-time data stream is formatted and uniformly stored (such as JSON or Parquet format), and is subjected to quality monitoring (such as packet loss and outlier detection) and preprocessing (such as denoising and interpolation to fill missing values) to generate the original data set.

[0032] It should also be noted that when storing data, ensure field consistency (such as timestamp, sensor type, data value, context label). Data preprocessing can use filtering algorithms (such as Kalman filtering) to remove sensor noise and fill in missing data appropriately. The final data set should be complete and consistent, and can provide reliable input for the next step of data analysis or modeling.

[0033] S2, align the time axes of different data sources in the original dataset to obtain structured behavioral and emotional feature datasets.

[0034] Furthermore, the time window is set for the data stream with unified timestamps, and the interpolation method is used to fill the low-frequency data; Among them, linear interpolation is performed on low-frequency physiological data to fill in the values ​​in each time window; the voice data is directly extracted according to the time window, and the facial data is also aligned through interpolation.

[0035] Specifically, time window setting refers to determining the minimum time unit based on the sensor with the highest sampling frequency (such as voice data) and dividing the time window into a unified time window (such as 10 milliseconds).

[0036] Extract physiological features, speech features, and facial features from the aligned data to form structured behavioral and emotional features; Specifically, physiological features: extract statistical features of physiological data such as heart rate; speech features: extract speech emotion features such as MFCC and pitch; facial features: extract key point change features of facial expressions.

[0037] The extracted features are combined with context labels to obtain structured behavioral and emotional feature datasets.

[0038] Specifically, the user behavior context (such as exercise, rest) is matched with the features of each time window, and the physiological, voice, facial features and situational labels are integrated to form a multidimensional dataset, namely, a structured behavioral feature and emotional feature dataset.

[0039] S3, based on structured behavioral feature and emotional feature datasets, builds users’ personalized behavioral baseline models and emotional semantic models to capture users’ behavioral patterns and emotional expression characteristics.

[0040] Furthermore, the heart rate variability features are extracted from the structured behavioral and emotional feature datasets and embedded into the multi-dimensional emotional feature space; Among them, the heart rate variability characteristic (HRV) refers to the degree of change in the time interval between consecutive heartbeats (i.e., RR interval). HRV is a common physiological characteristic that can reflect the activity of the autonomic nervous system, especially when under stress, anxiety or emotional fluctuations, heart rate variability is often affected.

[0041] Based on the behavioral and emotional feature datasets, a personalized behavior baseline model is constructed through nonlinear time series modeling. At the same time, a nonlinear integral equation is introduced to capture the complex dynamic changes of user behavior. The expression is: ; in, Indicates time The behavioral baseline Indicates that the user is at time behavioral characteristics, Represents a nonlinear function of behavioral characteristics changing over time, Indicates the adjustment of acceleration term A nonlinear function of Indicates behavioral characteristics About Time The second derivative of , that is, the acceleration change term, represents the time window length, Represents tiny time increments in calculus; Preferably, this nonlinear modeling approach can capture the gradual changes in user behavior and identify the mutation points of behavior. Compared with traditional linear models, this model can better handle complex nonlinear changes in user behavior, and is particularly suitable for identifying small but meaningful behavior changes under long-term data monitoring.

[0042] Different emotional feature sources are processed uniformly through multimodal emotional fusion, and the emotional baseline is output through the nonlinear fusion equation. The expression is: ; in, Indicates that the user is at time The emotional baseline of the moment, Indicates the user's first emotional feature Over time The nonlinear change function of Indicates the user's second emotional feature Over time The first derivative of Indicates the user's second emotional feature Over time Nonlinear function of Preferably, the sentiment semantic model can fuse sentiment data from different sources and capture the gradual and sudden changes of sentiment through nonlinear modeling. Compared with single-modality sentiment analysis, the use of multimodal fusion can more comprehensively and accurately capture the user's emotional fluctuations.

