Child psychological development analysis system and method based on large language model

Abstract

The application discloses a children's psychological development analysis system and method based on a large language model, which uses natural language understanding and processing technology based on a large language model and deep learning to respectively perform semantic encoding and time sequence feature aggregation on language communication data and behavior habit data of the children to be analyzed, so as to automatically evaluate whether the children to be analyzed are positive development according to a multi-modal fusion representation feature obtained by significant fusion between time sequence semantic aggregation features of the language communication data and the behavior habit data. In this way, by using the large language model and the deep learning technology, the system can automatically process and analyze a large amount of data, and reduce the subjectivity and workload of manual interpretation. Meanwhile, important dynamic changes and trends of language and behavior in the psychological development process of children can be captured, a more detailed basis is provided for children's psychological development analysis and psychological counseling, and the intelligent degree of children's psychological counseling is improved.

Description

Technical Field

[0001] This application relates to the field of intelligent analysis, and more specifically, to a system and method for analyzing children's psychological development based on a large language model. Background Technology

[0002] In today's society, with the accelerating pace of life and increasing competitive pressure, children face ever-growing psychological challenges. They not only have to cope with academic pressure but also face social influences from peers, the media, and other sources. These factors can significantly impact children's self-awareness, emotional management, and social skills. Therefore, understanding the patterns of child psychological development helps in the timely detection and prevention of problems, providing appropriate guidance and assistance to children, and promoting their overall healthy development.

[0003] However, existing methods for assessing children's psychological development often require subjective judgment by professionals based on standardized tests or observations. This not only increases the subjectivity of the assessment but also limits the general applicability of the results. In other words, due to the lack of unified quantitative standards, significant differences in interpretation may exist between different assessors, reducing the comparability of assessment results. Furthermore, the limitations of manual interpretation when processing large-scale data prevent the efficient identification and analysis of subtle but crucial complex relationships between data points. This can lead to the overlooking of important psychological patterns and behavioral trends, thus limiting the breadth and depth of our understanding of children's psychological development and affecting the timeliness and effectiveness of interventions.

[0004] Therefore, an optimized system for analyzing children's psychological development is desired. Summary of the Invention

[0005] To address the aforementioned technical problems, this application is proposed. Embodiments of this application provide a system and method for analyzing children's psychological development based on a large language model. This system employs natural language understanding and processing techniques based on a large language model and deep learning to perform semantic encoding and temporal feature aggregation on the language communication data and behavioral habit data of the child to be analyzed. Based on the multimodal fusion representation features obtained by significantly fusing the temporal semantic aggregation features of the language communication data and the behavioral habit data, the system automatically assesses whether the child is developing positively. Thus, by utilizing a large language model and deep learning technology, the system can automatically process and analyze large amounts of data, reducing the subjectivity and workload of manual interpretation. Simultaneously, it can capture important dynamic changes and trends in language and behavior during children's psychological development, providing more detailed evidence for children's psychological development analysis and psychological counseling, and improving the intelligence level of children's psychological counseling.

[0006] According to one aspect of this application, a child psychological development analysis system based on a large language model is provided, comprising:

[0007] The language communication behavior habit data acquisition module is used to acquire the time queue of language communication data and the time queue of behavior habit data of the children to be analyzed.

[0008] The language communication data semantic understanding module is used to perform semantic understanding on each language communication data in the time queue of the language communication data using a large model-based language communication semantic encoder to obtain the time queue of language communication data semantic encoding feature vectors.

[0009] The behavioral habit data semantic understanding module is used to perform semantic understanding on each behavioral habit data in the time queue of the behavioral habit data using a large model-based behavioral habit semantic encoder to obtain the time queue of behavioral habit data semantic encoded feature vectors.

[0010] The language behavior pattern temporal semantic aggregation module is used to input the time queue of the semantic encoding feature vector of the language communication data and the time queue of the semantic encoding feature vector of the behavior habit data into the feature aggregation network guided by the saliency decay of node features to obtain the language pattern temporal semantic aggregation representation vector and the behavior pattern temporal semantic aggregation representation vector.

[0011] The language-behavior multimodal fusion module is used to input the language pattern temporal semantic aggregation representation vector and the behavior pattern temporal semantic aggregation representation vector into a saliency fusion network based on feature principal component query matching to obtain a language-behavior multimodal saliency fusion representation vector;

[0012] The psychological development analysis result generation module is used to obtain psychological development analysis results based on the language-behavior multimodal saliency fusion representation vector, and the psychological development analysis results are used to indicate whether the development is positive.

[0013] According to another aspect of this application, a method for analyzing children's psychological development based on a large language model is provided, which includes:

[0014] Time queues for acquiring language communication data and behavioral habit data of the children to be analyzed;

[0015] A large-model-based language communication semantic encoder is used to perform semantic understanding on each language communication data in the time queue of the language communication data to obtain the time queue of language communication data semantic encoded feature vectors.

[0016] A large-model-based behavioral habit semantic encoder is used to perform semantic understanding on each behavioral habit data in the time queue of the behavioral habit data to obtain the time queue of behavioral habit data semantic encoded feature vectors.

[0017] The time queues of the semantic encoding feature vectors of the language communication data and the time queues of the semantic encoding feature vectors of the behavioral habit data are input into a feature aggregation network guided by the saliency decay of node features to obtain the temporal semantic aggregation representation vectors of the language pattern and the temporal semantic aggregation representation vectors of the behavioral pattern.

[0018] The language pattern temporal semantic aggregation representation vector and the behavior pattern temporal semantic aggregation representation vector are input into a saliency fusion network based on feature principal component query matching to obtain a language-behavior multimodal saliency fusion representation vector;

[0019] The language-behavior multimodal saliency fusion representation vector is used to obtain the psychological development analysis results, which are used to indicate whether the development is positive.

[0020] Compared with existing technologies, this application provides a child psychological development analysis system and method based on a large language model. It employs natural language understanding and processing techniques based on a large language model and deep learning to semantically encode and aggregate temporal features of the language communication data and behavioral habit data of the child to be analyzed. Based on the significant fusion of the temporal semantic aggregation features of the language communication data and the behavioral habit data, it automatically assesses whether the child is developing positively. Thus, by utilizing a large language model and deep learning technology, the system can automatically process and analyze large amounts of data, reducing the subjectivity and workload of manual interpretation. Simultaneously, it can capture important dynamic changes and trends in language and behavior during the child's psychological development, providing more detailed evidence for child psychological development analysis and counseling, and enhancing the intelligence level of child psychological counseling. Attached Figure Description

[0021] The above and other objects, features, and advantages of this application will become more apparent from the more detailed description of the embodiments of this application in conjunction with the accompanying drawings. The drawings are provided to further illustrate the embodiments of this application and form part of the specification. They are used together with the embodiments of this application to explain this application and do not constitute a limitation thereof. In the drawings, the same reference numerals generally represent the same components or steps.

