An emotion intervention system based on behavioral characteristics

Abstract

This invention discloses a behavioral feature-based emotion intervention system, belonging to the field of emotion intervention technology. It includes a feature fusion unit, a feature association analysis unit, a correlation group determination unit, a real-time data analysis unit, and an intervention plan output unit. Through continuous behavioral monitoring, this invention ensures that subtle deviations in behavioral patterns can be identified before the user is even consciously aware of them. This early warning capability for emotional risks breaks the traditional passive "problem occurs - proactively seek help" model, allowing for significantly earlier intervention and achieving true preventative mental health care. Furthermore, it can identify behavioral gaps and generate corresponding intervention actions, thus enabling personalized micro-interventions. From data collection to intervention execution, historical behavioral data is continuously used to optimize intervention strategies, thereby increasing the system's adaptability.

Description

Technical Field

[0001] This invention relates to the field of emotion intervention technology, specifically to an emotion intervention system based on behavioral characteristics. Background Technology

[0002] A behavior-based emotion intervention system continuously monitors user behavior data, establishes an individual baseline, intelligently identifies abnormal behavior patterns related to depression and anxiety, and automatically triggers personalized, tiered interventions. Patent application number 202510735650.9 discloses "an emotion intervention system, method, device, and medium based on multimodal analysis. The system includes an emotion analysis module, a behavior prediction module, and an intervention module. The emotion analysis module is configured to determine emotion characteristics based on user voice data; the behavior prediction module is configured to determine predicted behavior based on historical behavior data; the intervention module is configured to: determine whether to intervene and the intervention strategy based on emotion characteristics and predicted behavior; in response, determine a first audio parameter corresponding to the audio intervention and drive an audio device to play sound based on the first audio parameter; and / or determine a first lighting parameter corresponding to the lighting intervention and drive a lighting device to provide illumination based on the first lighting parameter."

[0003] The aforementioned existing technologies have solved problems such as the inability to intervene in users' emotions through multimodal means. However, when the system is running, it cannot provide early warnings and interventions before the user's emotional problems worsen. It can only wait for the user to actively seek help, thus missing the best intervention period. Furthermore, it cannot identify specific behavioral gaps and can only provide general suggestions, failing to address the user's real pain points and affecting the intervention effect. Summary of the Invention

[0004] The purpose of this invention is to provide an emotion intervention system based on behavioral characteristics to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: an emotion intervention system based on behavioral characteristics, including a real-time data analysis unit;

[0006] The feature fusion unit collects multi-source historical behavioral data and emotion scores of users, and analyzes the representation vector of the behavioral data.

[0007] The feature association analysis unit divides a day into multiple independent time periods, statistically analyzes the reference behavior data of all independent time periods, extracts the representation vector of the reference behavior data, and calculates the correlation coefficient between each feature in the representation vector.

[0008] The correlation group determination unit statistically analyzes the positive and negative correlation groups of each feature according to the correlation coefficient, and calculates the mean and standard deviation of each feature based on the reference behavior data of all independent time periods.

[0009] The intervention plan output unit obtains the impact coefficient of real-time behavioral data and outputs the corresponding behavioral gap type and intervention plan based on the impact coefficient.

[0010] Preferably, the feature fusion unit includes a historical data acquisition module, a feature extraction module, and a representation vector generation module;

[0011] The historical data acquisition module collects multi-source historical behavioral data of users and records the time, and obtains the user's self-reported emotion score through the ecological instantaneous assessment method;

[0012] The method of obtaining user self-reported emotion scores through ecological instantaneous assessment involves repeatedly issuing survey prompts through a smart device carried by the user. When a prompt appears, the user must immediately interrupt the current activity and complete a short electronic questionnaire. The electronic questionnaire includes a quantitative score of the current emotional state (e.g., 1-10 points). This method can capture the user's instantaneous emotions in a specific context in real time, effectively avoiding the errors caused by memory bias in traditional retrospective reports, thereby obtaining a dynamic emotion data stream with high ecological validity.

[0013] The feature extraction module uses a timestamp alignment algorithm to unify various behavioral data in the same time period onto the same time axis, cleans the data, and extracts different modal feature values ​​from the cleaned raw data.