[0043] S4, through a multi-dimensional behavior mutation network, monitors users’ behavior patterns and emotional expressions in real time and identifies abnormal mutation points that deviate from the personalized behavior baseline model.

[0044] Furthermore, real-time behavioral features and emotional features are extracted from the user's behavioral baseline and emotional baseline as input; Through the attention mechanism (AttentionMechanism) , acceleration term and emotional characteristics , Conduct information fusion; Based on the fused information, a bidirectional long short-term memory network (BiLSTM) is used to model the time series of behavioral and emotional features to capture the dynamic changes of user behavior and emotions over time. The specific steps are as follows: Through forward LSTM network and backward LSTM network processing, the historical information and future information of behavioral characteristics and emotional characteristics are captured respectively; Among them, the forward LSTM processes from time Time data to capture historical information; the backward LSTM network processes the information from time Time data to capture future information; Connect the captured historical information and future information to generate a comprehensive implicit representation of the time series; Among them, the implicit representation contains the dynamic change information of user behavior and emotion, which can reflect the user's behavior pattern and emotional state at different time points.

[0045] It should be noted that the implicit representation of the output is usually a high-dimensional vector, the dimension of which depends on the size of the hidden layer of the LSTM and the number of input features. This high-dimensional vector will serve as the input for the subsequent step (anomaly detection).

[0046] The mean square error (MSE) is used as the loss function, and the Adam optimizer is used to update the model parameters to ensure that the model can effectively learn the dynamic patterns of user behavior and emotions; After the model training is completed, the bidirectional LSTM will predict the input time series and output the continuous time series implicit representation; Among them, each time point has a corresponding high-dimensional implicit feature vector, which represents the behavior and emotional state at that moment.

[0047] Preferably, the bidirectional LSTM can consider both past and future sequence information, which is suitable for user behavior and sentiment data.

[0048] It should be noted that the main task of BiLSTM is to capture the dynamic change pattern in the time series. Here, the input is the user's behavior characteristics and emotional characteristics , And derived higher-order features, such as acceleration terms ; BiLSTM, through its bidirectional structure, is able to simultaneously consider the past and future contextual information of behavioral and emotional data in time, thereby modeling the entire time series.

[0049] The specific modeling process is as follows: BiLSTM can effectively capture important feature changes in the time series through its memory unit and gating mechanism, and generate an implicit representation of the time series, that is, mapping from the input features to a high-dimensional implicit space. This implicit representation can be regarded as a summary and abstraction of the original behavior and emotion data, and is used to represent the user's behavior and emotional state over a period of time.

[0050] The final output of BiLSTM is a high-dimensional latent vector of a time series, which contains rich information about the changes of user behavior characteristics and sentiment characteristics over time.

[0051] The Isolation Forest algorithm is further introduced into the bidirectional long short-term memory network to detect abnormal mutation points in real time; Specifically, Isolation Forest detects rare data points by randomly partitioning the feature space.

[0052] It should also be noted that the input of the Isolation Forest is not the original behavioral and emotional features, but the implicit representation of the time series output by the BiLSTM. In other words, the Isolation Forest relies on the output of the BiLSTM. The BiLSTM has modeled the time series and captured the time-dependent characteristics of user behavior and emotion. The Isolation Forest uses these high-dimensional implicit representations as input to determine which data points are "abnormal" - that is, points that are significantly different from normal behavioral and emotional patterns.

[0053] Preferably, BiLSTM can make full use of the characteristics of time series data to capture complex behavior and emotion patterns; Isolation Forest is good at processing high-dimensional data and can effectively detect anomalies in complex behavior and emotion patterns. The two work together to enhance anomaly detection capabilities and more accurately identify abnormal mutations in behavior or emotion.

