Behavioral intervention system for neurodevelopmental disorders based on large language model

Through large language models, analyzing children's behavioral timing data is provided, and personalized intervention plans are solved, which lacks scientific basis and multi-dimensional considerations in traditional intervention methods, improves targetedness and effectiveness, and promotes the all-round development of children.

CN118690330BActive Publication Date: 2024-11-12LITTLE NEURON (HANGZHOU) MEDICAL TECH CO LTD
View PDF 0 Cites 2 Cited by

Patent Information

Application Number
CN202411157330.1
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-08-22
Publication Date
2024-11-12
Estimated Expiration
2044-08-22

AI Technical Summary

Technical Problem

Traditional behavioral intervention methods for neurodevelopmental disabilities lack scientific basis, difficulty in assessing effectiveness, and neglecting the dynamic and multi-dimensional characteristics of children's behavior, resulting in the inability to effectively target individual needs.

Method used

A behavioral intervention system for neurodevelopmental disorders based on a large language model is adopted to collect and analyze the behavioral timing data of children, extract semantic characteristics and behavior patterns, and provide personalized intervention plans, including intervention goals, methods and steps.

Benefits of technology

It improves the pertinence and effectiveness of behavioral interventions in children with neurodevelopmental disorders, meets the unique needs of different children, and promotes their comprehensive development.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118690330B_ABST
    Figure CN118690330B_ABST
Patent Text Reader

Abstract

The present application discloses a neurodevelopmental disorder behavior intervention system based on a large language model, which collects the behavioral time series data of the child subject, including language, social interaction and behavioral performance patterns, and uses the large language model and artificial intelligence technology to extract the semantic features and behavior time series aggregation semantics of each behavior data of the child subject, so as to automatically identify the behavior pattern and dynamic change information of the child subject, and provide a scientific basis for the formulation and optimization of intervention strategies. In this way, it is possible to provide personalized intervention plans more intelligently based on the multi-dimensional behavior time series aggregation semantics of the child subject, including intervention goals, intervention methods and steps, so as to improve the pertinence of behavioral intervention for children with neurodevelopmental disorders, so as to better meet the needs of different children and promote their all-round development.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of intelligent intervention, and more specifically, to a neurodevelopmental disorder behavioral intervention system based on a large language model. Background Art

[0002] Neurodevelopmental disorders (NDDs) are a class of diseases that affect children's cognitive, social and behavioral development, including autism spectrum disorder (ASD), attention deficit hyperactivity disorder (ADHD), learning disabilities, etc. As the understanding of these disorders deepens, the importance of early intervention becomes more prominent. Studies have shown that early identification and intervention can significantly improve children's long-term development outcomes and enhance their social skills, language skills and academic performance.

[0003] However, traditional behavioral interventions for neurodevelopmental disorders often use generalized standards and usually rely on the doctor's subjective assessment and clinical experience, which makes the formulation of intervention strategies lack scientific basis and difficult to evaluate their effectiveness. At the same time, generalized interventions lack an in-depth understanding of the unique needs of each child, which means that subsequent interventions and programs may not effectively target the behavioral patterns and development needs of specific children. In addition, traditional behavioral intervention systems often ignore the dynamics of behavioral changes in children, and are unable to comprehensively consider multiple dimensions such as language, social interaction, and behavioral performance, resulting in a one-sided understanding of children's behavior and affecting the adaptability of behavioral intervention program decisions for children with neurodevelopmental disorders.

[0004] Therefore, an optimized behavioral intervention system for neurodevelopmental disorders is desired. Summary of the invention

[0005] In order to solve the above technical problems, the present application is proposed. The embodiment of the present application provides a neurodevelopmental disorder behavior intervention system based on a large language model, which collects the behavioral time series data of the child object, including language, social interaction and behavioral performance patterns, and uses the large language model and artificial intelligence technology to extract the semantic features and behavior time series aggregation semantics of each behavior data of the child object, so as to automatically identify the behavior pattern and dynamic change information of the child object, and provide a scientific basis for the formulation and optimization of intervention strategies. In this way, it is possible to provide personalized intervention plans more intelligently based on the multidimensional behavior time series aggregation semantics of the child object, including intervention goals, intervention methods and steps, thereby improving the pertinence of behavioral intervention for children with neurodevelopmental disorders, so as to better meet the needs of different children and promote their all-round development.

[0006] According to one aspect of the present application, a neurodevelopmental disorder behavior intervention system based on a large language model is provided, which includes: a patient subject behavior data acquisition module, which is used to obtain a time queue of the patient subject's behavior data; a patient subject behavior semantic encoding module, which is used to input each behavior data in the time queue of the behavior data into a behavior semantic encoder based on the large language model to obtain a time queue of behavior semantic encoding feature vectors; a behavior semantic temporal implicit feature extraction module, which is used to perform temporal encoding on the time queue of the behavior semantic encoding feature vectors based on behavior implicit semantics to obtain a time queue of behavior semantic implicit encoding feature vectors; a behavior semantic multi-scale feature attention fusion module, which is used to input each behavior data in the time queue of the behavior data into a behavior semantic encoder based on the large language model to obtain a time queue of behavior semantic encoding feature vectors; A feature value granularity association mask distinguishable attention fusion module is input into the time queue of semantic coding feature vectors and the time queue of the behavioral semantic implicit coding feature vectors for each corresponding group of behavioral semantic coding feature vectors and behavioral semantic implicit coding feature vectors to obtain a time queue of behavioral semantic multi-scale coding feature vectors; a behavioral semantic temporal aggregation module is used to input the time queue of the behavioral semantic multi-scale coding feature vectors into a feature aggregation network based on feature energy local significance gating to obtain a behavioral semantic temporal aggregation representation vector; an intervention plan generation module is used to determine an intervention plan generation result based on the behavioral semantic temporal aggregation representation vector, and the intervention plan generation result is a type label of the intervention plan.