[0014] The process of unifying various behavioral data within the same time period onto the same time axis using a timestamp alignment algorithm specifically involves:

[0015] S101. Collect raw data. Each data point carries a timestamp, but its precision and time zone may be inconsistent. Convert each data point into a standard time format and ensure that all data are under the same time zone reference to establish a common time language for subsequent alignment.

[0016] S102. Define the time window as the basic unit of alignment. Select the alignment strategy according to the analysis target. For high-frequency data, downsampling is usually performed. For low-frequency and event data, determine which time window it occurs in and classify it into that window.

[0017] S103. Slice the entire time period into segments using a defined time window, iterate through each data entry, and assign it to a specific time window based on its standardized timestamp.

[0018] S104. Within each time window, the various types of data allocated there will be aggregated to generate a unified data record for each time window that contains all behavioral dimension features, thereby forming a complete and aligned time series data for subsequent analysis.

[0019] The representation vector generation module utilizes a cross-modal attention mechanism to fuse features from different modalities at the same time step, and uses the fused result to generate a representation vector for user behavior data.

[0020] The specific method of utilizing cross-modal attention mechanism to fuse features from different modalities at the same time step is as follows:

[0021] S201. Project the feature values ​​of different modalities into a shared semantic subspace through independent linear transformations (fully connected layers) to make them have the same dimension, thereby achieving modal alignment and laying the foundation for subsequent interactive computation.

[0022] S202. Construct attention relationships for each pair of modalities that need to interact. For the target modality, define its projected feature vector as the query. Define the projected feature vectors of the source modality as the key and value, respectively. The key is used to calculate the relevance with the query, and the value is used to aggregate information. This process is bidirectional or multidirectional. Each modality can be used as the target modality in turn to capture the mutual influence between all modalities.

[0023] S203. Perform a dot product operation between the query vector and all key vectors to evaluate their relevance. After scaling, the calculation result is normalized by the Softmax function to generate a set of attention weights, which represent the importance distribution of source modality information to target modality.

[0024] S204. Use weights to perform a weighted summation of the value vectors to generate a context vector for the target modality;

[0025] S205. The generated context vector is fused with the original projection features of the target modality. The fusion operation usually adopts element-wise addition or vector concatenation. Addition can achieve compact integration of information, while concatenation can retain more complete features, thereby outputting a new enhanced feature representation that integrates multimodal information. This representation integrates the filtered complementary context from other modalities.

[0026] Preferably, the feature association analysis unit includes a historical data extraction module, a reference data determination module, and a vector analysis module;

[0027] The historical data extraction module divides a day into multiple independent time periods. After determining the corresponding independent time period according to the time of the user's historical behavior data, it sets the size of the review window and extracts the user's historical behavior data and emotion score in each independent time period according to the size of the review window.

[0028] The reference data determination module sets a threshold. Then, read the first... Independent time period User historical behavior data and mood rating ,in Indicates the first Independent time period The Individual user historical behavior data, Indicates the first Independent time period The Sentiment scores corresponding to each user's historical behavior data. Indicates the parameter, if the first Independent time period The sentiment score corresponding to the first user's historical behavior data Then the first Independent time period The first user historical behavior data Save to the review window, otherwise delete. Repeat the operation until the amount of data stored in the review window matches the window size, and treat all historical behavior data within the current window as an independent time period. Reference behavioral data;

[0029] The vector analysis module statistically analyzes the reference behavior data for all independent time periods and then extracts the representation vector of the reference behavior data.

[0030] Preferably, the feature association analysis unit further includes a matrix construction module and a correlation coefficient calculation module;

[0031] The matrix construction module will use independent time periods. The reference behavioral data representation vectors are stacked to form a The data matrix, where Indicates time period The total number of reference behavioral data, The matrix represents the dimension of the behavioral data representation vector, with each column representing the value of a feature across all reference behavioral data.

[0032] The correlation coefficient calculation module uses the correlation coefficient formula to calculate the degree of linear correlation between every two columns in the data matrix, thereby obtaining a... A symmetric matrix.