[0054] Compare real-time data with the user's behavioral and emotional baselines to calculate the degree of deviation between the current behavior and the baseline; Real-time behavioral characteristics and baseline behavioral mutation amplitude , the expression is: ; in, Indicates at time Real-time behavior, Indicates at time personalized behavior; Real-time emotional features and baseline emotional mutation amplitude , the expression is: ; in, Indicates at time emotional characteristics, Indicates at time emotional baseline; Through and Conduct analysis to identify breakpoints that deviate significantly from the personalized baseline; Setting the abnormal mutation point detection threshold ,if or Any one of the two exceeds the abnormal mutation point detection threshold , it is determined to be an abnormal mutation point; It should be noted that in order to cope with long-term changes in user behavior and emotions, the behavioral baseline and emotional baseline will be periodically updated through the incremental learning method.

[0055] Based on the sliding window approach, the baseline model is re-estimated using the latest data, ensuring that the model can adapt to long-term behavioral and sentiment changes.

[0056] S5, combines behavioral mutation points, user historical behavior data, and the mutation amplitude of emotional expression to develop personalized intervention strategies.

[0057] Furthermore, the magnitude of behavioral mutation and emotional mutation amplitude Through weighted linear combination, the comprehensive amplitude is formed; Combine historical behavior and sentiment data to analyze the differences and similarities between the current mutation point and historical trends and sentiment expression patterns, and determine whether the mutation is a periodic fluctuation or an anomaly; It should be noted that if the fluctuations are cyclical (i.e., the user's common emotional ups and downs), the deviation may not trigger overly aggressive intervention measures, and usually only minor intervention is required.

[0058] If judged to be an abnormal change, the magnitude of the deviation from baseline becomes more important, determining whether moderate or urgent intervention is required.

[0059] Compare the comprehensive amplitude with historical data. When the deviation of the comprehensive amplitude from the personalized baseline is less than K, take a slight intervention, such as sending an encouraging message or prompting the user to participate in relaxation activities. Among them, K is the threshold for distinguishing normal fluctuations from mild abnormalities.

[0060] When the deviation of the comprehensive amplitude from the personalized baseline is between K and P, it indicates that the user is under high stress, and moderate intervention is taken, such as pushing personalized emotion management and regulation suggestions to help users regulate their emotions; Among them, P is the threshold for judging high-risk behavior or abnormal emotions.

[0061] When the deviation of the comprehensive amplitude from the personalized baseline exceeds P, the user should be prompted to indicate that there is a high psychological crisis, and emergency intervention should be initiated immediately, such as notifying emergency contacts, providing psychological crisis hotline information, or triggering the alarm mechanism of smart devices.

[0062] Among them, K is the threshold used to distinguish normal fluctuations from mild abnormalities. Usually, behavioral characteristics and emotional characteristics will have a certain range of natural fluctuations around the personalized baseline, and these fluctuations can be classified as "normal fluctuations" or "mild emotional fluctuations."

[0063] Recommended range: Set to 5% to 10% deviation from the user's personalized baseline. For example, if the user's normal behavior and emotional baseline have slight fluctuations (such as heart rate changes, short-term mood swings), but not enough to indicate that the user is under stress or emotionally out of control, then it is defined as low risk.

[0064] Applicable scenarios: When the fluctuation range of the user's behavior or emotional characteristics is within this range, mild intervention can be taken, such as prompting the user to perform daily emotional regulation.

[0065] Among them, P is the threshold used to judge high-risk behavior or emotional abnormality. Exceeding this threshold indicates that the user may be in a state of psychological crisis and emergency intervention measures must be taken.

[0066] Recommended range: Set to a deviation of 30% or greater from the personalized baseline. For example, if a user’s behavioral patterns or emotional expressions differ from their baseline by more than 30%, this may indicate that the user is in a state of extreme stress or emotional breakdown and requires immediate intervention.

[0067] Applicable scenarios: When the user's behavioral and emotional characteristics deviate from the personalized baseline by more than P, emergency response measures should be triggered immediately, such as notifying emergency contacts or activating an alarm mechanism.