[0007] Compared with the prior art, the present application provides a neurodevelopmental disorder behavior intervention system based on a large language model, which collects the behavioral time series data of the child subject, including language, social interaction and behavioral performance patterns, and uses the large language model and artificial intelligence technology to extract the semantic features and behavior time series aggregation semantics of each behavior data of the child subject, so as to automatically identify the behavior pattern and dynamic change information of the child subject, and provide a scientific basis for the formulation and optimization of intervention strategies. In this way, it is possible to provide personalized intervention plans more intelligently based on the multi-dimensional behavior time series aggregation semantics of the child subject, including intervention goals, intervention methods and steps, thereby improving the pertinence of behavioral intervention for children with neurodevelopmental disorders, so as to better meet the needs of different children and promote their all-round development. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] By describing the embodiments of the present application in more detail in conjunction with the accompanying drawings, the above and other purposes, features and advantages of the present application will become more apparent. The accompanying drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0009] Figure 1It is a block diagram of a neurodevelopmental disorder behavioral intervention system based on a large language model according to an embodiment of the present application.

[0010] Figure 2 Schematic diagram of data flow of a neurodevelopmental disorder behavioral intervention system based on a large language model according to an embodiment of the present application.

[0011] Figure 3 It is a block diagram of a behavior semantic temporal aggregation module in a neurodevelopmental disorder behavior intervention system based on a large language model according to an embodiment of the present application.

[0012] Figure 4 It is a block diagram of another embodiment of the neurodevelopmental disorder behavioral intervention system based on a large language model according to an embodiment of the present application. DETAILED DESCRIPTION

[0013] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described here.

[0014] As shown in this application and claims, unless the context clearly indicates an exception, the words "a", "an", "an" and / or "the" do not refer to the singular and may also include the plural. Generally speaking, the terms "include" and "comprise" only indicate the inclusion of the steps and elements that have been clearly identified, and these steps and elements do not constitute an exclusive list. The method or device may also include other steps or elements.

[0015] Although the present application makes various references to certain modules in the system according to the embodiments of the present application, any number of different modules can be used and run on the user terminal and / or server. The modules are only illustrative, and different aspects of the system and method can use different modules.

[0016] Flowcharts are used in the present application to illustrate the operations performed by the system according to the embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed accurately in order. On the contrary, various steps may be processed in reverse order or simultaneously as required. Meanwhile, other operations may also be added to these processes, or a certain step or several steps of operations may be removed from these processes.

[0017] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described here.

[0018] Traditional behavioral interventions for neurodevelopmental disorders often use generalized standards and usually rely on the doctor's subjective assessment and clinical experience, which makes the formulation of intervention strategies lack scientific basis and difficult to evaluate their effectiveness. At the same time, generalized interventions lack an in-depth understanding of the unique needs of each child, which means that subsequent interventions and programs may not be able to effectively target the behavioral patterns and development needs of specific children. In addition, traditional behavioral intervention systems often ignore the dynamics of children's behavioral changes, and are unable to comprehensively consider multiple dimensions such as language, social interaction, and behavioral performance, resulting in a one-sided understanding of children's behavior and affecting the adaptability of behavioral intervention program decisions for children with neurodevelopmental disorders.

[0019] In recent years, large language models (LLMs) have made significant progress in natural language processing, sentiment analysis and other fields. Large language models can effectively capture the semantic information of children's behavioral data, covering language, social interaction and behavioral performance patterns, and perform complex behavioral pattern recognition and temporal dynamic analysis to more comprehensively understand the evolution of children's behavior. Applying large language models to the analysis of behavioral data can help identify children's behavioral patterns and changing trends, and provide a scientific basis for the formulation of intervention strategies.

[0020] In the technical solution of the present application, a neurodevelopmental disorder behavioral intervention system based on a large language model is proposed. Figure 1 It is a block diagram of a neurodevelopmental disorder behavioral intervention system based on a large language model according to an embodiment of the present application. Figure 2 FIG. 1 is a data flow diagram of a neurodevelopmental disorder behavior intervention system based on a large language model according to an embodiment of the present application. Figure 1 and Figure 2As shown, according to the embodiment of the present application, the neurodevelopmental disorder behavior intervention system based on the large language model includes: a patient subject behavior data acquisition module 310, which is used to obtain a time queue of the patient subject's behavior data; a patient behavior semantic encoding module 320, which is used to input each behavior data in the time queue of the behavior data into a behavior semantic encoder based on the large language model to obtain a time queue of the behavior semantic encoding feature vector; a behavior semantic temporal implicit feature extraction module 330, which is used to perform temporal encoding based on behavior implicit semantics on the time queue of the behavior semantic encoding feature vector to obtain a time queue of the behavior semantic implicit encoding feature vector; a behavior semantic multi-scale feature attention fusion module 340, which is used to input the behavior semantic encoding feature vector into a behavior semantic encoder based on the large language model to obtain a time queue of the behavior semantic implicit encoding feature vector; Each group of corresponding behavior semantic encoding feature vectors and behavior semantic implicit encoding feature vectors in the time queue of behavior semantic encoding feature vectors and the time queue of the behavior semantic implicit encoding feature vectors are input into the feature value granularity association mask distinguishable attention fusion module to obtain the time queue of behavior semantic multi-scale encoding feature vectors; a behavior semantic temporal aggregation module 350 is used to input the time queue of the behavior semantic multi-scale encoding feature vectors into a feature aggregation network based on feature energy local significance gating to obtain a behavior semantic temporal aggregation representation vector; an intervention plan generation module 360 ​​is used to determine the intervention plan generation result based on the behavior semantic temporal aggregation representation vector, and the intervention plan generation result is the type label of the intervention plan.

[0021] In particular, the patient object behavior data acquisition module 310 and the patient object behavior semantic encoding module 320 are used to obtain the time queue of the patient object's behavior data; and each behavior data in the time queue of the behavior data is input into the behavior semantic encoder based on the large language model to obtain the time queue of the behavior semantic encoding feature vector. It should be understood that since the behavior data of each patient object contains the behavior semantics and feature information about the child with neurodevelopmental disorders at a specific time point, however, the behavior data usually involves multiple dimensions, such as language expression, social interaction and emotional response. The large language model (LLM) can effectively understand the hidden semantics of the data and generate natural language, and has a strong data semantic understanding ability. Therefore, the large language model can comprehensively analyze these diverse behavior data, thereby generating rich semantic encoding features and enhancing the comprehensive and in-depth understanding of the behavior of the patient object. Based on this, in the technical solution of the present application, each behavior data in the time queue of the behavior data is further input into the behavior semantic encoder based on the large language model to obtain the time queue of the behavior semantic encoding feature vector. By inputting each behavioral data into the behavioral semantic encoder, the deeper semantic features of the behavioral data of the children at different time points can be extracted to help identify the children's behavioral patterns and the emotions and intentions behind them, and provide a scientific basis for the formulation of intervention strategies. In addition, the large language model has been trained with a large amount of data and has strong generalization capabilities. Inputting behavioral data into the model can make the generated feature vector more applicable to different children or different situations, thereby improving the effectiveness of the intervention plan.