[0033] Preferably, the relevant group determination unit includes a feature selection module, a mean calculation module, and a hole list generation module;

[0034] The feature selection module arbitrarily selects a feature as the target feature, traverses the entire row position value corresponding to the target feature in the symmetric matrix, and if the position value is greater than zero and the absolute value is greater than 0.5, the column number of the position value is added to the positive correlation group of the target feature. If the position value is less than zero and the absolute value is greater than 0.5, the column number of the position value is added to the negative correlation group of the target feature. The operation is repeated until all features are selected.

[0035] After the mean calculation module counts the positive and negative correlation groups for each feature, it calculates the mean and standard deviation for each feature based on the reference behavioral data representation vector for independent time periods.

[0036] The void list generation module constructs a corresponding behavioral void list for each independent time period. The behavioral void list contains the influence threshold of each feature and the void type to which it belongs. The void types include physical activity deficiency void, social withdrawal void, attention distraction void, and sleep disorder void.

[0037] Preferably, the real-time data analysis unit includes a real-time data reading module, an offset analysis module, and an adjacent feature acquisition module;

[0038] After the real-time data reading module determines the mean, standard deviation, positive correlation group and negative correlation group of each feature in the reference behavior data of all independent time periods, it obtains the current user's real-time behavior data, determines the independent time periods of the real-time behavior data, and extracts the representation vector of the real-time behavior data.

[0039] The offset analysis module is based on the mean of each feature within the reference behavior data for independent time periods. Standard deviation and the representation vector of real-time behavioral data ,in The offset between each feature value in the reference behavior data and the real-time behavior data is calculated. ,in , The first representing real-time behavioral data 1 eigenvalue, Indicates the first The mean value corresponding to each feature Indicates the first The standard deviation corresponding to each feature Indicates the first The offset corresponding to each feature This represents the dimension of the behavioral data representation vector;

[0040] If the adjacent feature acquisition module is... The absolute value of each feature offset Then extract the first The positive and negative correlation groups for each feature are determined according to the column numbers within the positive and negative correlation groups. Each feature corresponds to a positive neighboring feature and a negative neighboring feature.

[0041] Preferably, the real-time data analysis unit further includes a synergy coefficient output module and an influence coefficient calculation module;

[0042] The collaborative coefficient output module obtains the offset of positive adjacent features and determines... Is it greater than zero? If the offset is greater than zero, then iterate through the offsets of positive and negative adjacent features respectively, and count the offsets greater than zero. The number of positive adjacent features and the offset are less than The number of negative neighbor features is calculated, and the sum of the two is taken as the first value. The synergy coefficient of each feature, if If the offset is less than zero, then iterate through the offsets of positive and negative adjacent features respectively, and count the offsets less than zero. The number of positive adjacent features and the offset are greater than The number of negative neighbor features is calculated, and the sum of the two is taken as the first value. The synergy coefficient of each feature;

[0043] The influence coefficient calculation module reads the first... The absolute values ​​of the synergy coefficient and the offset of each feature. The characteristic ordinal number is used to calculate the influence coefficient of the characteristic based on the synergy coefficient and absolute value.

[0044] Preferably, the intervention plan output unit includes a type determination module, a plan level classification module, and an operation execution module;

[0045] The type determination module counts all offset absolute values ​​in the current independent time period that are greater than a preset value. After identifying the features, the influence coefficient of each feature is analyzed. Then, the list of behavioral gaps for independent time periods is retrieved. The influence threshold of the features in the gap list is compared with the corresponding influence coefficient. If the influence coefficient is greater than the threshold, the corresponding behavioral gap type is output.

[0046] The program level classification module stores multi-level intervention programs corresponding to different behavioral gap types in the database. The multi-level intervention programs include Level 1 basic intervention, Level 2 moderate intervention, and Level 3 deep intervention. The level is determined by the magnitude of the influence coefficient. Level 1 basic intervention mainly pushes emotional comfort content, Level 2 moderate intervention mainly distributes interactive guidance tasks, and Level 3 deep intervention mainly connects with professional psychological counseling.

[0047] The operation execution module extracts the corresponding intervention plan from the database according to the behavioral void type and the impact coefficient, and executes the intervention operation on the user terminal.