[0022] Figure 1 This is a block diagram of a child psychological development analysis system based on a large language model according to an embodiment of this application;

[0023] Figure 2 This is a schematic diagram of data flow in a child psychological development analysis system based on a large language model according to an embodiment of this application;

[0024] Figure 3 This is a flowchart of a child psychological development analysis method based on a large language model according to an embodiment of this application. Detailed Implementation

[0025] Hereinafter, exemplary embodiments according to this application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments of this application. It should be understood that this application is not limited to the exemplary embodiments described herein.

[0026] As indicated in this application and claims, unless the context clearly indicates otherwise, the words "a," "an," "an," and / or "the" are not specifically singular and may include plural forms. Generally speaking, the terms "comprising" and "including" only indicate the inclusion of explicitly identified steps and elements, which do not constitute an exclusive list, and the method or apparatus may also include other steps or elements.

[0027] While this application makes various references to certain modules of the systems according to embodiments of this application, any number of different modules can be used and run on user terminals and / or servers. The modules described are merely illustrative, and different aspects of the systems and methods may use different modules.

[0028] Flowcharts are used in this application to illustrate the operations performed by the system according to embodiments of this application. It should be understood that the preceding or following operations are not necessarily performed in exact order. Instead, various steps can be processed in reverse order or simultaneously as needed. Furthermore, other operations can be added to these processes, or one or more steps can be removed from them.

[0029] Hereinafter, exemplary embodiments according to this application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments of this application. It should be understood that this application is not limited to the exemplary embodiments described herein.

[0030] Existing methods for assessing children's psychological development typically require subjective judgment by professionals based on standardized tests or observations. This not only increases the subjectivity of the assessment but also limits the general applicability of the results. Specifically, due to the lack of unified quantitative standards, significant differences in interpretation may exist between different assessors, reducing the comparability of assessment results. Furthermore, the limitations of manual interpretation when processing large-scale data prevent the efficient identification and analysis of subtle but crucial complex relationships between data points. This can lead to the overlooking of important psychological patterns and behavioral trends, thus limiting the breadth and depth of our understanding of children's psychological development and affecting the timeliness and effectiveness of interventions. Therefore, an optimized system for analyzing children's psychological development is desired.

[0031] The technical solution of this application proposes a children's psychological development analysis system based on a large language model. Figure 1 This is a block diagram of a child psychological development analysis system based on a large language model, according to an embodiment of this application. Figure 2 This is a schematic diagram of data flow in a child psychological development analysis system based on a large language model, according to an embodiment of this application. Figure 1 and Figure 2 As shown, the child psychological development analysis system 300 based on a large language model according to an embodiment of this application includes: a language communication behavior habit data acquisition module 310, used to acquire a time queue of language communication data and a time queue of behavior habit data of the child to be analyzed; a language communication data semantic understanding module 320, used to perform semantic understanding on each language communication data in the time queue of the language communication data using a large-model-based language communication semantic encoder to obtain a time queue of language communication data semantic encoding feature vectors; a behavior habit data semantic understanding module 330, used to perform semantic understanding on each behavior habit data in the time queue of the behavior habit data using a large-model-based behavior habit semantic encoder to obtain a time queue of behavior habit data semantic encoding feature vectors; and a language behavior pattern time queue. The sequential semantic aggregation module 340 is used to input the time queue of the semantic encoding feature vector of the language communication data and the time queue of the semantic encoding feature vector of the behavior habit data into a feature aggregation network guided by the saliency decay of node features to obtain the temporal semantic aggregation representation vector of the language pattern and the temporal semantic aggregation representation vector of the behavior pattern; the language-behavior multimodal fusion module 350 is used to input the temporal semantic aggregation representation vector of the language pattern and the temporal semantic aggregation representation vector of the behavior pattern into a saliency fusion network based on feature principal component query matching to obtain the language-behavior multimodal saliency fusion representation vector; the psychological development analysis result generation module 360 ​​is used to obtain the psychological development analysis result based on the language-behavior multimodal saliency fusion representation vector, wherein the psychological development analysis result is used to indicate whether the development is positive.

[0032] Specifically, the language communication behavior habit data acquisition module 310 and the language communication data semantic understanding module 320 are used to acquire the time queue of language communication data and the time queue of behavior habit data of the children to be analyzed; and to use a large-model-based language communication semantic encoder to perform semantic understanding on each language communication data in the time queue of the language communication data to obtain the time queue of language communication data semantic encoding feature vectors. Considering that each language communication data in the time queue of the language communication data has semantic relationships between contexts, and that the large language model, after pre-training on massive corpora, has very powerful natural language understanding capabilities, it can accurately capture the semantic information in the language communication data and extract semantically rich feature vectors. Based on this, in the technical solution of this application, a large-model-based language communication semantic encoder is used to perform semantic understanding on each language communication data in the time queue of the language communication data to obtain the time queue of language communication data semantic encoding feature vectors.

[0033] Specifically, the behavioral habit data semantic understanding module 330 is used to perform semantic understanding on each behavioral habit data in the time queue of the behavioral habit data using a large-model-based behavioral habit semantic encoder to obtain a time queue of behavioral habit data semantic encoded feature vectors. Similarly, considering that each behavioral habit data in the time queue of the behavioral habit data also has semantic interrelationships with the context of behavioral information, the technical solution of this application uses a large-model-based behavioral habit semantic encoder to perform semantic understanding on each behavioral habit data in the time queue of the behavioral habit data to obtain a time queue of behavioral habit data semantic encoded feature vectors.

[0034] Specifically, the language behavior pattern temporal semantic aggregation module 340 is used to input the time queues of the semantic encoding feature vectors of the language communication data and the semantic encoding feature vectors of the behavior habit data into a feature aggregation network guided by node feature saliency decay to obtain the language pattern temporal semantic aggregation representation vector and the behavior pattern temporal semantic aggregation representation vector. Considering that both language communication and behavior habits are dynamic processes that change over time, and that different time points have different importance and saliency, in order to effectively capture these temporal dynamic features and automatically identify key time points in language communication and behavior habits to enhance the semantic representation capability of the entire time dimension, the technical solution of this application uses a feature aggregation network guided by node feature saliency decay to perform saliency temporal feature aggregation on the time queues of the semantic encoding feature vectors of the language communication data and the behavior habit data to obtain the language pattern temporal semantic aggregation representation vector and the behavior pattern temporal semantic aggregation representation vector. Specifically, the feature aggregation network guided by node feature saliency decay is a feature extraction and aggregation model based on feature saliency analysis. Here, we will take the time queue of the semantic encoding feature vector of the language communication data as an example for detailed explanation.