[0022] In particular, the behavior semantic temporal implicit feature extraction module 330 is used to perform temporal encoding based on behavior implicit semantics on the time queue of the behavior semantic encoding feature vector to obtain a time queue of the behavior semantic implicit encoding feature vector. In a specific example of the present application, the time queue of the behavior semantic encoding feature vector is input into a behavior semantic temporal encoder based on a Bi-LSTM model to obtain a time queue of the behavior semantic implicit encoding feature vector. Considering that children's behavior data usually presents time series characteristics and exhibits dynamic characteristics that change over time. Therefore, in order to be able to more comprehensively understand the evolution process and dynamic semantics of the child's behavior, in the technical solution of the present application, the time queue of the behavior semantic encoding feature vector is further input into a behavior semantic temporal encoder based on a Bi-LSTM model to obtain a time queue of the behavior semantic implicit encoding feature vector. It should be understood that Bi-LSTM (bidirectional long short-term memory network) can effectively process time series data, while taking into account past and future behavior semantic information, and capturing the behavior dependencies between different time points. Therefore, through the processing of the behavioral semantics temporal encoder based on the Bi-LSTM model, the temporal context of behavioral semantics can be more comprehensively understood, the understanding and recognition accuracy of the children's behavioral patterns and behavioral evolution processes can be improved, and the system can be helped to effectively identify the temporal trends, periodic changes and abnormal behaviors of the children's behaviors, providing an important basis for subsequent analysis and intervention.

[0023] In particular, the behavior semantic multi-scale feature attention fusion module 340 is used to input each group of corresponding behavior semantic coding feature vectors and behavior semantic implicit coding feature vectors in the time queue of the behavior semantic coding feature vectors and the time queue of the behavior semantic implicit coding feature vectors into the feature value granularity association mask distinguishable attention fusion module to obtain the time queue of the behavior semantic multi-scale coding feature vectors. It should be understood that each behavior semantic coding feature vector in the time queue of the behavior semantic coding feature vectors respectively contains the behavior semantic coding features of the child object at each time point. Each behavior semantic implicit coding feature vector in the time queue of the behavior semantic implicit coding feature vectors respectively contains the behavior semantic temporal context implicit association features of the child object at each time point, which contains the temporal context semantic information of the child's behavior. In particular, considering that children's behavioral data often contain multiple features (such as language, social interaction, and emotional response, etc.), if we want to better understand the behavioral semantics of children, it is beneficial to improve the adaptability of the subsequent intervention plan to meet the personalized needs of different children. In the technical solution of the present application, each group of corresponding behavioral semantic encoding feature vectors and behavioral semantic implicit encoding feature vectors in the time queue of the behavioral semantic encoding feature vector and the time queue of the behavioral semantic implicit encoding feature vector are further input into the feature value granularity association mask distinguishable attention fusion module to obtain the time queue of behavioral semantic multi-scale encoding feature vectors. Through the processing of the feature value granularity association mask distinguishable attention fusion module, the behavioral semantic encoding features of children and the behavioral temporal context semantic association features can be effectively fused to form a more comprehensive representation of the behavioral semantic features of children, thereby improving the ability to understand children's behavior. In addition, the feature value granularity association mask distinguishable attention fusion module can also automatically identify and focus on the most important features for the behavior analysis of children and the formulation of subsequent intervention strategies, so as to dynamically adjust the focus point according to the fine-grained importance between the input behavior semantic encoding feature vector and the behavior semantic implicit encoding feature vector. This selective attention can help filter out noise and information irrelevant to subsequent tasks, improve the model's ability to understand the behavior patterns and semantic information of children, and provide more comprehensive data support for the formulation of personalized intervention plans. This can help intervention personnel better understand the specific needs of children and thus formulate more effective intervention strategies.

[0024] In an embodiment of the present application, each group of corresponding behavior semantic encoding feature vectors and behavior semantic implicit encoding feature vectors in the time queue of the behavior semantic encoding feature vectors and the time queue of the behavior semantic implicit encoding feature vectors is input into a feature value granularity association mask distinguishable attention fusion module to obtain a time queue of behavior semantic multi-scale encoding feature vectors, including: calculating a global eigenvalue association matrix between a group of corresponding behavior semantic encoding feature vectors and behavior semantic implicit encoding feature vectors in the time queue of the behavior semantic encoding feature vectors and the time queue of the behavior semantic implicit encoding feature vectors to obtain a behavior semantic multi-scale global eigenvalue association matrix; The behavior semantics multi-scale global eigenvalue association matrix is ​​input into a learnable gating function to obtain a behavior semantics multi-scale correlation distinguishable weight matrix; using the behavior semantics multi-scale correlation distinguishable weight matrix as a fusion mask matrix, the matrix products of the behavior semantics encoding feature vector and the behavior semantics implicit encoding feature vector and the behavior semantics multi-scale correlation distinguishable weight matrix are calculated respectively to obtain a distinguishing and strengthening behavior semantics encoding feature vector and a distinguishing and strengthening behavior semantics implicit encoding feature vector; the distinguishing and strengthening behavior semantics encoding feature vector and the distinguishing and strengthening behavior semantics implicit encoding feature vector are fused to obtain a behavior semantics multi-scale encoding feature vector.

[0025] Among them, the process of inputting the behavior semantics multi-scale global eigenvalue association matrix into a learnable gating function to obtain a behavior semantics multi-scale correlation distinguishable weight matrix includes: using the negative number of the eigenvalues ​​at each position in the behavior semantics multi-scale global eigenvalue association matrix as the exponential power, calculating the natural exponential function value with the natural constant e as the base to obtain the behavior semantics multi-scale global eigenvalue association class support matrix; calculating the reciprocal of the sum of the eigenvalues ​​at each position in the behavior semantics multi-scale global eigenvalue association class support matrix and the constant 1 to obtain the behavior semantics multi-scale global eigenvalue association weight matrix; calculating the positional point multiplication between the behavior semantics multi-scale global eigenvalue association weight matrix and the behavior semantics multi-scale global eigenvalue association matrix to obtain the behavior semantics multi-scale correlation distinguishable weight matrix.