[0048] Compared with the prior art, the beneficial effects of the present invention are:

[0049] 1. This invention, through continuous behavioral monitoring, ensures that subtle deviations in behavioral patterns can be identified before the user is even consciously aware of them. This early warning capability for emotional risks breaks the traditional passive model of "problem occurs - actively seeking help," allowing for intervention to be initiated much earlier, thereby achieving true preventive mental health care. Furthermore, through multimodal data fusion and collaborative verification, its evaluation results have higher credibility and validity, providing a reliable basis for intervention.

[0050] 2. This invention can identify behavioral gaps and generate corresponding intervention operations, thereby realizing personalized micro-interventions. From data collection to intervention execution, historical behavioral data is continuously used to optimize intervention strategies, thereby increasing the system's adaptability. Attached Figure Description

[0051] Figure 1 A schematic diagram of the overall system flow is provided for embodiments of the present invention;

[0052] Figure 2 This is an internal module block diagram of the feature fusion unit provided in an embodiment of the present invention;

[0053] Figure 3 This is an internal module block diagram of the related group determination unit provided in an embodiment of the present invention;

[0054] Figure 4 This is an internal module block diagram of the intervention scheme output unit provided in an embodiment of the present invention.

[0055] In the diagram: 1. Feature Fusion Unit; 101. Historical Data Acquisition Module; 102. Feature Extraction Module; 103. Representation Vector Generation Module; 2. Feature Association Analysis Unit; 201. Historical Data Extraction Module; 202. Reference Data Determination Module; 203. Vector Analysis Module; 204. Matrix Construction Module; 205. Correlation Coefficient Calculation Module; 3. Correlation Group Determination Unit; 301. Feature Selection Module; 302. Mean Calculation Module; 303. Hollow List Generation Module; 4. Real-time Data Analysis Unit; 401. Real-time Data Reading Module; 402. Offset Analysis Module; 403. Adjacent Feature Acquisition Module; 404. Synergy Coefficient Output Module; 405. Influence Coefficient Calculation Module; 5. Intervention Plan Output Unit; 501. Type Determination Module; 502. Plan Level Classification Module; 503. Operation Execution Module. Detailed Implementation

[0056] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0057] Please see Figure 1 - Figure 4 The present invention provides a technical solution: an emotion intervention system based on behavioral characteristics, including a real-time data analysis unit 4;

[0058] Feature fusion unit 1 collects multi-source historical behavioral data and emotion scores of users, and analyzes the representation vector of the behavioral data;

[0059] Feature correlation analysis unit 2 divides a day into multiple independent time periods, statistically analyzes the reference behavior data of all independent time periods, and calculates the correlation coefficient between each feature in the representation vector.

[0060] Correlation group determination unit 3: Correlation group determination unit 3 counts the positive and negative correlation groups of each feature according to the correlation coefficient, and calculates the mean and standard deviation of each feature based on the reference behavior data of all independent time periods;

[0061] Intervention plan output unit 5: After obtaining the impact coefficient of real-time behavioral data, intervention plan output unit 5 outputs the corresponding behavioral gap type and intervention plan based on the impact coefficient.

[0062] Feature fusion unit 1 includes a historical data acquisition module 101, a feature extraction module 102, and a representation vector generation module 103;

[0063] The historical data acquisition module 101 collects multi-source historical behavioral data of users and records the time, and obtains the user's self-reported emotion score through the ecological instantaneous assessment method;

[0064] The method of obtaining user self-reported emotion scores through ecological instantaneous assessment involves repeatedly issuing survey prompts through a smart device carried by the user. When a prompt appears, the user must immediately interrupt the current activity and complete a short electronic questionnaire. The electronic questionnaire includes a quantitative score of the current emotional state (e.g., 1-10 points). This method can capture the user's instantaneous emotions in a specific context in real time, effectively avoiding the errors caused by memory bias in traditional retrospective reports, thereby obtaining a dynamic emotion data stream with high ecological validity.

[0065] The feature extraction module 102 uses a timestamp alignment algorithm to unify various behavioral data in the same time period onto the same time axis, cleans the data, and extracts different modal feature values ​​from the cleaned raw data.