[0035] In the embodiments of this application, the time queues of the semantic encoding feature vectors of the language communication data and the time queues of the semantic encoding feature vectors of the behavioral habit data are input into a feature aggregation network guided by the saliency decay of node features to obtain the temporal semantic aggregation representation vectors of language patterns and the temporal semantic aggregation representation vectors of behavioral patterns. This includes: firstly, calculating the feature saliency descriptor factor of each semantic encoding feature vector in the time queue of the language communication data, wherein the feature saliency descriptor factor is related to the mean and variance of each semantic encoding feature vector; that is, the mean and variance of each feature in the language communication data reflect the average importance and fluctuation of that feature throughout the entire time series, and are important statistics describing feature saliency. Furthermore, by calculating the relative mean and variance of each feature, the importance of that feature relative to other features can be obtained, which is beneficial for highlighting key features. Next, based on the distance span between each language communication data semantic encoding feature vector in the time queue and the current language communication data semantic encoding feature vector, a feature saliency decay factor is constructed for each language communication data semantic encoding feature vector. Here, the feature saliency decay factor is constructed to represent that as time changes, the influence of language patterns that are far removed from the current time is smaller, thereby highlighting the importance of recent features and assigning them relatively higher weights. Subsequently, based on the feature saliency decay factors of each language communication data semantic encoding feature vector, the feature saliency descriptor factors of each language communication data semantic encoding feature vector are modulated to obtain a sequence of language communication data semantic feature saliency decay descriptor factors. Then, the sequence of language communication data semantic feature saliency decay descriptor factors is input into a gating mask module to obtain a sequence of language communication data semantic feature saliency decay weight factors. Finally, using the sequence of language communication data semantic feature saliency decay weight factors as the weight sequence, the weighted sum of the time queue of the language communication data semantic encoding feature vectors is calculated to obtain the temporal semantic aggregation representation vector of the language pattern.

[0036] The process of calculating the feature saliency descriptor of each language communication data semantic coding feature vector in the time queue includes: calculating the mean and variance of the language communication data semantic coding feature vector to obtain the mean and variance of the language communication data semantic features; calculating the positional difference between the language communication data semantic coding feature vector and the mean of the language communication data semantic features to obtain the language communication data semantic difference vector; calculating the fourth power of each feature value in the language communication data semantic difference vector to obtain the language communication data semantic modulation difference vector; calculating the expected value of the language communication data semantic modulation difference vector to obtain the language communication data expected value; and dividing the language communication data expected value by the square of the language communication data semantic feature variance to obtain the feature saliency descriptor corresponding to the language communication data semantic coding feature vector.

[0037] More specifically, the process of constructing the feature significance decay factor of each language communication data semantic coding feature vector based on the distance span between each language communication data semantic coding feature vector in the time queue and the current language communication data semantic coding feature vector includes: extracting the maximum value from each language communication data semantic coding feature vector to obtain a sequence of language communication data semantic feature maximum values; extracting the maximum value from the current language communication data semantic coding feature vector to obtain the current language communication data semantic coding feature maximum value; calculating the sequence of language communication data semantic feature maximum values ​​and the current language communication data semantic coding feature vector. The sequence of semantic feature offset values ​​for language communication data is obtained by subtracting the maximum values ​​of the encoded features positionally; the distance span between each semantic encoded feature vector of language communication data and the current semantic encoded feature vector of language communication data is calculated to obtain a sequence of semantic distance span values ​​for language communication data, and the square of each value in the sequence of semantic distance span values ​​for language communication data is calculated to obtain a sequence of semantic distance span modulation values ​​for language communication data; the sequence of semantic feature offset values ​​for language communication data and the sequence of semantic distance span modulation values ​​for language communication data are divided positionally to obtain the feature saliency attenuation factor of each semantic encoded feature vector of language communication data.

[0038] More specifically, the process of inputting the sequence of descriptors for the saliency decay of the semantic features of the language communication data into a gating mask module to obtain the sequence of weight factors for the saliency decay of the semantic features of the language communication data includes: performing sigmoid-based normalization on each descriptor in the sequence of descriptors for the saliency decay of the semantic features of the language communication data to obtain a set of normalized descriptors for the saliency decay of the semantic features of the language communication data; and inputting each normalized descriptor in the set of normalized descriptors for the saliency decay of the semantic features of the language communication data into a gating function for masking to obtain the sequence of weight factors for the saliency decay of the semantic features of the language communication data.

[0039] In summary, in the above embodiments, inputting the time queues of the semantic encoding feature vectors of the language communication data and the semantic encoding feature vectors of the behavioral habit data into a feature aggregation network guided by node feature saliency decay to obtain the temporal semantic aggregation representation vectors of the language pattern and the temporal semantic aggregation representation vectors of the behavioral pattern includes: inputting the time queues of the semantic encoding feature vectors of the language communication data into a feature aggregation network guided by node feature saliency decay, and processing them with the following feature aggregation formula to obtain the temporal semantic aggregation representation vectors of the language pattern; wherein, the feature aggregation formula is:

[0040] X = {x1, x2, ..., x} k ,...,x n}

[0041]

[0042] w i =sigmoid(k i )

[0043] ws i =mask(w i )

[0044]

[0045] Where X represents the time queue of the semantic encoding feature vector of the language communication data, x k x n and x i These are the k-th, n-th, and i-th semantic encoding feature vectors of the language communication data in the time queue, respectively. max(·) represents the maximum value of the extracted vector, and Count(x) represents the maximum value of the extracted vector. n →x i ) represents x n With x iThe distance span between them, τ i x is the feature saliency attenuation factor of the semantic encoding feature vector of the i-th language communication data. i (j) is the feature value at the (j)th position of the semantic encoding feature vector of the i-th language communication data, μ i and σ i 4 These are the mean and squared variance of the semantic encoding feature vector of the i-th language communication data, respectively, k i It is the semantic feature saliency decay descriptor of the i-th language communication data semantic encoding feature vector, and sigmoid(·) is the sigmoid function. i It is the i-th normalized language communication data semantic feature saliency attenuation descriptor in the set of normalized language communication data semantic feature saliency attenuation descriptors, mask(·) is the masking process, ws i θ is the i-th semantic feature saliency attenuation weight factor in the sequence of semantic feature saliency attenuation weight factors of the language communication data, θ is a predetermined threshold, and V1 is the temporal semantic aggregation representation vector of the language pattern. Similarly, the encoding method of the temporal queue of the semantic encoding feature vector of the behavioral habit data is also as shown in the above formula.