[0026] In summary, in the above embodiment, each group of corresponding behavior semantic coding feature vectors and behavior semantic implicit coding feature vectors in the time queue of the behavior semantic coding feature vectors and the time queue of the behavior semantic implicit coding feature vectors are input into the feature value granularity association mask distinguishable attention fusion module to obtain the time queue of the behavior semantic multi-scale coding feature vectors, including: each group of corresponding behavior semantic coding feature vectors and behavior semantic implicit coding feature vectors in the time queue of the behavior semantic coding feature vectors and the time queue of the behavior semantic implicit coding feature vectors are input into the feature value granularity association mask distinguishable attention fusion module to process with the following distinguishable attention fusion formula to obtain the time queue of the behavior semantic multi-scale coding feature vectors; wherein, the distinguishable attention fusion formula is: ;in, and The time queue of the behavior semantic coding feature vector and the time queue of the behavior semantic implicit coding feature vector are respectively the behavior semantic coding feature vector and the behavior semantic implicit coding feature vector corresponding to each group, is the behavior semantic multi-scale global eigenvalue association matrix, For The exponential operation with base , is the association weight matrix of the multi-scale global eigenvalues ​​of behavioral semantics, is the behavioral semantic multi-scale correlation distinguishable weight matrix, and They are the semantic encoding feature vector for distinguishing and reinforcing behavior and the implicit encoding feature vector for distinguishing and reinforcing behavior semantics, is the cascade operation of vectors, Each behavior semantics multi-scale encoding feature vector in the time queue of the behavior semantics multi-scale encoding feature vector.

[0027] In particular, the behavior semantic temporal aggregation module 350 is used to input the time queue of the behavior semantic multi-scale encoding feature vector into a feature aggregation network based on feature energy local significant gating to obtain a behavior semantic temporal aggregation representation vector. Since each behavior semantic multi-scale encoding feature vector in the time queue of the behavior semantic multi-scale encoding feature vector contains multi-scale encoding feature information about the behavior semantics of the child, it contains both the behavior semantic features of the child at each time point and the correlation relationship and contextual dynamic connection of such behavior semantics in time sequence. In particular, each behavior semantic multi-scale encoding feature vector has a temporal correlation and importance semantics in the time dimension. Therefore, in order to aggregate the behavior semantics of the child at these different time points, so as to more comprehensively capture and understand the joint semantics and dynamic patterns of the child's complex behavior in time sequence, and provide a basis for the subsequent formulation of intervention plans, in the technical solution of the present application, the time queue of the behavior semantic multi-scale encoding feature vector is input into a feature aggregation network based on feature energy local significant gating to obtain a behavior semantic temporal aggregation representation vector. The feature aggregation network based on local significance gating of feature energy can quantify and aggregate the local energy distribution of the child's actions to refine the feature representation, so as to identify the most important features for behavior analysis at a specific point in time. This ability to focus on significant features based on energy can help the model focus on the key behavior patterns of the child and filter out unimportant or redundant information, so as to more effectively capture the temporal change patterns and dynamic characteristics of the child's behavioral semantics, so as to generate a more representative temporal aggregation representation vector, thereby providing strong support for subsequent behavior analysis and personalized intervention for the child. In particular, in a specific example of the present application, such as Figure 3As shown, the behavior semantic temporal aggregation module 350 includes: a feature energy level coefficient calculation unit 351, which is used to determine the feature energy level coefficient of each behavior semantic multi-scale encoding feature vector based on the maximum value, minimum value, mean value and variance of each behavior semantic multi-scale encoding feature vector in the time queue of the behavior semantic multi-scale encoding feature vector to obtain a sequence of behavior semantic multi-scale feature energy level coefficients; a local neighborhood significance descriptor calculation unit 352, which is used to determine the scale of the behavior semantic multi-scale local neighborhood, calculate the local neighborhood significance descriptor of each behavior semantic multi-scale feature energy level coefficient in the behavior semantic multi-scale feature energy level coefficient. A mask gated probabilistic unit 353 is used to input the sequence of the behavioral semantics multi-scale feature energy level significance descriptors into a mask module based on a gating function to obtain a sequence of mask gated probabilistic behavioral semantics multi-scale feature energy level significance descriptors; a behavioral semantics temporal aggregation representation unit 354 is used to use the sequence of mask gated probabilistic behavioral semantics multi-scale feature energy level significance descriptors as a sequence of weights, calculate the position-weighted sum of the time queue of the behavioral semantics multi-scale encoding feature vector to obtain the behavioral semantics temporal aggregation representation vector.

[0028] Specifically, the feature energy level coefficient calculation unit 351 is used to determine the feature energy level coefficients of each behavior semantics multi-scale encoding feature vector based on the maximum value, minimum value, mean value and variance of each behavior semantics multi-scale encoding feature vector in the time queue of the behavior semantics multi-scale encoding feature vector to obtain a sequence of behavior semantics multi-scale feature energy level coefficients. In a specific example, the maximum value, minimum value, average value and variance of the behavior semantics multi-scale encoding feature vector are first calculated to obtain the behavior semantics multi-scale maximum value, the behavior semantics multi-scale minimum value, the behavior semantics multi-scale average value and the behavior semantics multi-scale variance; then, the sum of the behavior semantics multi-scale variance and the hyperparameter is calculated to obtain the first behavior semantics multi-scale local energy statistical factor; then, the difference between the behavior semantics multi-scale maximum value and the behavior semantics multi-scale average value is calculated to obtain the behavior semantics multi-scale maximum deviation value; the difference between the behavior semantics multi-scale minimum value and the behavior semantics multi-scale average value is calculated to obtain the behavior semantics multi-scale minimum deviation value; further, the sum of the squares of the behavior semantics multi-scale maximum deviation value and the behavior semantics multi-scale minimum deviation value is calculated and added to the hyperparameter to obtain the second behavior semantics multi-scale local energy statistical factor; finally, the division between the first behavior semantics multi-scale local energy statistical factor and the second behavior semantics multi-scale local energy statistical factor is calculated to obtain the behavior semantics multi-scale feature energy level coefficient.

[0029] Specifically, the local neighborhood significant descriptor calculation unit 352 is used to determine the scale of the behavior semantics multi-scale local neighborhood, and calculate the local neighborhood significant descriptor of each behavior semantics multi-scale feature energy level coefficient in the behavior semantics multi-scale feature energy level coefficient to obtain a sequence of behavior semantics multi-scale feature energy level significant descriptors. In a specific example, the scale of the behavior semantics multi-scale local neighborhood is used as the mean calculation range, and the local neighborhood mean of each behavior semantics multi-scale feature energy level coefficient is calculated respectively to obtain the sequence of behavior semantics multi-scale feature energy level significant descriptors.