[0066] The process of unifying various behavioral data within the same time period onto the same time axis using a timestamp alignment algorithm specifically involves:

[0067] S101. Collect raw data. Each data point carries a timestamp, but its precision and time zone may be inconsistent. Convert each data point into a standard time format and ensure that all data are under the same time zone reference to establish a common time language for subsequent alignment.

[0068] S102. Define the time window as the basic unit of alignment. Select the alignment strategy according to the analysis target. For high-frequency data, downsampling is usually performed. For low-frequency and event data, determine which time window it occurs in and classify it into that window.

[0069] S103. Slice the entire time period into segments using a defined time window, iterate through each data entry, and assign it to a specific time window based on its standardized timestamp.

[0070] S104. Within each time window, the various types of data allocated there will be aggregated to generate a unified data record for each time window that contains all behavioral dimension features, thereby forming a complete and aligned time series data for subsequent analysis.

[0071] The representation vector generation module 103 utilizes a cross-modal attention mechanism to fuse features from different modalities at the same time step and uses the fused result to generate a representation vector for user behavior data.

[0072] The specific method of utilizing cross-modal attention mechanism to fuse features from different modalities at the same time step is as follows:

[0073] S201. Project the feature values ​​of different modalities into a shared semantic subspace through independent linear transformations (fully connected layers) to make them have the same dimension, thereby achieving modal alignment and laying the foundation for subsequent interactive computation.

[0074] S202. Construct attention relationships for each pair of modalities that need to interact. For the target modality, define its projected feature vector as the query. Define the projected feature vectors of the source modality as the key and value, respectively. The key is used to calculate the relevance with the query, and the value is used to aggregate information. This process is bidirectional or multidirectional. Each modality can be used as the target modality in turn to capture the mutual influence between all modalities.

[0075] S203. Perform a dot product operation between the query vector and all key vectors to evaluate their relevance. After scaling, the calculation result is normalized by the Softmax function to generate a set of attention weights, which represent the importance distribution of source modality information to target modality.

[0076] S204. Use weights to perform a weighted summation of the value vectors to generate a context vector for the target modality;

[0077] S205. The generated context vector is fused with the original projection features of the target modality. The fusion operation usually adopts element-wise addition or vector concatenation. Addition can achieve compact integration of information, while concatenation can retain more complete features, thereby outputting a new enhanced feature representation that integrates multimodal information. This representation integrates the filtered complementary context from other modalities.

[0078] Feature association analysis unit 2 includes a historical data extraction module 201, a reference data determination module 202, and a vector analysis module 203;

[0079] The historical data extraction module 201 divides a single day into multiple independent time periods. After determining the corresponding independent time period according to the time of the user's historical behavior data, it sets the size of the review window and extracts the user's historical behavior data and sentiment scores in each independent time period according to the size of the review window.

[0080] Reference data determination module 202 sets threshold Then, read the first... Independent time period User historical behavior data and mood rating ,in Indicates the first Independent time period The Individual user historical behavior data, Indicates the first Independent time period The Sentiment scores corresponding to each user's historical behavior data. Indicates the parameter, if the first Independent time period The sentiment score corresponding to the first user's historical behavior data Then the first Independent time period The first user historical behavior data Save to the review window, otherwise delete. Repeat the operation until the amount of data stored in the review window matches the window size, and treat all historical behavior data within the current window as an independent time period. Reference behavioral data;

[0081] After the vector analysis module 203 statistically analyzes the reference behavior data for all independent time periods, it extracts the representation vector of the reference behavior data.

[0082] Feature association analysis unit 2 also includes matrix construction module 204 and correlation coefficient calculation module 205;

[0083] Matrix construction module 204 will have independent time periods The reference behavioral data representation vectors are stacked to form a The data matrix, where Indicates time period The total number of reference behavioral data, The matrix represents the dimension of the behavioral data representation vector, with each column representing the value of a feature across all reference behavioral data.