[0046] Specifically, the language-behavior multimodal fusion module 350 is used to input the language pattern temporal semantic aggregation representation vector and the behavior pattern temporal semantic aggregation representation vector into a saliency fusion network based on feature principal component query matching to obtain a language-behavior multimodal saliency fusion representation vector. Considering the semantic interrelationship between the language pattern temporal semantic aggregation representation vector and the behavior pattern temporal semantic aggregation representation vector, which includes the main component features between language and behavior pattern language features, but also some less important feature information, in order to enable the model to focus on those principal component features that are highly correlated between language and behavior, and to highlight the role of these key features in the fusion process, in the technical solution of this application, the language pattern temporal semantic aggregation representation vector and the behavior pattern temporal semantic aggregation representation vector are input into a saliency fusion network based on feature principal component query matching to obtain a language-behavior multimodal saliency fusion representation vector. It is worth mentioning that the saliency fusion network based on feature principal component query matching is a feature extraction and fusion technique based on feature sparsification, principal component analysis, and saliency fusion, which is used to generate sparse and saliency feature association representations between feature vectors.

[0047] In the above embodiments, the language pattern temporal semantic aggregation representation vector and the behavior pattern temporal semantic aggregation representation vector are input into a saliency fusion network based on feature principal component query matching to obtain a language-behavior multimodal saliency fusion representation vector. This includes: First, standardizing the language pattern temporal semantic aggregation representation vector and the behavior pattern temporal semantic aggregation representation vector to obtain standardized language pattern temporal semantic aggregation representation vectors and standardized behavior pattern temporal semantic aggregation representation vectors; standardization eliminates the influence of differences in dimensions and numerical ranges, ensuring that each feature dimension is on the same numerical scale. Next, the sample covariance matrices of the standardized language pattern temporal semantic aggregation representation vectors and the standardized behavior pattern temporal semantic aggregation representation vectors are calculated to obtain language pattern sample covariance matrices and behavior pattern sample covariance matrices; and feature vector extraction based on matrix factorization is performed on the language pattern sample covariance matrix and the behavior pattern sample covariance matrix to obtain a set of language pattern principal component feature vectors and a set of behavior pattern principal component feature vectors; wherein, principal component features typically have stronger expressive power and can better capture the potential key semantic information contained in the original features. Subsequently, the sets of language pattern principal component feature vectors and the sets of behavior pattern principal component feature vectors are input into a maximum approximation query matching network to obtain a set of optimal matching pairs of language pattern principal component feature vectors and behavior pattern principal component feature vectors. That is, maximum approximation query matching is performed on the two sets based on the feature vectors, and the most relevant semantic query matching is performed by calculating the similarity between each feature, thereby finding the optimal matching pair. Furthermore, the optimal matching pairs of each language pattern principal component feature vector and behavior pattern principal component feature vector in the set of optimal matching pairs are input into a semantic fine-grained gating joint module to obtain a set of language-behavior pattern principal component fusion feature vectors. Here, a gating mechanism is used to perform fine-grained semantic joint on the optimal matching pairs to enhance and strengthen the expression of semantic relevance and importance in each matching pair, resulting in a set of language-behavior pattern principal component fusion feature vectors. Finally, the set of language-behavior pattern principal component fusion feature vectors is concatenated to obtain the language-behavior multimodal saliency fusion representation vector.

[0048] The process of standardizing the language pattern temporal semantic aggregation representation vector and the behavior pattern temporal semantic aggregation representation vector to obtain standardized language pattern temporal semantic aggregation representation vectors and standardized behavior pattern temporal semantic aggregation representation vectors includes: calculating the mean and standard deviation of the language pattern temporal semantic aggregation representation vectors to obtain the mean and standard deviation of the language pattern temporal semantic aggregation representation features; subtracting the language pattern temporal semantic aggregation representation vector from the mean of the language pattern temporal semantic aggregation representation features by position, and then calculating the language pattern temporal semantic aggregation representation offset vector. The standardized language pattern temporal semantic aggregation representation vector is obtained by dividing the standard deviation of the pattern temporal semantic aggregation representation feature by position. The mean and standard deviation of the behavior pattern temporal semantic aggregation representation vector are calculated to obtain the mean and standard deviation of the behavior pattern temporal semantic aggregation representation feature. The behavior pattern temporal semantic aggregation representation vector is subtracted from the mean of the behavior pattern temporal semantic aggregation representation feature by position, and the calculated behavior pattern temporal semantic aggregation representation offset vector is divided by the standard deviation of the behavior pattern temporal semantic aggregation representation feature by position to obtain the standardized behavior pattern temporal semantic aggregation representation vector.

[0049] More specifically, the process of calculating the sample covariance matrices of the standardized language pattern temporal semantic aggregation representation vector and the standardized behavior pattern temporal semantic aggregation representation vector to obtain the language pattern sample covariance matrix and the behavior pattern sample covariance matrix includes: multiplying the transpose of the standardized language pattern temporal semantic aggregation representation vector with the standardized language pattern temporal semantic aggregation representation vector, and then dividing the resulting standardized language pattern temporal semantic aggregation correlation matrix by the value obtained by subtracting one from the length of the standardized language pattern temporal semantic aggregation representation vector to obtain the language pattern sample covariance matrix; and multiplying the transpose of the standardized behavior pattern temporal semantic aggregation representation vector with the standardized behavior pattern temporal semantic aggregation representation vector, and then dividing the resulting standardized behavior pattern temporal semantic aggregation correlation matrix by the value obtained by subtracting one from the length of the standardized behavior pattern temporal semantic aggregation representation vector to obtain the behavior pattern sample covariance matrix.

[0050] More specifically, the process of inputting the set of language pattern principal component feature vectors and the set of behavior pattern principal component feature vectors into a maximum approximation query matching network to obtain a set of optimal matching pairs of language pattern principal component feature vectors and behavior pattern principal component feature vectors includes: extracting predetermined language pattern principal component feature vectors from the set of language pattern principal component feature vectors; calculating the cosine similarity between the predetermined language pattern principal component feature vectors and each behavior pattern principal component feature vector in the set of behavior pattern principal component feature vectors to obtain a set of matching query similarities; and taking the behavior pattern principal component feature vector corresponding to the largest matching query similarity in the set of matching query similarities and the predetermined language pattern principal component feature vectors as the optimal matching pair of the predetermined language pattern principal component feature vectors and behavior pattern principal component feature vectors.

