Children-oriented text grading method and system based on artificial intelligence

Abstract

This invention discloses a method and system for grading text for children based on artificial intelligence. The method comprises preprocessing the text content to be evaluated to generate a standardized text representation; extracting a multidimensional feature vector from the standardized text representation using a pretrained model, including thematic, sentiment, and language complexity features; generating a content suitability score based on the multidimensional feature vector using a multi-layer perceptron; and predicting an age rating based on the suitability score and the multidimensional feature vector using a multi-classification model. Compared to traditional methods, this method can identify implicit and harmful information and improve the safety and efficiency of children's content review through automated review, thus possessing significant innovation and social benefits.

Description

Technical Field

[0001] The present invention belongs to the field of artificial intelligence technology, and in particular relates to an artificial intelligence-based text grading method and system for children. Background Art

[0002] With the development of LLM technology, children's educational and entertainment content is increasingly dominated by personalized text generated in real time, such as instantly created children's stories or interactive dialogues. However, traditional review methods are completely unable to adapt to this change. Manual review is inefficient and unable to process the massive amount of data generated in real time. Keyword filtering, lacking contextual understanding, is unable to identify obscure negative information or determine the suitability of complex educational content. Therefore, existing technologies cannot protect children from potential risks, and innovative solutions are urgently needed. Traditional keyword filtering solutions have the following significant flaws:

[0003] Insufficient efficiency and coverage: Traditional manual review is time-consuming, unable to process the massive amount of data generated in real time, and has limited coverage, which can no longer meet the growing demand.

[0004] Consistency issues: Manual evaluations are subject to inconsistent standards due to differences in reviewer experience and subjective bias, making it difficult to ensure the reliability of the results.

[0005] Insufficient analysis depth: Existing automated tools often rely on simple keyword matching, which can overlook subtle and harmful information. For example, content may not contain explicitly offensive words but convey negative emotions or values that are inconsistent with social norms through metaphors or narratives, potentially negatively impacting children's psychological development and behavioral habits.

[0006] Misjudgment of complex content: For complex educational content suitable for children (such as science books), keyword filtering may misjudge it as inappropriate due to the high complexity of the language, limiting children's opportunities to obtain useful information. Summary of the Invention

[0007] In response to the above-mentioned problems, the present invention provides a text grading method and system for children based on artificial intelligence, aiming to provide an intelligent system that can comprehensively analyze the theme, emotional tendency, language complexity and potential value influence of the text, and accurately grade it according to the age group of children, so as to improve the safety, age appropriateness and positive guidance of the content received by children.

[0008] According to a first aspect of an embodiment of the present disclosure, a method for text grading for children based on artificial intelligence is provided, the method comprising the following steps:

[0009] Preprocessing the text content to be evaluated to generate a standardized text representation, wherein the preprocessing includes text segmentation, stop word filtering, and word form normalization;

[0010] The standardized text representation is represented by using a pre-trained model to generate a semantic representation of the text, and a multi-dimensional feature vector is extracted based on the semantic representation, wherein the multi-dimensional feature vector includes topic classification, sentiment tendency and language complexity features; the standardized text representation is represented by using a pre-trained model to extract a multi-dimensional feature vector, specifically comprising: based on the input of the standardized text representation, using a pre-trained BERT model to obtain a multi-dimensional vector of CLS tags and obtain a topic classification feature vector; based on the input of the standardized text representation, using a pre-trained BERT model for encoding, generating a multi-dimensional semantic representation for each token to obtain sentiment tendency confidence; based on the input of the standardized text representation, using a pre-trained BERT model for encoding, generating a multi-dimensional semantic for each token, adding a dual affine attention layer, predicting the dependency between each token, quantifying the language complexity to obtain language complexity features;

[0011] Inputting the multi-dimensional feature vector into a multi-layer perceptron, wherein the multi-layer perceptron processes the nonlinear mapping of the multi-dimensional feature vector through a Sigmoid function to generate a content suitability score from 0 to 100;

[0012] The content suitability score is normalized, fused with the multi-dimensional feature vector, and then input into the Softmax classifier. Finally, the trained Softmax classifier outputs the age-appropriate classification label prediction result.

[0013] In some embodiments, extracting a multi-dimensional feature vector from the standardized text representation using a pre-trained model specifically includes:

[0014] Based on the input of standardized text representation, a pre-trained or BERT model is used to obtain a 768-dimensional vector of CLS tags, which is used as the feature vector for topic classification and input into the support vector machine for topic classification;

[0015] Based on the input of the standardized text representation, the pre-trained BERT model is used for encoding to generate a 768-dimensional semantic representation for each token. A fully connected layer is added to map the 768-dimensional semantic representation to the probability distribution of the sentiment label and output the confidence score through the softmax function.