[0084] The correlation coefficient calculation module 205 uses the correlation coefficient formula to calculate the degree of linear correlation between every two columns in the data matrix, thereby obtaining a correlation coefficient. The symmetric matrix is ​​defined where the value in the first row and second column represents the correlation coefficient between the first and second features. The specific formula for the correlation coefficient is as follows:

[0085]

[0086] in, Representation of features With features The correlation coefficient between them Representation of features The One value, Representation of features The One value, Representation of features The mean, Representation of features The mean, This represents the total number of eigenvalues. Indicates parameters;

[0087] The relevant group determination unit 3 includes a feature selection module 301, a mean calculation module 302, and a hole list generation module 303;

[0088] The feature selection module 301 arbitrarily selects a feature as the target feature, traverses the entire row position value corresponding to the target feature in the symmetric matrix, if the position value is greater than zero and the absolute value is greater than 0.5, then the column number of the position value is added to the positive correlation group of the target feature, if the position value is less than zero and the absolute value is greater than 0.5, then the column number of the position value is added to the negative correlation group of the target feature, and the operation is repeated until all features are selected;

[0089] After the mean calculation module 302 counts the positive and negative correlation groups for each feature, it calculates the mean and standard deviation for each feature based on the reference behavioral data representation vector for independent time periods.

[0090] The void list generation module 303 constructs a corresponding behavioral void list for each independent time period. The behavioral void list contains the influence threshold of each feature and the void type to which it belongs. The void types include physical activity deficiency void, social withdrawal void, attention distraction void, and sleep disorder void.

[0091] The real-time data analysis unit 4 includes a real-time data reading module 401, an offset analysis module 402, and an adjacent feature acquisition module 403;

[0092] After the real-time data reading module 401 determines the mean, standard deviation, positive correlation group and negative correlation group of each feature in the reference behavior data of all independent time periods, it obtains the current user's real-time behavior data, determines the independent time period of the real-time behavior data, and extracts the representation vector of the real-time behavior data.

[0093] The offset analysis module 402 calculates the mean of each feature within the reference behavioral data for independent time periods. Standard deviation and the representation vector of real-time behavioral data ,in The offset between each feature value in the reference behavior data and the real-time behavior data is calculated. ,in , The first representing real-time behavioral data 1 eigenvalue, Indicates the first The mean value corresponding to each feature Indicates the first The standard deviation corresponding to each feature Indicates the first The offset corresponding to each feature This represents the dimension of the behavioral data representation vector;

[0094] If the adjacent feature acquisition module 403 is the first The absolute value of each feature offset Then extract the first The positive and negative correlation groups for each feature are determined according to the column numbers within the positive and negative correlation groups. Each feature corresponds to a positive neighboring feature and a negative neighboring feature;

[0095] The real-time data analysis unit 4 also includes a synergy coefficient output module 404 and an influence coefficient calculation module 405;

[0096] The collaborative coefficient output module 404 obtains the offset of positively adjacent features and determines... Is it greater than zero? If the offset is greater than zero, then iterate through the offsets of positive and negative adjacent features respectively, and count the offsets greater than zero. The number of positive adjacent features and the offset are less than The number of negative neighbor features is calculated, and the sum of the two is taken as the first value. The synergy coefficient of each feature, if If the offset is less than zero, then iterate through the offsets of positive and negative adjacent features respectively, and count the offsets less than zero. The number of positive adjacent features and the offset are greater than The number of negative neighbor features is calculated, and the sum of the two is taken as the first value. The synergy coefficient of each feature;

[0097] Influence coefficient calculation module 405 reads the first The absolute values ​​of the synergy coefficient and the offset of each feature. The feature ordinal number is used to calculate the influence coefficient of the feature based on the synergy coefficient and absolute value.

[0098] The intervention plan output unit 5 includes a type determination module 501, a plan level classification module 502, and an operation execution module 503;

[0099] The type determination module 501 counts all offsets in the current independent time period whose absolute values ​​are greater than a preset value. After identifying the features, the influence coefficient of each feature is analyzed. Then, the list of behavioral gaps for independent time periods is retrieved. The influence threshold of the features in the gap list is compared with the corresponding influence coefficient. If the influence coefficient is greater than the threshold, the corresponding behavioral gap type is output.