[0051] More specifically, the process of inputting the best-matching pairs of each language pattern principal component feature vector and behavior pattern principal component feature vector from the set of best-matching pairs of language pattern principal component feature vectors and behavior pattern principal component feature vectors into the semantic fine-grained gating joint module to obtain the set of language-behavior pattern principal component fusion feature vectors includes: calculating the positional difference, positional dot product, and positional addition between the best-matching pairs of the language pattern principal component feature vectors and behavior pattern principal component feature vectors respectively to obtain the language-behavior principal component pattern difference vector, the language-behavior principal component pattern dot product vector, and the language-behavior principal component pattern summation vector; concatenating the language-behavior principal component pattern difference vector, the language-behavior principal component pattern dot product vector, and the language-behavior principal component pattern summation vector and then performing one-dimensional convolutional encoding to obtain the language-behavior principal component pattern multi-dimensional fusion vector; and performing local window-based max pooling processing on the language-behavior principal component pattern multi-dimensional fusion vector to obtain the language-behavior pattern principal component fusion feature vector.

[0052] In summary, in the above embodiments, inputting the language pattern temporal semantic aggregation representation vector and the behavior pattern temporal semantic aggregation representation vector into a saliency fusion network based on feature principal component query matching to obtain a language-behavior multimodal saliency fusion representation vector includes: inputting the language pattern temporal semantic aggregation representation vector and the behavior pattern temporal semantic aggregation representation vector into a saliency fusion network based on feature principal component query matching, and processing them with the following saliency fusion formula to obtain the language-behavior multimodal saliency fusion representation vector; wherein, the saliency fusion formula is:

[0053]

[0054] C1=U1Λ1U1 T

[0055] C2=U2Λ2U2 T

[0056] U1 = [v 11 ,v 12 ,…,v 1m ]

[0057] Λ1=diag(λ 11 ,λ 12 ,…,λ 1m )

[0058] U2 = [v 21 ,v 22 ,…,v 2m ]

[0059] Λ2=diag(λ 21 ,λ 22 ,…,λ 2m )

[0060]

[0061]

[0062] v f =[v p1 ;v p2 ;...v pm ]

[0063] Where V1 and V2 represent the language pattern temporal semantic aggregation representation vector and the behavior pattern temporal semantic aggregation representation vector, respectively; μ1 and μ2 are the means of the language pattern temporal semantic aggregation representation vector and the behavior pattern temporal semantic aggregation representation vector, respectively; and σ1 and σ2 are the standard deviations of the language pattern temporal semantic aggregation representation vector and the behavior pattern temporal semantic aggregation representation vector, respectively. 1norm and V 2norm V represents the standardized language pattern temporal semantic aggregation representation vector and the standardized behavior pattern temporal semantic aggregation representation vector, respectively. 1norm T and V 2norm T V 1norm and V 2norm The transpose of the vector, n1 and n2 are the number of eigenvalues ​​in the standardized language pattern temporal semantic aggregation representation vector and the standardized behavior pattern temporal semantic aggregation representation vector, respectively; C1 and C2 are the language pattern sample covariance matrix and the behavior pattern sample covariance matrix, respectively; U1 and U2 are the language pattern principal component orthogonal matrix and the behavior pattern principal component orthogonal matrix, respectively; Λ1 and Λ2 are the language pattern diagonal matrix and the behavior pattern diagonal matrix, respectively; diag(λ)11 ,λ 12 ,…,λ 1m ) is a matrix whose diagonal element is λ 11 ,λ 12 ,…,λ 1m Language pattern diagonal matrix, λ 11 ,λ 12 ,…,λ 1m These are the weights of the principal component feature vectors for each language pattern, diag(λ). 21 ,λ 22 ,…,λ 2m ) is a matrix whose diagonal element is λ 21 ,λ 22 ,…,λ 2m Behavioral pattern diagonal matrix, λ 21 ,λ 22 ,…,λ 2m These are the weights of the principal component feature vectors for each behavioral pattern, U1. T and U2 T The transposes of U1 and U2 are respectively, v 11 ,v 12 ,…,v 1m v represents the principal component feature vector of each language pattern in the set of principal component feature vectors of the language pattern. 21 ,v 22 ,…,v 2m The principal component feature vectors of each behavior pattern are in the set of principal component feature vectors of the behavior pattern. <v 1i ·v 2j >To calculate the v 1i and v 2j The dot product between vectors, ||·||, is used to calculate the 1-norm of the vectors. To return the j value corresponding to the maximum value, k is the maximum approximate matching value. ⊙ and These represent positional difference, positional dot product, and positional addition, respectively. [·,·,·] represents concatenation. conv1D(·) is a one-dimensional convolutional encoding operation, and MaxPool(·) is a max-pooling operation. pi is the i-th language-behavior pattern principal component fusion feature vector in the set of language-behavior pattern principal component fusion feature vectors, m is the number of feature vectors in the set of language-behavior pattern principal component fusion feature vectors, v f It is the language-behavior multimodal saliency fusion representation vector.

[0064] Specifically, the psychological development analysis result generation module 360 ​​is used to obtain psychological development analysis results based on the language-behavior multimodal saliency fusion representation vector, which indicates whether the child is developing positively. In a specific example of this application, the language-behavior multimodal saliency fusion representation vector is input into a classifier-based psychological development analysis module to obtain psychological development analysis results, which indicate whether the child is developing positively. That is, the language-behavior multimodal saliency fusion representation vector obtained by saliency fusion of the language pattern temporal semantic aggregation representation vector and the behavior pattern temporal semantic aggregation representation vector is used for classification processing to automatically assess whether the child being analyzed is developing positively. In this way, by using a large language model and deep learning technology, the system can automatically process and analyze large amounts of data, reducing the subjectivity and workload of manual interpretation. At the same time, it can capture important dynamic changes and trends in the process of children's psychological development, providing more detailed evidence for psychological counseling, thereby improving the intelligence level of children's psychological counseling.

[0065] Preferably, considering that the language pattern temporal semantic aggregation representation vector and the behavior pattern temporal semantic aggregation representation vector represent the global temporal guided aggregation features based on the significant decay of temporal nodes of local temporal semantic coding features of language communication data and behavioral habit data, respectively, when performing significant fusion based on feature principal component query matching, the significant difference in feature principal component query matching caused by the difference in the local-global temporal aggregation distribution of the source semantics will lead to fine-grained alignment conflict in semantic aggregation, thereby causing the loss of key aggregation information. This results in poor convergence mapping iteration consistency of the language-behavior multimodal significant fusion representation vector to the predetermined class probability category representation in the probability space, reducing the accuracy of the classification results.