[0030] Specifically, the mask-gated probabilization unit 353 is used to input the sequence of the behavior semantics multi-scale feature energy level significance descriptors into a mask module based on a gating function to obtain a sequence of mask-gated probabilized behavior semantics multi-scale feature energy level significance descriptors. In a specific example, first, the negative number of each behavior semantics multi-scale feature energy level significance descriptor in the sequence of the behavior semantics multi-scale feature energy level significance descriptors is used as an exponential power to calculate the exponential function value with the natural constant e as the base to obtain a sequence of behavior semantics multi-scale feature energy level class support significance descriptors; then, the positional sum between the sequence of the behavior semantics multi-scale feature energy level class support significance descriptors and the constant 1 is calculated to obtain a sequence of linearly modulated behavior semantics multi-scale feature energy level class support significance descriptors; the reciprocal of each linearly modulated behavior semantics multi-scale feature energy level class support significance descriptor in the sequence of linearly modulated behavior semantics multi-scale feature energy level class support significance descriptors is calculated to obtain a sequence of behavior semantics multi-scale gated probabilized feature energy level significance descriptors. column; further, each behavior semantics multi-scale gated probabilistic feature energy level significance descriptor in the sequence of the behavior semantics multi-scale gated probabilistic feature energy level significance descriptors is masked to obtain the sequence of the masked gated probabilistic behavior semantics multi-scale feature energy level significance descriptors; wherein, each behavior semantics multi-scale gated probabilistic feature energy level significance descriptor in the sequence of the behavior semantics multi-scale gated probabilistic feature energy level significance descriptors is masked to obtain the sequence of the masked gated probabilistic behavior semantics multi-scale feature energy level significance descriptors, including: in response to the behavior semantics multi-scale gated probabilistic feature energy level significance descriptor being greater than a predetermined threshold, taking the original value of the behavior semantics multi-scale gated probabilistic feature energy level significance descriptor, otherwise setting it to 0.

[0031] Specifically, the behavior semantic temporal aggregate representation unit 354 is used to calculate the position-weighted sum of the time queue of the behavior semantic multi-scale encoding feature vector using the sequence of the mask-gated probabilistic behavior semantic multi-scale feature energy level significance descriptors as a sequence of weights to obtain the behavior semantic temporal aggregate representation vector. It should be understood that the position-weighted sum can fuse the temporal information of the mask-gated probabilistic behavior semantic multi-scale feature energy level significance and the behavior semantic multi-scale encoding feature vector.

[0032] In summary, in the above embodiment, the time queue of the behavior semantic multi-scale encoding feature vector is input into the feature aggregation network based on feature energy local significance gating to obtain the behavior semantic temporal aggregation representation vector, including: inputting the time queue of the behavior semantic multi-scale encoding feature vector into the feature aggregation network based on feature energy local significance gating to process it with the following feature aggregation formula to obtain the behavior semantic temporal aggregation representation vector; wherein the feature aggregation formula is: ;in, is the temporal sequence of the multi-scale encoding feature vectors of the behavior semantics, is the first in the time queue of the multi-scale encoding feature vector of the behavior semantics Behavior semantic multi-scale encoding feature vector, and are the first and Behavior semantic multi-scale encoding feature vector, For the The mean of the multi-scale encoding feature vectors of behavioral semantics, For the said The variance of the multi-scale encoding feature vector of the behavior semantics, is a hyperparameter, Indicates the The maximum value of the multi-scale encoding feature vector of behavioral semantics, Indicates the The minimum value of the multi-scale encoding feature vector of behavioral semantics, For the said The characteristic energy level coefficients of the behavior semantic multi-scale encoding feature vector, Indicated by the characteristic energy level coefficient The number of characteristic energy coefficients in the local neighborhood centered at For the The characteristic energy level coefficients of the behavior semantic multi-scale encoding feature vector, For the said The feature energy level saliency descriptor of the multi-scale encoding feature vector of behavioral semantics, For the said The gated probabilistic feature energy level saliency descriptor of the multi-scale encoding feature vector of behavioral semantics, For masking operation, is the predetermined threshold, For the said The mask-gated probabilistic feature energy-level saliency descriptor of the multi-scale encoding feature vector of behavioral semantics, For the said The saliency-enhanced behavior semantics multi-scale encoding feature vector corresponding to the behavior semantics multi-scale encoding feature vector, The number of feature vectors in the time queue of the multi-scale encoding feature vector for saliency enhancement behavior semantics is A vector representing the behavior semantic temporal aggregation.

[0033] It is worth mentioning that in other specific examples of the present application, the time queue of the behavioral semantic multi-scale encoding feature vector can also be input into a feature aggregation network based on feature energy local significance gating in other ways to obtain a behavioral semantic temporal aggregation representation vector, for example: input the time queue of the behavioral semantic multi-scale encoding feature vector; calculate the energy of each feature vector in the time queue; based on the feature energy, calculate a local significance gate to control the contribution of each feature vector to the aggregation result; multiply the local significance gate by the feature vector, and then sum them to obtain the behavioral semantic temporal aggregation representation vector.

[0034] In particular, the intervention plan generation module 360 ​​is used to determine the intervention plan generation result based on the behavior semantic temporal aggregation representation vector, and the intervention plan generation result is the type label of the intervention plan. In a specific example of the present application, the behavior semantic temporal aggregation representation vector is input into the classifier-based intervention plan generation module to obtain the intervention plan generation result, and the intervention plan generation result is the type label of the intervention plan. That is, the multi-scale semantic temporal aggregation features of the child's behavior are used for classification processing to generate an intervention plan. In this way, it is possible to provide personalized intervention plans more intelligently based on the multi-dimensional behavior temporal aggregation semantics of the child object, including intervention goals, intervention methods and steps, thereby improving the pertinence of behavioral intervention for children with neurodevelopmental disorders, so as to better meet the different needs of children and promote their all-round development.