[0100] The program level classification module 502 stores multi-level intervention programs corresponding to different behavioral gap types in the database. The multi-level intervention programs include Level 1 basic intervention, Level 2 moderate intervention and Level 3 deep intervention. The level is determined by the magnitude of the impact coefficient. Among them, Level 1 basic intervention mainly pushes emotional comfort content, Level 2 moderate intervention mainly distributes interactive guidance tasks, and Level 3 deep intervention mainly connects with professional psychological counseling.

[0101] The operation execution module 503 extracts the corresponding intervention plan from the database according to the behavioral void type and the impact coefficient, and executes the intervention operation on the user terminal.

[0102] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0103] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An emotion intervention system based on behavioral characteristics, comprising a real-time data analysis unit (4), characterized in that: Feature fusion unit (1) collects multi-source historical behavioral data and emotion scores of users, and analyzes the representation vector of behavioral data; Feature association analysis unit (2) divides a day into multiple independent time periods, statistically analyzes the reference behavior data of all independent time periods, and calculates the correlation coefficient between each feature in the representation vector. The correlation group determination unit (3) counts the positive and negative correlation groups of each feature according to the correlation coefficient, and calculates the mean and standard deviation of each feature based on the reference behavior data of all independent time periods. Intervention program output unit (5): After obtaining the impact coefficient of real-time behavioral data, the intervention program output unit (5) outputs the corresponding behavioral gap type and intervention program according to the impact coefficient.

2. The emotion intervention system based on behavioral characteristics according to claim 1, characterized in that: The feature fusion unit (1) includes a historical data acquisition module (101), a feature extraction module (102), and a representation vector generation module (103). The historical data acquisition module (101) collects multi-source historical behavior data of users and records the time, and obtains the user's self-reported emotion score through the ecological instantaneous assessment method; The feature extraction module (102) uses a timestamp alignment algorithm to unify various behavioral data in the same time period onto the same time axis, cleans the data, and extracts different modal feature values ​​from the cleaned raw data. The representation vector generation module (103) utilizes a cross-modal attention mechanism to fuse different modal features at the same time step and uses the fused result to generate a representation vector of user behavior data.

3. The emotion intervention system based on behavioral characteristics according to claim 1, characterized in that: The feature association analysis unit (2) includes a historical data extraction module (201), a reference data determination module (202), and a vector analysis module (203). The historical data extraction module (201) divides a day into multiple independent time periods. After determining the corresponding independent time period according to the time of the user's historical behavior data, it sets the size of the review window and extracts the user's historical behavior data and emotion score in each independent time period according to the size of the review window. The reference data determination module (202) sets a threshold. Then, read the first... Independent time period User historical behavior data and mood rating ,in Indicates the first Independent time period The Individual user historical behavior data, Indicates the first Independent time period The Sentiment scores corresponding to each user's historical behavior data. Indicates the parameter, if the first Independent time period The sentiment score corresponding to the first user's historical behavior data Then the first Independent time period The first user historical behavior data Save to the review window, otherwise delete. Repeat the operation until the amount of data stored in the review window matches the window size, and treat all historical behavior data within the current window as an independent time period. Reference behavioral data; The vector analysis module (203) statistically analyzes the reference behavior data for all independent time periods and then analyzes the representation vector of the reference behavior data.

4. The emotion intervention system based on behavioral characteristics according to claim 3, characterized in that: The feature association analysis unit (2) also includes a matrix construction module (204) and a correlation coefficient calculation module (205). The matrix construction module (204) will separate time periods The reference behavioral data representation vectors are stacked to form a The data matrix, where Indicates time period The total number of reference behavioral data, The matrix represents the dimension of the behavioral data representation vector, with each column representing the value of a feature across all reference behavioral data. The correlation coefficient calculation module (205) uses the correlation coefficient formula to calculate the degree of linear correlation between every two columns in the data matrix, thereby obtaining a correlation coefficient. A symmetric matrix.