[0068] Among them, the range between K and P is the medium risk range, indicating that the user may be experiencing greater stress or emotional fluctuations, but has not yet reached a crisis state.

[0069] Recommended range: 10% to 30% deviation from personalized baseline. At this point, the user may need external help to regulate emotions or cope with stress, but it is not yet enough to trigger crisis intervention.

[0070] Applicable scenarios: When the user's behavior or emotional characteristics fluctuate within this range, emotional management advice should be provided, or the user should be advised to contact a mental health professional.

[0071] It should be noted that the calculation of mutation amplitude and , and combined with the judgment results in the previous step (periodic fluctuations or abnormal changes) to further assess the risk level. Sudden changes in periodic fluctuations, even if the deviation is large, may be considered a normal behavior pattern, and therefore may not necessarily require high-intensity intervention. Abnormal changes mean that the deviation is not only significant in magnitude, but also rare in user behavior and emotional patterns, which may indicate a higher psychological risk and require stronger intervention.

[0072] The present embodiment also provides a psychological crisis early warning intervention system based on behavior analysis, including: a data acquisition module, a timeline alignment module, a model building module, a monitoring module and a strategy formulation module; the data acquisition module is used to collect the user's behavior data and emotional semantic data through multiple sensors to generate an original data set; the timeline alignment module is used to perform timeline alignment on different data sources in the original data set to obtain a structured behavior feature and emotional feature data set; the model building module is used to build a personalized behavior baseline model and emotional semantic model of the user based on the structured behavior feature and emotional feature data set to capture the user's behavior pattern and emotional expression characteristics; the monitoring module is used to monitor the user's behavior pattern and emotional expression in real time through a multi-dimensional behavior mutation network to identify abnormal mutation points that deviate from the personalized behavior baseline model; the strategy formulation module is used to combine the behavior mutation points, the user's historical behavior data and the mutation amplitude of emotional expression to formulate a personalized intervention strategy.

[0073] This embodiment also provides a computer device, which is suitable for the situation of the psychological crisis early warning intervention method based on behavioral analysis, including: a memory and a processor; the memory is used to store computer executable instructions, and the processor is used to execute computer executable instructions to implement the psychological crisis early warning intervention method based on behavioral analysis proposed in the above embodiment.

[0074] The computer device may be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covered on the display screen, or a key, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse, etc.

[0075] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, it implements the psychological crisis early warning intervention method based on behavioral analysis as proposed in the above embodiment; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (Static Random Access Memory, referred to as SRAM), electrically erasable programmable read-only memory (Electrically Erasable Programmable Read-Only Memory, referred to as EEPROM), erasable programmable read-only memory (Erasable Programmable Read-Only Memory, referred to as EPROM), programmable read-only memory (Programmable Red-Only Memory, referred to as PROM), read-only memory (Read-Only Memory, referred to as ROM), magnetic storage, flash memory, magnetic disk or optical disk.

[0076] In summary, the present invention collects the user's behavioral data and emotional semantic data through multiple sensors, generates an original data set, and aligns the time axes of different data sources to obtain structured behavioral feature and emotional feature data sets, and constructs the user's personalized behavioral baseline model and emotional semantic model, which can effectively capture the user's behavioral pattern and emotional expression characteristics; uses a multidimensional behavioral mutation network to monitor the user's behavior and emotion in real time, and identifies abnormal mutation points that deviate from the personalized behavioral baseline model, so that effective intervention measures can be taken in time at the early stage of the psychological crisis to reduce the risks brought about by the psychological crisis.

[0077] Example 2, referring to Table 1, is the second example of the present invention. To further verify the technical solution of the present invention, experimental simulation data of a psychological crisis early warning intervention method based on behavior analysis is provided.