[0035] In a preferred example, when the time queue of the behavior semantic encoding feature vector and the time queue of the behavior semantic implicit encoding feature vector respectively express the local time domain semantic encoding features and the global time domain short-range-long-range bidirectional contextual semantic temporal association features of the behavior data, the time queue of the behavior semantic multi-scale encoding feature vector can perform distinguishable attention fusion of multi-dimensional temporal features based on the local time domain granularity, and perform distribution aggregation based on the local time domain feature energy significance under the global time domain by a feature aggregation network based on feature energy local significance gating. However, this will also cause the obtained behavior semantic temporal aggregation representation vector to have a distribution regression pattern offset caused by the difference in energy significance aggregation, that is, it will cause the behavior semantic temporal aggregation representation vector to be input into the classifier-based intervention plan generation module to obtain an overflow of the intervention plan generation result, affecting its accuracy.

[0036] Based on this, the applicant of the present application optimizes the behavior semantic temporal aggregation representation vector when inputting the behavior semantic temporal aggregation representation vector into the intervention plan generation module based on the classifier to obtain the intervention plan generation result. The optimization process specifically includes the following steps: determining the behavior semantic temporal aggregation representation vector input into the intervention plan generation module based on the classifier and obtaining the same The intervention type labels correspond to label probability values, and calculate the The square root of the sum of the squares of the label probability values ​​is obtained to obtain the type distribution probability value; the behavior semantic temporal aggregation representation vector is passed through a probabilistic activation function to obtain a probabilistic behavior semantic temporal aggregation representation vector; the probabilistic behavior semantic temporal aggregation representation vector is subtracted from the unit feature vector, and the feature vector resulting from the subtraction is used as the base to calculate a power function with the type distribution probability value as the exponent; the power function vector is multiplied by the probabilistic behavior semantic temporal aggregation representation vector to obtain a behavior semantic temporal aggregation intermediate feature vector; the absolute value of each eigenvalue of the probabilistic behavior semantic temporal aggregation representation vector is used as the base to calculate a power function with the inverse of the type distribution probability value as the exponent, and all eigenvalues ​​are summed to obtain a behavior semantic temporal aggregation bias eigenvalue; the behavior semantic temporal aggregation bias eigenvalue is multiplied by a weight as a hyperparameter, and then multiplied by the behavior semantic temporal aggregation intermediate feature vector to obtain an optimized behavior semantic temporal aggregation representation vector.

[0037] Wherein, the probabilistic behavior semantic temporal aggregation representation vector is recorded as , then the optimized behavior semantic temporal aggregation representation vector It is expressed as: ,in, ,in, is the probabilistic behavior semantic temporal aggregation representation vector, is the unit eigenvector, and , is the eigenvalue of the probabilistic motion pattern temporal semantic feature vector, is a power function that calculates the eigenvalues ​​of each position in the vector with the probability value of the type distribution as an exponent, represents the type distribution probability value, The first eigenvalues, Indicates point addition, Indicates point reduction, represents dot product, represents the weight as a hyperparameter, is the optimized behavioral semantic temporal aggregation representation vector.

[0038] That is, for the behavior semantic temporal aggregation representation vector, based on the convergence limit of the power series probability distribution of the Cauchy-Hadamard form, the feature set of the behavior semantic temporal aggregation representation vector is represented by a power series based on the convergence probability, and a set convergence bounded bias based on each eigenvalue of the behavior semantic temporal aggregation representation vector as a convergence radius representation is added, so as to avoid the behavior semantic temporal aggregation representation vector in the regression process of the intervention plan generation module based on the classifier under the unified probability convergence limit standard. The behavior semantic temporal aggregation representation vector is caused by the gradient propagation of the model parameters based on the convergence probability. The distribution mode deviation causes the invalid overflow of the classification result, thereby improving the accuracy of the intervention plan generation result obtained by the intervention plan generation module based on the classifier. In this way, it is possible to provide personalized intervention plans more intelligently based on the multi-dimensional behavior temporal aggregation semantics of the child object, thereby improving the pertinence of the behavioral intervention for children with neurodevelopmental disorders, so as to better meet the needs of different children and promote their all-round development.

[0039] Figure 4 FIG. 4 is a block diagram of another embodiment of a neurodevelopmental disorder behavior intervention system based on a large language model according to an embodiment of the present application. Figure 4 As shown, further, in another embodiment of the present application, the neurodevelopmental disorder behavior intervention system based on the large language model also includes: a data collection module 370 for the environment where the patient is located and an interactive behavior analysis module 380. Among them, the data collection module 370 for the environment where the patient is located is used to obtain the time series of the external environment data of the patient, such as the family environment, school environment, etc., which provide background information of the patient's behavior. The interactive behavior analysis module 380 is used to collect and analyze the time series of interactive behavior data of other people (such as parents, teachers, and peers) who interact with the patient. This module evaluates the impact of these interactions on the patient's behavior.

[0040] Furthermore, in another embodiment, the child behavior semantic encoding module 320 is used to perform semantic encoding based on a large language model on each behavior data in the time queue of the behavior data, each external environment data in the time series of the external environment data, and each interactive behavior data in the time series of the interactive behavior data to obtain a time queue of behavior semantic encoding feature vectors, a time queue of external environment semantic encoding feature vectors, and a time queue of interactive behavior semantic encoding feature vectors, and to perform weighted fusion on the time queue of the behavior semantic encoding feature vectors, the time queue of the external environment semantic encoding feature vectors, and the time queue of the interactive behavior semantic encoding feature vectors to obtain a time queue of behavior semantic encoding feature vectors containing behavior information, environment information, and interactive information. Then, the behavior semantic temporal implicit feature extraction module 330, the behavior semantic multi-scale feature attention fusion module 340, and the behavior semantic temporal aggregation module 350 process the time queue of behavior semantic encoding feature vectors containing behavior information, environment information, and interactive information with reference to the encoding method of the above embodiment to obtain a behavior semantic temporal aggregation representation vector.

[0041] Further, in another embodiment, the intervention plan generation module 360 ​​is used to determine the intervention plan generation result based on the behavior semantic temporal aggregation representation vector, and the intervention plan generation result is the type label of the intervention plan. It should be understood that in another embodiment, the intervention plan generation result is obtained by dividing the time queue of the behavior semantic encoding feature vector containing behavior information, environmental information and interactive information, that is, the intervention plan generation module generates a personalized intervention plan based on the analysis results of the child's behavior data, environmental data and interactive behavior data.