[0066] Based on this, in the preferred example, when the language-behavior multimodal saliency fusion representation vector is input into the classifier-based psychosocial development analysis module to obtain psychosocial development analysis results, the language-behavior multimodal saliency fusion representation vector is optimized. Specifically, the optimization process includes the following steps: optimizing the first feature value v of the language-behavior multimodal saliency fusion representation vector... i Second eigenvalue v jThe relative probability value is obtained by calculating the absolute value of the difference and dividing it by the sum of absolute values, where the correction makes the relative probability value greater than zero and less than one. A first product is calculated of the first feature value multiplied by the relative probability value, and a second product is calculated of the second feature value multiplied by the absolute value of the difference between the relative probability value and one. A first quotient of the first feature value divided by the absolute value of the difference between the relative probability value and one and a second quotient of the second feature value divided by the relative probability value are also calculated. The first product and the second product are added together, and the difference between the first quotient and the second quotient is further subtracted to obtain the result relative to the first feature value v. i Second eigenvalue v j The corresponding eigenvalues ​​m of the optimization matrix i,j ; and, multiply the optimization matrix with the language-behavior multimodal saliency fusion representation vector to obtain the optimized language-behavior multimodal saliency fusion representation vector.

[0067] Specifically, it is expressed as:

[0068]

[0069] m i,j ∈M,v i ∈V,v j(j≠i) ∈V

[0070]

[0071] Where V represents the language-behavior multimodal saliency fusion representation vector, v i and v j p represents the first and second eigenvalues ​​of the language-behavior multimodal saliency fusion representation vector, respectively. i,j m represents the relative probability value. i,j Let M represent the corresponding eigenvalues ​​of the optimization matrix, and M represent the optimization matrix. V' represents matrix multiplication, and V' represents the optimized language-behavior multimodal saliency fusion representation vector.

[0072] Wherein, the first eigenvalue v i Second eigenvalue v j The corresponding relative probability value p i,j Represented as:

[0073]

[0074] Among them, v i and v j These represent the first and second eigenvalues ​​of the language-behavior multimodal saliency fusion representation vector, respectively, where ε represents a predetermined hyperparameter, and p... i,jThis represents the relative probability value. Finally, in the preferred example, the optimized language-behavior multimodal saliency fusion representation vector is input into a classifier-based psychosocial development analysis module to obtain the psychosocial development analysis results.

[0075] Here, regarding the region-boundary integral relationship of the high-dimensional feature manifold of the language-behavior multimodal saliency fusion representation vector, the probability distribution of the overall feature set of the language-behavior multimodal saliency fusion representation vector is defined as the manifold constraint boundary to approximate the simply connected region representation composed of eigenvalue pairs of the high-dimensional manifold of the feature set of the language-behavior multimodal saliency fusion representation vector. This avoids ambiguity in mapping the diverse feature representations of the language-behavior multimodal saliency fusion representation vector to the local manifold representation within the high-dimensional feature classification and regression convergence space, improves the consistency of execution iteration of each local feature distribution in the mapping task, thereby improving the convergence effect of feature classification and regression and enhancing the accuracy of the psychological development analysis results obtained by inputting the language-behavior multimodal saliency fusion feature vector into the classifier-based psychological development analysis module. In this way, by utilizing large language models and deep learning technology, the system can automatically process and analyze large amounts of data, reducing the subjectivity and workload of manual interpretation. Simultaneously, it can capture important dynamic changes and trends in the process of children's psychological development, providing more detailed evidence for psychological counseling, thereby improving the intelligence level of children's psychological counseling.

[0076] As described above, the child psychological development analysis system 300 based on a large language model according to embodiments of this application can be implemented in various wireless terminals, such as servers with child psychological development analysis algorithms based on large language models. In one possible implementation, the child psychological development analysis system 300 based on a large language model according to embodiments of this application can be integrated into the wireless terminal as a software module and / or hardware module. For example, the child psychological development analysis system 300 based on a large language model can be a software module in the operating system of the wireless terminal, or it can be an application developed for the wireless terminal; of course, the child psychological development analysis system 300 based on a large language model can also be one of many hardware modules of the wireless terminal.

[0077] Alternatively, in another example, the child psychological development analysis system 300 based on a large language model and the wireless terminal can also be separate devices, and the child psychological development analysis system 300 based on a large language model can be connected to the wireless terminal via wired and / or wireless networks, and transmit interactive information in accordance with an agreed data format.

[0078] Furthermore, a method for analyzing children's psychological development based on a large language model is also provided.

[0079] Figure 3 This is a flowchart of a child psychological development analysis method based on a large language model, according to an embodiment of this application. Figure 3 As shown, the child psychological development analysis method based on a large language model according to an embodiment of this application includes the following steps: S1, acquiring a time queue of language communication data and a time queue of behavioral habit data of the child to be analyzed; S2, using a large-model-based language communication semantic encoder to perform semantic understanding on each language communication data in the time queue of the language communication data to obtain a time queue of language communication data semantic encoding feature vectors; S3, using a large-model-based behavioral habit semantic encoder to perform semantic understanding on each behavioral habit data in the time queue of the behavioral habit data to obtain a time queue of behavioral habit data semantic encoding feature vectors; S4, converting the language... The time queues of the semantic encoding feature vectors of communication data and the time queues of the semantic encoding feature vectors of behavioral habit data are input into a feature aggregation network guided by the saliency decay of node features to obtain the temporal semantic aggregation representation vectors of language patterns and behavioral patterns; S5, the temporal semantic aggregation representation vectors of language patterns and behavioral patterns are input into a saliency fusion network based on feature principal component query matching to obtain a language-behavioral multimodal saliency fusion representation vector; S6, the language-behavioral multimodal saliency fusion representation vector is used to obtain a psychological development analysis result, which is used to indicate whether the development is positive.

[0080] In summary, the child psychological development analysis method based on a large language model, according to embodiments of this application, is explained. It employs natural language understanding and processing techniques based on a large language model and deep learning to semantically encode and aggregate temporal features of the child's language communication data and behavioral habit data, respectively. Based on the significant fusion of the temporal semantic aggregation features of the language communication data and the behavioral habit data, it automatically assesses whether the child is developing positively. Thus, by utilizing a large language model and deep learning technology, the system can automatically process and analyze large amounts of data, reducing the subjectivity and workload of manual interpretation. Simultaneously, it can capture important dynamic changes and trends in language and behavior during child psychological development, providing more detailed evidence for child psychological development analysis and counseling, and enhancing the intelligence level of child psychological counseling.