[0016] Based on the input of standardized text representation, the pre-trained BERT model is used for encoding, generating a 768-dimensional semantic representation for each token. A dual affine attention layer is added to predict the dependencies between tokens and generate a syntactic tree. The language complexity is quantified based on the topological analysis and information entropy calculation of the syntactic tree.

[0017] The extracted feature vectors are normalized to ensure that the dimensions of features in different dimensions are consistent.

[0018] In some embodiments, the multi-dimensional feature vector also includes social value features, which are extracted through graph embedding technology of knowledge graph.

[0019] In some embodiments, a content suitability score is generated based on a multi-dimensional feature vector using a multi-layer perceptron. The specific expression is:

[0020] ,

[0021] in represents the content suitability score, represents the Sigmoid function, is the weight matrix of the neural network, x represents the multi-dimensional feature vector, Represents the bias vector, and ReLU represents the rectified linear unit function.

[0022] In some embodiments, using a classification model to predict age rating based on the suitability score and the feature vector includes:

[0023] Fusion of suitability score and feature vector into input vector;

[0024] Output the probability distribution of each age level through the Softmax classifier;

[0025] The age category with the highest probability is selected as the prediction result.

[0026] In some embodiments, the content suitability scoring process further includes:

[0027] The MLP model is pre-trained and its parameters are optimized using a labeled dataset that includes both appropriate and inappropriate children's text samples.

[0028] In some embodiments, the age rating further includes:

[0029] Predefine the cognitive and psychological characteristics of children of different age groups, including vocabulary comprehension and emotional receptivity;

[0030] Train the Softmax classifier using historical data.

[0031] According to a second aspect of an embodiment of the present disclosure, there is provided an artificial intelligence-based text rating system for children, the system comprising:

[0032] A preprocessing module, configured to preprocess the text content to be evaluated and generate a standardized text representation, wherein the preprocessing includes text segmentation, stop word filtering, and word form normalization;

[0033] A feature extraction module is used to generate a semantic representation of the text using a pre-trained model for the standardized text representation, and extract a multi-dimensional feature vector based on the semantic representation, wherein the multi-dimensional feature vector includes topic classification, sentiment tendency and language complexity features; extracting a multi-dimensional feature vector from the standardized text representation using a pre-trained model specifically includes: based on the input of the standardized text representation, using a pre-trained BERT model to obtain a multi-dimensional vector of CLS tags and obtain a topic classification feature vector; based on the input of the standardized text representation, using a pre-trained BERT model for encoding, generating a multi-dimensional semantic representation for each token to obtain sentiment tendency confidence; based on the input of the standardized text representation, using a pre-trained BERT model for encoding, generating a multi-dimensional semantic for each token, adding a dual affine attention layer, predicting the dependency between each token, quantifying the language complexity to obtain language complexity features;

[0034] A content suitability scoring module, configured to input the multi-dimensional feature vector into a multi-layer perceptron, wherein the multi-layer perceptron processes the nonlinear mapping of the multi-dimensional feature vector using a Sigmoid function to generate a content suitability score ranging from 0 to 100;

[0035] The age rating module is used to normalize the content suitability score, fuse it with the multi-dimensional feature vector, and input it into the Softmax classifier. Finally, the trained Softmax classifier outputs the age-appropriate rating label prediction result.

[0036] According to a third aspect of an embodiment of the present disclosure, an electronic device is provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps of the above-mentioned artificial intelligence-based text grading method for children are implemented.

[0037] According to a fourth aspect of an embodiment of the present disclosure, a non-transitory computer-readable storage medium is provided, on which computer instructions are stored. When the instructions are executed by a processor, the steps of the above-mentioned artificial intelligence-based text grading method for children are implemented.

[0038] The embodiments of the present disclosure provide a method, system, electronic device, and storage medium for text grading based on artificial intelligence for children, which utilizes multi-dimensional feature extraction and deep learning technology to achieve efficient and accurate automated review. The system includes text feature extraction, suitability scoring, and age grading modules, equipped with a processor and memory, and uses natural language processing and machine learning to analyze text themes, emotions, and language complexity to generate suitability scores and predict age-appropriate ratings. Compared with traditional methods, this system can identify obscure and harmful information, improve the safety and review efficiency of children's content through automated review, and has significant innovation and social benefits. Specifically, the beneficial effects of the present invention include:

[0039] Multi-dimensional analysis: Integrates multiple features such as themes, sentiment, language complexity, and values, going beyond traditional keyword matching methods to ensure a comprehensive and in-depth assessment.