[0042] As described above, the neurodevelopmental disorder behavior intervention system 300 based on a large language model according to an embodiment of the present application can be implemented in various wireless terminals, such as a server with a neurodevelopmental disorder behavior intervention algorithm based on a large language model. In a possible implementation, the neurodevelopmental disorder behavior intervention system 300 based on a large language model according to an embodiment of the present application can be integrated into a wireless terminal as a software module and / or a hardware module. For example, the neurodevelopmental disorder behavior intervention system 300 based on a large language model can be a software module in the operating system of the wireless terminal, or can be an application developed for the wireless terminal; of course, the neurodevelopmental disorder behavior intervention system 300 based on a large language model can also be one of the many hardware modules of the wireless terminal.

[0043] Alternatively, in another example, the large language model-based behavioral intervention system for neurodevelopmental disorders 300 and the wireless terminal may also be separate devices, and the large language model-based behavioral intervention system for neurodevelopmental disorders 300 may be connected to the wireless terminal via a wired and / or wireless network and transmit interactive information in accordance with an agreed data format.

[0044] In one example, the neurodevelopmental disorder behavioral intervention system based on a large language model of the present application mainly achieves comprehensive intervention effects through the following key modules, improves the behavioral and emotional state of children, and provides important support for professional intervention personnel: 1. Evaluation and program formulation: (1) Preliminary assessment: Collect preliminary behavioral data of families and children through online questionnaires and interviews, videos, pictures, etc., covering aspects such as language, social interaction and behavioral expression patterns.

[0045] (2) Assessment support: Use large language models to analyze collected data to assist professionals in making accurate judgments and identifying the specific needs and problem areas of children and their families.

[0046] (3) Personalized plan formulation: Based on the judgment results, the system generates a personalized intervention plan, which is manually reviewed by professionals. The plan will cover specific intervention goals, intervention methods and steps.

[0047] 2. Personalized intervention plan: (1) Dynamic adjustment: The system dynamically adjusts the intervention plan based on the feedback and behavioral data of the intervention practitioners to ensure continuous optimization of the intervention plan.

[0048] (2) Multidimensional intervention: Combining multiple disciplinary intervention techniques such as behavioral management, cognitive training, social skills, and emotional management to comprehensively improve the capabilities of intervention practitioners and children.

[0049] 3. Natural Language Interaction: (1) Virtual Assistant: Provides 24 / 7 intelligent support, answers intervention practitioners’ questions through natural language interaction, and provides immediate help and emotional support.

[0050] (2) Simulation training: Simulating real-life problem scenarios to help intervention practitioners practice intervention skills and enhance their practical application-level intervention capabilities.

[0051] 4. Education and Training: (1) Online courses: Provide modular online courses covering behavior management strategies, cognitive training, emotion regulation, etc., to help parents master effective intervention methods.

[0052] (2) Interactive learning: Enhance adult learning outcomes and ensure practical application of knowledge through interactive exercises and quizzes.

[0053] 5. Data analysis and tracking: (1) Behavior records: The intervention practitioners record the behavioral data through mobile applications and generate detailed behavioral data reports.

[0054] (2) Progress tracking: Regularly evaluate the progress of the child, identify behavioral patterns and changing trends through data analysis, and dynamically adjust the intervention plan.

[0055] (3) Personalized recommendations: Based on data analysis results, the system provides personalized intervention recommendations to help parents and professionals optimize intervention strategies.

[0056] 6. Support professional intervention personnel: (1) Needs assessment: By analyzing large amounts of data, we can provide intervention assistance to help professionals more accurately assess the situation of children and their families.

[0057] (2) Intervention recommendations: Provide personalized intervention recommendations based on the child’s behavioral data and intervention progress to help professionals develop and adjust intervention plans.

[0058] (3) Resource integration: Integrate educational resources and training materials to facilitate professionals to obtain the latest research results and intervention methods.

[0059] The embodiments of the present disclosure have been described above, and the above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and changes will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The selection of terms used herein is intended to best explain the principles of the embodiments, practical applications, or improvements to the technology in the market, or to enable other persons of ordinary skill in the art to understand the embodiments disclosed herein.