[0081] The various embodiments of this disclosure have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles, practical application, or improvement of the technology in the market, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. A child psychological development analysis system based on a large language model, characterized in that, include: The language communication behavior habit data acquisition module is used to acquire the time queue of language communication data and the time queue of behavior habit data of the children to be analyzed. The language communication data semantic understanding module is used to perform semantic understanding on each language communication data in the time queue of the language communication data using a large model-based language communication semantic encoder to obtain the time queue of language communication data semantic encoding feature vectors. The behavioral habit data semantic understanding module is used to perform semantic understanding on each behavioral habit data in the time queue of the behavioral habit data using a large model-based behavioral habit semantic encoder to obtain the time queue of behavioral habit data semantic encoded feature vectors. The language behavior pattern temporal semantic aggregation module is used to input the time queue of the semantic encoding feature vector of the language communication data and the time queue of the semantic encoding feature vector of the behavior habit data into the feature aggregation network guided by the saliency decay of node features to obtain the language pattern temporal semantic aggregation representation vector and the behavior pattern temporal semantic aggregation representation vector. The language-behavior multimodal fusion module is used to input the language pattern temporal semantic aggregation representation vector and the behavior pattern temporal semantic aggregation representation vector into a saliency fusion network based on feature principal component query matching to obtain a language-behavior multimodal saliency fusion representation vector; The psychological development analysis result generation module is used to obtain psychological development analysis results based on the language-behavior multimodal saliency fusion representation vector, and the psychological development analysis results are used to indicate whether the development is positive; Specifically, when inputting the language-behavior multimodal saliency fusion representation vector into a classifier-based psychological development analysis module to obtain psychological development analysis results, the language-behavior multimodal saliency fusion representation vector is optimized. The optimization process includes the following steps: optimizing the first feature value of the language-behavior multimodal saliency fusion representation vector... Second eigenvalue The relative probability value is obtained by calculating the absolute value of the difference and dividing it by the sum of the absolute values, where the correction makes the relative probability value greater than zero and less than one. A first product is calculated by multiplying the first feature value by the relative probability value, and a second product is calculated by multiplying the second feature value by the absolute value of the difference between the relative probability value and one. A first quotient is calculated for the first feature value divided by the absolute value of the difference between the relative probability value and one, and a second quotient is calculated for the second feature value divided by the relative probability value. The first product and the second product are added together, and the difference between the first quotient and the second quotient is further subtracted to obtain the result relative to the first feature value. Second eigenvalue The corresponding eigenvalues ​​of the optimization matrix ; and, multiply the optimization matrix with the language-behavior multimodal saliency fusion representation vector to obtain the optimized language-behavior multimodal saliency fusion representation vector.

2. The child psychological development analysis system based on a large language model according to claim 1, characterized in that, The language behavior pattern temporal semantic aggregation module includes: The semantic feature saliency descriptor calculation unit for language communication data is used to calculate the feature saliency descriptor of each language communication data semantic encoding feature vector in the time queue of the language communication data semantic encoding feature vector, wherein the feature saliency descriptor is related to the mean and variance of each language communication data semantic encoding feature vector; The language communication data semantic feature saliency decay factor construction unit is used to construct the feature saliency decay factor of each language communication data semantic feature vector based on the distance span between each language communication data semantic feature vector in the time queue and the current language communication data semantic feature vector. The language communication data semantic feature saliency attenuation factor modulation unit is used to modulate the feature saliency descriptor of each language communication data semantic encoding feature vector based on the feature saliency attenuation factor of each language communication data semantic encoding feature vector to obtain a sequence of language communication data semantic feature saliency attenuation descriptor factors. The language communication data semantic feature saliency attenuation factor masking unit is used to input the sequence of the language communication data semantic feature saliency attenuation descriptor factors into the gated masking module to obtain the sequence of language communication data semantic feature saliency attenuation weight factors. The language pattern temporal semantic aggregation representation unit is used to calculate the weighted sum of the time queues of the language communication data semantic encoding feature vectors, using the sequence of the saliency decay weight factors of the semantic features of the language communication data as the sequence of weights, to obtain the language pattern temporal semantic aggregation representation vector.

3. The child psychological development analysis system based on a large language model according to claim 2, characterized in that, The semantic feature saliency descriptor calculation unit for language communication data is used for: The mean and variance of the semantic encoding feature vector of the language communication data are calculated respectively to obtain the mean and variance of the semantic features of the language communication data. The positional difference between the semantic encoding feature vector of the language communication data and the mean of the semantic features of the language communication data is calculated to obtain the semantic difference vector of the language communication data. Calculate the fourth power of each feature value in the semantic difference vector of the language communication data to obtain the semantic modulation difference vector of the language communication data; Calculate the expected value of the semantic modulation difference vector of the language communication data to obtain the expected value of the language communication data; The expected value of the language communication data is divided by the square of the variance of the semantic features of the language communication data to obtain the feature saliency descriptor corresponding to the semantic encoding feature vector of the language communication data.

4. The child psychological development analysis system based on a large language model according to claim 3, characterized in that, The semantic feature saliency attenuation factor construction unit of the language communication data is used for: Extract the maximum value from the semantic encoding feature vectors of each language communication data to obtain a sequence of maximum values ​​for the semantic features of the language communication data; Extract the maximum value of the semantic coding feature vector of the current language communication data to obtain the maximum value of the semantic coding feature of the current language communication data; The sequence of maximum values ​​of the semantic features of the language communication data is calculated by subtracting the maximum value of the semantic encoding features of the current language communication data from the sequence of maximum values ​​of the semantic features of the language communication data by position to obtain the sequence of offset values ​​of the semantic features of the language communication data. The distance span values ​​between the semantic coding feature vectors of each language communication data and the current semantic coding feature vector of the language communication data are statistically analyzed to obtain a sequence of semantic distance span values ​​of language communication data. The square of each value in the sequence of semantic distance span values ​​of language communication data is calculated to obtain a sequence of semantic distance span modulation values ​​of language communication data. The sequence of semantic feature offset values ​​of the language communication data is divided by the sequence of semantic distance span modulation values ​​of the language communication data according to position to obtain the feature saliency attenuation factor of each semantic coding feature vector of the language communication data.

5. The child psychological development analysis system based on a large language model according to claim 4, characterized in that, The semantic feature saliency attenuation factor masking unit for language communication data is used for: For each of the language communication data semantic feature saliency decay descriptor in the sequence of language communication data semantic feature saliency decay descriptor, a normalization process based on the Sigmoid function is performed to obtain a set of normalized language communication data semantic feature saliency decay descriptor. Each normalized language communication data semantic feature saliency attenuation descriptor in the set of normalized language communication data semantic feature saliency attenuation descriptors is input into a gating function for masking to obtain a sequence of language communication data semantic feature saliency attenuation weight factors.