[0040] Intelligent identification of implicit information: Through semantic understanding and context analysis, it can identify implicit negative content or negative tendencies and improve review accuracy.

[0041] Efficient automation: Real-time evaluation and grading of text content is achieved, reducing manual intervention and improving review efficiency.

[0042] Accurate age grading: Based on the cognitive and psychological needs of children of different age groups, the system outputs age-appropriate grading results with high pertinence.

[0043] Transparent feedback: Provides detailed evaluation reports to facilitate content creators to optimize texts, while supporting effective supervision by parents and regulators.

[0044] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present invention and, together with the description, serve to explain the principles of the present invention.

[0046] Figure 1 This is a flow chart of a text grading method for children based on artificial intelligence in an embodiment of the present invention;

[0047] Figure 2 1 is a schematic diagram of a flow chart for preprocessing text content to be evaluated in an embodiment of the present invention;

[0048] Figure 3 1 is a schematic diagram of the structure of a text grading system for children based on artificial intelligence in an embodiment of the present invention;

[0049] Figure 4 It is a schematic diagram of an electronic device in an embodiment of the present invention. DETAILED DESCRIPTION

[0050] The present invention will be further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are intended only to illustrate the present invention and are not intended to limit the present invention. It should also be noted that, for ease of description, the accompanying drawings only illustrate portions relevant to the present invention, not all structures.

[0051] Before discussing the exemplary embodiments in more detail, it should be mentioned that some exemplary embodiments are described as processes or methods depicted as flow charts. Although the flow charts describe the steps as sequential processes, many of the steps can be implemented in parallel, concurrently, or simultaneously. In addition, the order of the steps can be rearranged. The process can be terminated when its operation is completed, but can also have additional steps not included in the accompanying drawings. The process can correspond to a method, function, procedure, subroutine, subprogram, etc.

[0052] The present invention provides the following embodiments for a method, system, electronic device, and storage medium for text grading based on artificial intelligence for children:

[0053] A text grading method based on artificial intelligence for children, such as Figure 1 As shown, the method includes the following steps:

[0054] S1. Preprocessing the text content to be evaluated to generate a standardized text representation, wherein the preprocessing includes text segmentation, stop word filtering, and word form normalization;

[0055] S2. Generate a semantic representation of the text using a pre-trained model for the standardized text representation, and extract a multi-dimensional feature vector based on the semantic representation, wherein the multi-dimensional feature vector includes topic classification, sentiment tendency, and language complexity features; extracting a multi-dimensional feature vector from the standardized text representation using a pre-trained model specifically includes: based on the input of the standardized text representation, using a pre-trained BERT model to obtain a multi-dimensional vector of CLS tags and obtain a topic classification feature vector; based on the input of the standardized text representation, using a pre-trained BERT model for encoding, generating a multi-dimensional semantic representation for each token to obtain sentiment tendency confidence; based on the input of the standardized text representation, using a pre-trained BERT model for encoding, generating a multi-dimensional semantic for each token, adding a dual affine attention layer, predicting the dependency between each token, quantifying the language complexity to obtain language complexity features;

[0056] S3. Inputting the multidimensional feature vector into a multilayer perceptron, wherein the multilayer perceptron processes the nonlinear mapping of the multidimensional feature vector through a Sigmoid function to generate a content suitability score from 0 to 100;

[0057] S4. Normalize the content suitability score, fuse it with the multi-dimensional feature vector, and input it into the Softmax classifier. Finally, the trained Softmax classifier outputs the age-appropriate classification label prediction result.

[0058] The preprocessing in S1 includes word segmentation based on statistical models, stop word filtering and word form normalization. In the preferred embodiment, it also includes sentence segmentation and part-of-speech tagging of the text. The specific implementation steps are as follows: Figure 2 Shown, including:

[0059] S11, word segmentation: divide the text into word units;

[0060] S12, stop word filtering: remove meaningless common words;

[0061] S13, word form normalization: unify the word form;

[0062] S14. Syntactic analysis: Generate syntactic structure.

[0063] Furthermore, the input text is segmented into Chinese words (based on the Conditional Random Field (CRF) model), stop words are removed, and word form normalization is performed. The word segmentation model is optimized using the maximum entropy criterion, and the transition probability is calculated using the following formula:

[0064] ,

[0065] in represents the transition probability, exp represents the exponential function, ∑k represents the sum of index k, is the weight parameter, is the characteristic function, Indicates all possible Summation.