Claims

1. A neurodevelopmental disorder behavioral intervention system based on a large language model, characterized in that: include: A patient subject behavior data collection module, used to obtain a time queue of the patient subject's behavior data; A child behavior semantic encoding module, used to input each behavior data in the time queue of the behavior data into a behavior semantic encoder based on a large language model to obtain a time queue of behavior semantic encoding feature vectors; a behavior semantic temporal implicit feature extraction module, used to perform temporal encoding based on behavior implicit semantics on the time queue of the behavior semantic encoding feature vectors to obtain a time queue of behavior semantic implicit encoding feature vectors; a behavior semantic multi-scale feature attention fusion module, used to input each group of corresponding behavior semantic encoding feature vectors and behavior semantic implicit encoding feature vectors in the time queue of the behavior semantic encoding feature vectors and the time queue of the behavior semantic implicit encoding feature vectors into a feature value granularity association mask distinguishable attention fusion module to obtain a time queue of behavior semantic multi-scale encoding feature vectors; a behavior semantic temporal aggregation module, used to input the time queue of the behavior semantic multi-scale encoding feature vectors into a feature aggregation network based on feature energy local significance gating to obtain a behavior semantic temporal aggregation representation vector; an intervention plan generation module, used to determine an intervention plan generation result based on the behavior semantic temporal aggregation representation vector, and the intervention plan generation result is a type label of the intervention plan; Among them, the behavior semantics multi-scale feature attention fusion module includes: a global feature value association calculation unit, which is used to calculate the global feature value association matrix between a group of corresponding behavior semantics encoding feature vectors and behavior semantics implicit encoding feature vectors in the time queue of the behavior semantics encoding feature vector and the time queue of the behavior semantics implicit encoding feature vector to obtain the behavior semantics multi-scale global feature value association matrix; a gating calculation unit, which is used to input the behavior semantics multi-scale global feature value association matrix into a learnable gating function to obtain the behavior semantics multi-scale correlation distinguishable weight matrix; a distinguishable weight fusion mask unit, which is used to use the behavior semantics multi-scale correlation distinguishable weight matrix as a fusion mask matrix, and respectively calculate the matrix product between the behavior semantics encoding feature vector and the behavior semantics implicit encoding feature vector and the behavior semantics multi-scale correlation distinguishable weight matrix to obtain the distinguishing and strengthening behavior semantics encoding feature vector and the distinguishing and strengthening behavior semantics implicit encoding feature vector; a behavior semantics multi-scale feature fusion unit, which is used to fuse the distinguishing and strengthening behavior semantics encoding feature vector and the distinguishing and strengthening behavior semantics implicit encoding feature vector to obtain the behavior semantics multi-scale encoding feature vector; Wherein, the gated calculation unit is used to: use the negative number of each position eigenvalue in the behavior semantics multi-scale global eigenvalue association matrix as the exponential power, calculate the natural exponential function value with the natural constant e as the base to obtain the behavior semantics multi-scale global eigenvalue association class support matrix; calculate the reciprocal of the sum of each position eigenvalue in the behavior semantics multi-scale global eigenvalue association class support matrix and the constant 1 to obtain the behavior semantics multi-scale global eigenvalue association weight matrix; calculate the position point multiplication between the behavior semantics multi-scale global eigenvalue association weight matrix and the behavior semantics multi-scale global eigenvalue association matrix to obtain the behavior semantics multi-scale correlation distinguishable weight matrix; The behavior semantic temporal aggregation module includes: a characteristic energy level coefficient calculation unit, which is used to determine the characteristic energy level coefficients of each behavior semantic multi-scale coding feature vector based on the maximum value, minimum value, mean value and variance of each behavior semantic multi-scale coding feature vector in the time queue of the behavior semantic multi-scale coding feature vector to obtain a sequence of behavior semantic multi-scale characteristic energy level coefficients; a local neighborhood significance descriptor calculation unit, which is used to determine the scale of the behavior semantic multi-scale local neighborhood, calculate the local neighborhood significance descriptor of each behavior semantic multi-scale characteristic energy level coefficient in the behavior semantic multi-scale characteristic energy level coefficient. descriptors to obtain a sequence of behavioral semantic multi-scale feature energy level significance descriptors; a mask-gated probabilistic unit, used to input the sequence of behavioral semantic multi-scale feature energy level significance descriptors into a mask module based on a gating function to obtain a sequence of mask-gated probabilistic behavioral semantic multi-scale feature energy level significance descriptors; a behavioral semantic temporal aggregation representation unit, used to use the sequence of mask-gated probabilistic behavioral semantic multi-scale feature energy level significance descriptors as a sequence of weights, calculate the position-weighted sum of the time queue of the behavioral semantic multi-scale encoding feature vector to obtain the behavioral semantic temporal aggregation representation vector; Wherein, the characteristic energy level coefficient calculation unit is used to: calculate the maximum value, minimum value, average value and variance of the behavior semantics multi-scale encoding feature vector to obtain the behavior semantics multi-scale maximum value, behavior semantics multi-scale minimum value, behavior semantics multi-scale average value and behavior semantics multi-scale variance; calculate the sum of the behavior semantics multi-scale variance and the hyperparameter to obtain the first behavior semantics multi-scale local energy statistical factor; calculate the difference between the behavior semantics multi-scale maximum value and the behavior semantics multi-scale average value to obtain the behavior semantics multi-scale maximum deviation value; calculate the difference between the behavior semantics multi-scale minimum value and the behavior semantics multi-scale average value to obtain the behavior semantics multi-scale minimum deviation value; calculate the sum of the squares of the behavior semantics multi-scale maximum deviation value and the behavior semantics multi-scale minimum deviation value and add it to the hyperparameter to obtain the second behavior semantics multi-scale local energy statistical factor; calculate the division between the first behavior semantics multi-scale local energy statistical factor and the second behavior semantics multi-scale local energy statistical factor to obtain the behavior semantics multi-scale feature energy level coefficient; The local neighborhood significant descriptor calculation unit is used to: use the scale of the behavior semantics multi-scale local neighborhood as the mean calculation range, and respectively calculate the local neighborhood mean of each behavior semantics multi-scale feature energy level coefficient to obtain a sequence of the behavior semantics multi-scale feature energy level significant descriptors; Among them, the mask gated probabilization unit is used to: use the negative number of each behavior semantics multi-scale feature energy level significance descriptor in the sequence of the behavior semantics multi-scale feature energy level significance descriptor as the exponential power, calculate the exponential function value with the natural constant e as the base to obtain a sequence of behavior semantics multi-scale feature energy level class support significance descriptors; calculate the positional sum between the sequence of behavior semantics multi-scale feature energy level class support significance descriptors and the constant 1 to obtain a sequence of linear modulation behavior semantics multi-scale feature energy level class support significance descriptors; calculate the reciprocal of each linear modulation behavior semantics multi-scale feature energy level class support significance descriptor in the sequence of linear modulation behavior semantics multi-scale feature energy level class support significance descriptors to obtain the behavior semantics multi-scale gated probabilization feature energy level significance descriptor. descriptors; performing masking processing on each behavior semantics multi-scale gated probabilistic feature energy level significance descriptor in the sequence of behavior semantics multi-scale gated probabilistic feature energy level significance descriptors to obtain the sequence of masked gated probabilistic behavior semantics multi-scale feature energy level significance descriptors; wherein, performing masking processing on each behavior semantics multi-scale gated probabilistic feature energy level significance descriptor in the sequence of behavior semantics multi-scale gated probabilistic feature energy level significance descriptors to obtain the sequence of masked gated probabilistic behavior semantics multi-scale feature energy level significance descriptors comprises: in response to the behavior semantics multi-scale gated probabilistic feature energy level significance descriptor being greater than a predetermined threshold, taking the original value of the behavior semantics multi-scale gated probabilistic feature energy level significance descriptor, otherwise setting it to 0.

2. The neurodevelopmental disorder behavior intervention system based on a large language model according to claim 1, characterized in that: The behavior semantic temporal implicit feature extraction module is used to: input the time queue of the behavior semantic coding feature vector into a behavior semantic temporal encoder based on a Bi-LSTM model to obtain the time queue of the behavior semantic implicit coding feature vector.

3. The neurodevelopmental disorder behavior intervention system based on a large language model according to claim 2, characterized in that: The intervention plan generation module is used to: input the behavior semantic temporal aggregation representation vector into the classifier-based intervention plan generation module to obtain an intervention plan generation result, and the intervention plan generation result is a type label of the intervention plan.

Citation Information

Cited By

  • Mixed reality assessment intervention system and method for child developmental disorder

    CN120977477A

  • Intelligent intervention strategy optimization method, system and equipment based on dynamic physiological parameters

    CN121506363A