5. The emotion intervention system based on behavioral characteristics according to claim 1, characterized in that: The related group determination unit (3) includes a feature selection module (301), a mean calculation module (302), and a hole list generation module (303). The feature selection module (301) arbitrarily selects a feature and uses it as the target feature. Traverse the entire row position value corresponding to the target feature in the symmetric matrix. If the position value is greater than zero and the absolute value is greater than 0.5, add the column number of the position value to the positive correlation group of the target feature. If the position value is less than zero and the absolute value is greater than 0.5, add the column number of the position value to the negative correlation group of the target feature. Repeat the operation until all features are selected. The mean calculation module (302) counts the positive and negative correlation groups for each feature, and then calculates the mean and standard deviation for each feature based on the reference behavioral data representation vector for independent time periods. The void list generation module (303) constructs a corresponding behavioral void list for each independent time period. The behavioral void list contains the influence threshold and void type of each feature. The void types include physical activity deficiency void, social withdrawal void, attention distraction void, and sleep disorder void.

6. The emotion intervention system based on behavioral characteristics according to claim 1, characterized in that: The real-time data analysis unit (4) includes a real-time data reading module (401), an offset analysis module (402), and an adjacent feature acquisition module (403). After the real-time data reading module (401) determines the mean, standard deviation, positive correlation group and negative correlation group of each feature in the reference behavior data of all independent time periods, it obtains the current user's real-time behavior data, determines the independent time period of the real-time behavior data, and extracts the representation vector of the real-time behavior data. The offset analysis module (402) calculates the mean value of each feature within the reference behavior data for independent time periods. Standard deviation and the representation vector of real-time behavioral data ,in The offset between each feature value in the reference behavior data and the real-time behavior data is calculated. ,in , The first representing real-time behavioral data 1 eigenvalue, Indicates the first The mean value corresponding to each feature Indicates the first The standard deviation corresponding to each feature Indicates the first The offset corresponding to each feature This represents the dimension of the behavioral data representation vector; If the adjacent feature acquisition module (403) has the following characteristics: The absolute value of each feature offset Then extract the first The positive and negative correlation groups for each feature are determined according to the column numbers within the positive and negative correlation groups. Each feature corresponds to a positive neighboring feature and a negative neighboring feature.

7. The emotion intervention system based on behavioral characteristics according to claim 6, characterized in that: The real-time data analysis unit (4) also includes a synergy coefficient output module (404) and an influence coefficient calculation module (405). The collaborative coefficient output module (404) obtains the offset of positive adjacent features and determines... Is it greater than zero? If the offset is greater than zero, then iterate through the offsets of positive and negative adjacent features respectively, and count the offsets greater than zero. The number of positive adjacent features and the offset are less than The number of negative neighbor features is calculated, and the sum of the two is taken as the first value. The synergy coefficient of each feature, if If the offset is less than zero, then iterate through the offsets of positive and negative adjacent features respectively, and count the offsets less than zero. The number of positive adjacent features and the offset are greater than The number of negative neighbor features is calculated, and the sum of the two is taken as the first value. The synergy coefficient of each feature; The influence coefficient calculation module (405) reads the first... The absolute values ​​of the synergy coefficient and the offset of each feature. The characteristic ordinal number is used to calculate the influence coefficient of the characteristic based on the synergy coefficient and absolute value.

8. The emotion intervention system based on behavioral characteristics according to claim 1, characterized in that: The intervention plan output unit (5) includes a type determination module (501), a plan level classification module (502), and an operation execution module (503). The type determination module (501) counts all absolute values ​​of offsets in the current independent time period that are greater than a preset value. After identifying the features, the influence coefficient of each feature is analyzed. Then, the list of behavioral gaps for independent time periods is retrieved. The influence threshold of the features in the gap list is compared with the corresponding influence coefficient. If the influence coefficient is greater than the threshold, the corresponding behavioral gap type is output. The program level classification module (502) stores multi-level intervention programs corresponding to different behavioral void types in the database. The multi-level intervention programs include Level 1 basic intervention, Level 2 moderate intervention and Level 3 deep intervention. The level is determined by the magnitude of the influence coefficient. Level 1 basic intervention mainly pushes emotional comfort content, Level 2 moderate intervention mainly distributes interactive guidance tasks, and Level 3 deep intervention mainly connects with professional psychological counseling. The operation execution module (503) extracts the corresponding intervention plan from the database according to the behavioral void type and the impact coefficient, and then executes the intervention operation on the user terminal.

Images