[0078] A total of 40 volunteers aged between 18 and 40 were selected as subjects for the experiment, which lasted for 30 days. The volunteers wore smart wearable devices, including heart rate monitors, respiratory rate monitors, smart bracelets (for recording exercise and sleep data), voice recording devices (for capturing language emotional characteristics) and cameras (for capturing facial expression data). These devices were used to collect physiological, emotional and behavioral data of each volunteer in real time.

[0079] Furthermore, existing technologies for psychological crisis warning and intervention usually rely on a single data source or simple sentiment analysis methods, such as judging the psychological state through speech analysis, facial expression recognition or behavior monitoring. Common methods include using traditional speech sentiment analysis technology based on simplified sentiment feature models, or relying on a single physiological data (such as heart rate or skin conductivity) to monitor the user's psychological state.

[0080] Before the experiment began, each volunteer underwent a preliminary mental health assessment to determine their initial state. At the same time, a variety of different situational labels were set, such as work pressure, social interaction, and rest status, in order to better capture their emotional and behavioral characteristics during the experiment.

[0081] During the experiment, all data were marked with a unified timestamp to ensure that different data sources could be accurately aligned on the timeline. During the data collection phase, data such as heart rate, respiratory rate, voice intonation, and facial expressions were standardized, and low-frequency sampled data were interpolated to ensure data integrity. After preprocessing, the real-time data stream was generated into an original data set. Subsequently, the data was sorted by time window, behavioral and emotional features were extracted, and a structured data set was formed.

[0082] Based on these structured data sets, a personalized behavioral baseline model and emotional semantic model were constructed for each volunteer. The behavioral and emotional characteristics of each volunteer were monitored in real time through a multidimensional behavioral mutation network, and the abnormal mutation points of behavior and emotion were detected using the isolation forest algorithm. In order to capture the potential psychological crisis signals in these mutation points, a bidirectional long short-term memory network was used to model the time series of behavioral and emotional characteristics.

[0083] During the mutation point detection stage, the behavioral baseline and emotional baseline were calculated respectively, and different intervention strategies were set: when the mutation amplitude was small, mild intervention was adopted; when the mutation amplitude was medium, medium intervention was adopted; when the mutation amplitude was large, emergency intervention was initiated immediately.

[0084] The details are shown in Table 1 below: Table 1 Experimental data comparison table Parameter name User A (Existing Technology) User A (the present invention) User B (Existing Technology) User B (the present invention) Heart rate variability (ms) 45 60 50 65 Sentiment score (1-10) 4 7 5 8 Number of mutation points (per day) 3 1 4 1 Number of interventions (times / day) 2 1 3 1 Psychological stress index (1-100) 70 50 75 45 Through the analysis of experimental data, it can be clearly seen that the psychological crisis early warning intervention method of the present invention has shown significant advantages in many aspects. First, heart rate variability (HRV) is an important physiological indicator for measuring psychological stress and autonomic nervous system function. In the experiment, the heart rate variability of users using the method of the present invention was significantly improved, and the HRV of user A and user B increased from 45ms and 50ms to 60ms and 65ms respectively. This shows that the method of the present invention can effectively relieve the psychological stress of users and improve the function of the autonomic nervous system.

[0085] Secondly, the improvement of the emotion score further demonstrates the significant effect of the method of the present invention in improving the emotional state of the user. The emotion score is obtained by performing emotion analysis on the user's voice data and facial expression data. Under the intervention of the present invention, the emotion score of user A increased from 4 points to 7 points, and the emotion score of user B increased from 5 points to 8 points. This shows that through the personalized behavior baseline model and emotion semantic model of the present invention, the emotional fluctuations of the user can be captured more accurately, and timely intervention can be carried out when the mutation point occurs, which significantly improves the emotional state of the user.

[0086] In addition, the comparison between the number of mutation points and the number of interventions also illustrates the advantages of the present invention in terms of timely warning and intervention. Users A and B using the prior art detected 3 and 4 mutation points respectively on average per day, while users using the present invention only detected 1 mutation point per day. This shows that the present invention can capture potential crisis signals earlier and reduce the frequency of serious crises. At the same time, the number of interventions is also significantly reduced, indicating that the present invention has significant advantages in reducing unnecessary interventions.