6. The child psychological development analysis system based on a large language model according to claim 5, characterized in that, The language-behavior multimodal fusion module includes: The language behavior pattern temporal semantic feature standardization unit is used to standardize the language pattern temporal semantic aggregated representation vector and the behavior pattern temporal semantic aggregated representation vector to obtain the standardized language pattern temporal semantic aggregated representation vector and the standardized behavior pattern temporal semantic aggregated representation vector. The language behavior pattern sample covariance matrix calculation unit is used to calculate the sample covariance matrix of the standardized language pattern temporal semantic aggregation representation vector and the standardized behavior pattern temporal semantic aggregation representation vector to obtain the language pattern sample covariance matrix and the behavior pattern sample covariance matrix. The language-behavior pattern principal component feature extraction unit is used to extract feature vectors based on matrix decomposition from the language pattern sample covariance matrix and the behavior pattern sample covariance matrix to obtain a set of language pattern principal component feature vectors and a set of behavior pattern principal component feature vectors. The language-behavior maximum approximation query unit is used to input the set of language pattern principal component feature vectors and the set of behavior pattern principal component feature vectors into the maximum approximation query matching network to obtain the set of best matching pairs of language pattern principal component feature vectors and behavior pattern principal component feature vectors. The language-behavior pattern principal component fusion unit is used to input each best-matching pair of language pattern principal component feature vectors and behavior pattern principal component feature vectors from the set of best-matching pairs of language pattern principal component feature vectors and behavior pattern principal component feature vectors into the semantic fine-grained gating joint module to obtain a set of language-behavior pattern principal component fusion feature vectors. The language-behavior multimodal saliency fusion unit is used to concatenate the set of language-behavior mode principal component fusion feature vectors to obtain the language-behavior multimodal saliency fusion representation vector; The language behavior pattern sample covariance matrix calculation unit is used for: Multiply the transpose of the standardized language pattern temporal semantic aggregation representation vector with the standardized language pattern temporal semantic aggregation representation vector, and then divide the resulting standardized language pattern temporal semantic aggregation correlation matrix by the value obtained by subtracting one from the length of the standardized language pattern temporal semantic aggregation representation vector to obtain the language pattern sample covariance matrix. After multiplying the transpose of the standardized behavior pattern temporal semantic aggregation representation vector with the standardized behavior pattern temporal semantic aggregation representation vector, the resulting standardized behavior pattern temporal semantic aggregation association matrix is ​​divided by the value obtained by subtracting one from the length of the standardized behavior pattern temporal semantic aggregation representation vector in terms of position to obtain the behavior pattern sample covariance matrix.

7. The child psychological development analysis system based on a large language model according to claim 6, characterized in that, The language behavior pattern temporal semantic feature standardization unit is used for: The mean and standard deviation of the language pattern temporal semantic aggregation representation vector are calculated respectively to obtain the mean and standard deviation of the language pattern temporal semantic aggregation representation features; The standardized language pattern temporal semantic aggregation representation vector is obtained by subtracting the mean of the language pattern temporal semantic aggregation representation features from the vector by position, and then dividing the calculated language pattern temporal semantic aggregation representation offset vector by the standard deviation of the language pattern temporal semantic aggregation representation features by position. The mean and standard deviation of the behavioral pattern temporal semantic aggregation representation vector are calculated respectively to obtain the mean and standard deviation of the behavioral pattern temporal semantic aggregation representation features; The standardized behavioral pattern temporal semantic aggregated representation vector is obtained by subtracting the mean of the behavioral pattern temporal semantic aggregated representation features from the mean at position, and then dividing the calculated behavioral pattern temporal semantic aggregated representation offset vector by the standard deviation of the behavioral pattern temporal semantic aggregated representation features at position.

8. The child psychological development analysis system based on a large language model according to claim 7, characterized in that, The language-behavior maximum approximation query unit is used for: Extract the predetermined language pattern principal component feature vectors from the set of language pattern principal component feature vectors; Calculate the cosine similarity between the predetermined language pattern principal component feature vector and each behavior pattern principal component feature vector in the set of behavior pattern principal component feature vectors to obtain a set of matching query similarities; The behavioral pattern principal component feature vector corresponding to the largest matching query similarity in the set of matching query similarities is taken as the optimal matching pair between the predetermined language pattern principal component feature vector and the behavioral pattern principal component feature vector.

9. The child psychological development analysis system based on a large language model according to claim 8, characterized in that, The language-behavior pattern principal component fusion unit is used for: Calculate the positional difference, positional dot product, and positional addition between the best matching pairs of the language pattern principal component feature vector and the behavior pattern principal component feature vector to obtain the language-behavior principal component pattern difference vector, the language-behavior principal component pattern dot product vector, and the language-behavior principal component pattern summation vector; The language-behavior principal component pattern difference vector, the language-behavior principal component pattern dot product vector, and the language-behavior principal component pattern summation vector are concatenated and then subjected to one-dimensional convolutional encoding to obtain a language-behavior principal component pattern multi-dimensional fusion vector. The language-behavior principal component pattern multidimensional fusion vector is subjected to local window-based max pooling to obtain the language-behavior pattern principal component fusion feature vector.

10. A method for analyzing children's psychological development based on a large language model, characterized in that, include: Time queues for acquiring language communication data and behavioral habit data of the children to be analyzed; A large-model-based language communication semantic encoder is used to perform semantic understanding on each language communication data in the time queue of the language communication data to obtain the time queue of language communication data semantic encoded feature vectors. A large-model-based behavioral habit semantic encoder is used to perform semantic understanding on each behavioral habit data in the time queue of the behavioral habit data to obtain the time queue of behavioral habit data semantic encoded feature vectors. The time queues of the semantic encoding feature vectors of the language communication data and the time queues of the semantic encoding feature vectors of the behavioral habit data are input into a feature aggregation network guided by the saliency decay of node features to obtain the temporal semantic aggregation representation vectors of the language pattern and the temporal semantic aggregation representation vectors of the behavioral pattern. The language pattern temporal semantic aggregation representation vector and the behavior pattern temporal semantic aggregation representation vector are input into a saliency fusion network based on feature principal component query matching to obtain a language-behavior multimodal saliency fusion representation vector; The language-behavior multimodal saliency fusion representation vector is used to obtain the psychological development analysis results, which are used to indicate whether the development is positive; Specifically, when inputting the language-behavior multimodal saliency fusion representation vector into a classifier-based psychological development analysis module to obtain psychological development analysis results, the language-behavior multimodal saliency fusion representation vector is optimized. The optimization process includes the following steps: optimizing the first feature value of the language-behavior multimodal saliency fusion representation vector... Second eigenvalue The relative probability value is obtained by calculating the absolute value of the difference and dividing it by the sum of the absolute values, where the correction makes the relative probability value greater than zero and less than one. A first product is calculated by multiplying the first feature value by the relative probability value, and a second product is calculated by multiplying the second feature value by the absolute value of the difference between the relative probability value and one. A first quotient is calculated for the first feature value divided by the absolute value of the difference between the relative probability value and one, and a second quotient is calculated for the second feature value divided by the relative probability value. The first product and the second product are added together, and the difference between the first quotient and the second quotient is further subtracted to obtain the result relative to the first feature value. Second eigenvalue The corresponding eigenvalues ​​of the optimization matrix ; and, multiply the optimization matrix with the language-behavior multimodal saliency fusion representation vector to obtain the optimized language-behavior multimodal saliency fusion representation vector.

Images