[0066] In S2, the standardized text representation is extracted using a pre-trained model to extract multi-dimensional feature vectors, including:

[0067] Based on the input of standardized text representation, the pre-trained BERT model is used to obtain the 768-dimensional vector of CLS tags, which is used as the feature vector for topic classification and input into the support vector machine for topic classification;

[0068] Specifically, topic classification combines traditional statistical methods with deep semantic representation technology to build a high-dimensional topic classification model. The semantic representation of the pre-trained BERT (Bidirectional Encoder Representation) model is optimized, and the multi-category classification task is completed using a support vector machine (SVM).

[0069] Further preferably, a pre-trained BERT model is loaded, a text sequence after word segmentation is input, and a 768-dimensional vector of CLS tags is obtained as a feature vector for topic classification; the feature vector is input into SVM, and the kernel function is a radial basis function (RBF):

[0070] ,

[0071] in represents a similarity measure; Represents the feature vector extracted by BERT; represents the squared Euclidean distance; Represents the parameter that controls the influence of distance; exp represents the exponential function.

[0072] The optimization goal is:

[0073] ,

[0074] Where w represents the weight vector; b represents the bias term; represents the square of the weight norm; C represents the regularization parameter; represents the slack variable; represents the true label; represents the distance from the sample to the hyperplane; n represents the total number of samples.

[0075] Based on the input of the standardized text representation, the pre-trained BERT model is used for encoding to generate a 768-dimensional semantic representation for each token. A fully connected layer is added to map the 768-dimensional semantic representation to the probability distribution of the sentiment label and output the confidence score through the softmax function.

[0076] Specifically, we use the BERT model combined with the self-attention mechanism to build a sentiment classification model and quantify sentiment polarity and intensity. The specific implementation steps include:

[0077] Text encoding: Use the pre-trained BERT model to encode the input text and generate a 768-dimensional embedding vector for each token.

[0078] CLS token output: Extract the CLS token vector (768 dimensions) output by the BERT model as the semantic representation of the entire text.

[0079] Sentiment classification: Add a fully connected layer to the CLS token output, map it to the probability distribution of sentiment labels (such as positive, negative, neutral), and output the confidence through the softmax function.

[0080] Model fine-tuning: Fine-tune the BERT model on sentiment analysis tasks and use a child-oriented annotated sentiment dataset to optimize model parameters to adapt to the language characteristics and emotional expressions of children's texts.

[0081] Based on the input of standardized text representation, the pre-trained BERT model is used for encoding, generating a 768-dimensional semantic representation for each token. A dual affine attention layer is added to predict the dependencies between tokens and generate a syntactic tree. The language complexity is quantified based on the topological analysis and information entropy calculation of the syntactic tree.

[0082] Specifically, we combine dependency parsing and information entropy to quantify the complexity of syntactic structure and lexical diversity. The specific implementation steps include:

[0083] Syntactic analysis: Use the BERT-based dependency parsing model to generate a syntactic tree. The specific steps are as follows:

[0084] Use the pre-trained BERT model to encode the text and obtain a 768-dimensional semantic representation for each token;

[0085] A biaffine attention layer is added to the BERT output to predict the dependency between tokens and generate a syntactic tree.

[0086] Syntactic complexity, including: tree depth: calculate the maximum depth of the syntax tree (tree_depth); branching factor: calculate the average branching factor of the syntax tree ; Comprehensive syntactic complexity: , where β is the weighting coefficient of the branching factor.

[0087] Lexical entropy: Calculate the lexical entropy of the text, denoted as vocab_entropy, which is used to measure the diversity of vocabulary.

[0088] Complexity fusion: The final language complexity score is generated through weighted summation. The formula is as follows:

[0089]

[0090] Among them, w1 and w2 are the weight coefficients of syntactic complexity and lexical entropy respectively.

[0091] Preferably, the multi-dimensional feature vector also includes social value features, which are extracted through graph embedding technology of the knowledge graph.

[0092] Specifically, based on the knowledge graph and graph embedding technology (TransE), combined with the sentiment dictionary, value orientation is extracted. The specific implementation steps include:

[0093] Knowledge graph construction: Contains positive and negative value entities and their relationships.

[0094] Graph Embedding: ,

[0095] Where min means that the optimization goal is to minimize the value of the entire function; Represents all correct triples in the knowledge graph Perform summation; Embedding vectors representing head entities, relations, and tail entities, learned by the TransE model; Indicates the wrong head entity and tail entity generated by negative sampling; The embedding vector distance representing the correct triple is calculated using the squared L2 norm, which should be small. The embedding vector distance representing the error triple is also calculated using the squared L2 norm, and the value should be large; Represents the margin, which is a hyperparameter (such as 0.1 or 1) used to control the difference in the distance between the correct triples and the incorrect triples.