[0087] Finally, the comparison of psychological stress index further shows the significant effect of the present invention in reducing the psychological stress of users. The psychological stress index dropped from 70 to 50, and from 75 to 45, showing that the present invention can help users better manage and relieve psychological stress through multidimensional behavior mutation network and personalized intervention strategy.

[0088] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A psychological crisis early warning intervention method based on behavioral analysis, characterized in that: include: Collect user behavior data and emotional semantic data through multiple sensors to generate original data sets; Align the time axis of different data sources in the original dataset to obtain structured behavioral and emotional feature datasets; Based on the structured behavioral feature and emotional feature datasets, we build a personalized behavioral baseline model and emotional semantic model for users to capture their behavioral patterns and emotional expression characteristics. Through the multi-dimensional behavior mutation network, the user's behavior patterns and emotional expressions are monitored in real time to identify abnormal mutation points that deviate from the personalized behavior baseline model; Combine behavioral mutation points, user historical behavior data, and the mutation amplitude of emotional expression to develop personalized intervention strategies.

2. A psychological crisis early warning intervention method based on behavioral analysis as claimed in claim 1, characterized in that: The method collects the user's behavior data and emotional semantic data through multiple sensors to generate an original data set. The specific steps are as follows: Acquire physiological data, voice data, and facial data; Connect smart wearable devices to central data acquisition equipment and collect and transmit data through data acquisition programs; Set the sampling frequency for each sensor and unify the time interval; The sensor data stream is given a unified time stamp and context label and transmitted to the central server in the form of real-time stream; The real-time transmitted data stream is formatted and uniformly stored, and after quality monitoring and preprocessing, the original data set is generated.

3. A psychological crisis early warning intervention method based on behavioral analysis as claimed in claim 2, characterized in that: The time axis alignment of different data sources in the original data set is performed to obtain a structured behavioral feature and emotional feature data set. The specific steps are as follows: For the data stream with unified timestamp, set the time window and use the interpolation method to fill the low-frequency data; Extract physiological features, speech features, and facial features from the aligned data to form structured behavioral and emotional features; The extracted features are combined with context labels to obtain structured behavioral and emotional feature datasets.

4. A psychological crisis early warning intervention method based on behavioral analysis as claimed in claim 3, characterized in that: Based on the structured behavioral feature and emotional feature data set, the personalized behavioral baseline model and emotional semantic model of the user are constructed to capture the user's behavioral pattern and emotional expression characteristics. The specific steps are as follows: Extract heart rate variability features from structured behavioral and emotional feature datasets and embed them into a multi-dimensional emotional feature space; Based on the behavioral and emotional feature datasets, a personalized behavior baseline model is constructed through nonlinear time series modeling. At the same time, a nonlinear integral equation is introduced to capture the complex dynamic changes of user behavior. The expression is: ； in, Indicates time The behavioral baseline Indicates that the user is at time behavioral characteristics, Represents a nonlinear function of behavioral characteristics changing over time, Indicates the adjustment of acceleration term A nonlinear function of Indicates behavioral characteristics About Time The second-order derivative of represents the time window length, Represents tiny time increments in calculus; Different emotional feature sources are processed uniformly through multimodal emotional fusion, and the emotional baseline is output through the nonlinear fusion equation. The expression is: ； in, Indicates that the user is at time The emotional baseline of the moment, Indicates the user's first emotional feature Over time The nonlinear change function of Indicates the user's second emotional feature Over time The first derivative of Indicates the user's second emotional feature Over time The nonlinear function of .

5. A psychological crisis early warning intervention method based on behavioral analysis as claimed in claim 4, characterized in that: The multi-dimensional behavior mutation network is used to monitor the user's behavior patterns and emotional expressions in real time, and to identify abnormal mutation points that deviate from the personalized behavior baseline model. The specific steps are as follows: Extract real-time behavior features and emotion features from the user's behavior baseline and emotion baseline as input; The behavioral features are integrated into the , acceleration term and emotional characteristics , Conduct information fusion; Based on the fused information, a bidirectional long short-term memory network is used to model the time series of behavioral and emotional features. Furthermore, the isolation forest algorithm is introduced into the bidirectional long short-term memory network to detect abnormal mutation points in real time; Compare real-time data with the user's behavioral and emotional baselines to calculate the degree of deviation between the current behavior and the baseline; Real-time behavioral characteristics and baseline behavioral mutation amplitude , the expression is: ； in, Indicates at time Real-time behavior, Indicates at time personalized behavior; Real-time emotional features and baseline emotional mutation amplitude , the expression is: ； in, Indicates at time emotional characteristics, Indicates at time emotional baseline; Through and Conduct analysis to identify breakpoints that deviate significantly from the personalized baseline; Setting the abnormal mutation point detection threshold ,if or Any one of the two exceeds the abnormal mutation point detection threshold , it is determined to be an abnormal mutation point.

6. A psychological crisis early warning intervention method based on behavioral analysis as claimed in claim 5, characterized in that: The bidirectional long short-term memory network is used to model the time series of behavioral characteristics and emotional characteristics. The specific steps are as follows: Through forward LSTM network and backward LSTM network processing, the historical information and future information of behavioral characteristics and emotional characteristics are captured respectively; Connect the captured historical information and future information to generate a comprehensive implicit representation of the time series; Use mean square error as the loss function and use Adam optimizer to update model parameters; After the model training is completed, the bidirectional LSTM will predict the input time series and output a continuous time series implicit representation.

7. A psychological crisis early warning intervention method based on behavioral analysis as claimed in claim 6, characterized in that: The behavior mutation point, the user's historical behavior data, and the mutation amplitude of emotional expression are combined to formulate a personalized intervention strategy. The specific steps are as follows: Change the behavior and emotional mutation amplitude Through weighted linear combination, the comprehensive amplitude is formed; Combine historical behavior and sentiment data to analyze the differences and similarities between the current mutation point and historical trends and sentiment expression patterns, and determine whether the mutation is a periodic fluctuation or an anomaly; Compare the comprehensive amplitude with historical data. When the deviation of the comprehensive amplitude from the personalized baseline is less than K, a slight intervention is taken, where K is the threshold for distinguishing normal fluctuations from mild abnormalities. When the deviation of the comprehensive amplitude from the personalized baseline is between K and P, moderate intervention is adopted, where P is the threshold for judging high-risk behavior or emotional abnormality; When the combined magnitude deviates from the personalized baseline by more than P, emergency intervention should be initiated immediately.

8. A psychological crisis early warning intervention system based on behavior analysis, based on a psychological crisis early warning intervention method based on behavior analysis according to any one of claims 1 to 7, characterized in that: include: Data collection module, timeline alignment module, model building module, monitoring module and strategy formulation module; The data collection module is used to collect user behavior data and emotional semantic data through multiple sensors to generate an original data set; The timeline alignment module is used to perform timeline alignment on different data sources in the original data set to obtain a structured behavioral feature and emotional feature data set; The model building module is used to build a user's personalized behavior baseline model and emotional semantic model based on the structured behavior feature and emotional feature data set to capture the user's behavior pattern and emotional expression characteristics; The monitoring module is used to monitor the user's behavior pattern and emotional expression in real time through a multi-dimensional behavior mutation network, and identify abnormal mutation points that deviate from the personalized behavior baseline model; The strategy formulation module is used to combine the behavior mutation point, the user's historical behavior data and the mutation amplitude of emotional expression to formulate a personalized intervention strategy.

Images