[0096] Text Representation: Generated using BERT , knowledge graph entities are embedded through TransE ;

[0097] Similarity calculation: ;

[0098] Where sim represents the cosine similarity score, ranging from -1 to 1. The closer the value is to 1, the more similar the text and entity semantics are; BERT embedding vectors representing text capture the semantic information of the text; TransE embedding vector representing the knowledge graph entity, representing positive or negative values; Represents the dot product of two vectors, measuring their similarity in direction; Represents the L2 norm of text embedding vectors and entity embedding vectors, used for normalization.

[0099] Tendency judgment: Determine value tendencies based on thresholds.

[0100] It should be noted that the extracted feature vectors are normalized to ensure that the dimensions of features of different dimensions are consistent. That is, the feature vectors output by each submodule are normalized and then weighted fused to generate the feature vector x=[T,E,C,V] 。Where T represents the topic classification features of the text, such as the probability distribution or classification label of the text belonging to categories such as education, entertainment, violence, etc. These features are usually generated by pre-trained BERT models and support vector machines (SVM) to help the system understand the main content and type of the text. E represents the emotional tendency features of the text, such as emotional polarity (positive, negative, neutral) and its intensity. These features are generated by the BERT model combined with the self-attention mechanism to evaluate the impact of the text on children's emotions. C represents the language complexity quantification indicators of the text, such as complexity scores calculated based on syntactic tree depth, branching factor and lexical entropy. These features are generated through dependency syntax analysis and information entropy calculation to help the system determine whether the language difficulty of the text is suitable for children. V represents the social value tendency features of the text, such as positive or negative value tendencies extracted through knowledge graphs and graph embedding technology (TransE). These features are used to evaluate whether the text conveys values that conform to social norms and children's educational needs. These features are generated separately by the four sub-modules of the text feature extraction module and integrated into a comprehensive feature vector x , as the input of the suitability scoring module and the age grading module. Through this multi-dimensional feature extraction, the system can comprehensively analyze the text content to ensure the accuracy and scientificity of the evaluation.

[0101] In S3, content suitability scores are generated by a multi-layer perceptron based on multi-dimensional feature vectors. The specific expression is:

[0102] ,

[0103] in represents the content suitability score, which ranges from 0 to 100. The higher the score, the more suitable the content is for children. x represents the feature vector, including topic classification, emotional tendency, and language complexity. : The weight matrix of the neural network; Represents the bias vector; ReLU(z)=max(0,z) represents the rectified linear unit function; : Sigmoid function, which maps the output to the range of 0 to 1.

[0104] The content suitability scoring process also includes pre-training the MLP model and optimizing its parameters using a labeled dataset containing samples of appropriate and inappropriate text for children. During the inference phase, feature vectors are input in real time and the suitability score is output.

[0105] Specifically, the dynamic suitability score uses a multi-layer perceptron (MLP) architecture to perform nonlinear fusion on feature vectors to generate a suitability score (0-100). The specific implementation process includes:

[0106] Feature Embedding: , where We is the weight matrix of the fully connected layer, be is the bias vector, and x is the comprehensive feature vector of the input;

[0107] Single-head self-attention: (where Q = WQ⋅e, K = WK⋅e, and V = WV⋅e are the linear transformations of query, key, and value, respectively, and d is the feature dimension).

[0108] Feedforward Network (FFN): , where W1 is the first layer weight matrix and b1 is the bias vector);

[0109] Output: , where W2 is the second layer weight matrix, b2 is the bias vector, and σ is the Sigmoid function.

[0110] In S4, a classification model is used to predict age classification based on the suitability score and feature vector, including:

[0111] Fusion of suitability score and feature vector into input vector;

[0112] Output the probability distribution of each age level through the Softmax classifier;

[0113] The age category with the highest probability is selected as the prediction result.

[0114] Furthermore, age classification includes:

[0115] Predefine the cognitive and psychological characteristics of children of different age groups, including vocabulary comprehension and emotional receptivity;

[0116] Train the Softmax classifier using historical data.

[0117] Specifically, age classification is based on a multi-classification model (such as a Softmax classifier) that combines the suitability score and feature vector to predict the appropriate age classification. The model is trained using a cross-entropy loss and introduces an L2 regularization term to prevent overfitting. The implementation steps include:

[0118] (1) Feature fusion: The input of the age grading module is the suitability score and the feature vector. First, the suitability score (Score) and the feature vector (x) are fused to form a unified input vector , the specific operations are as follows:

[0119] Normalize the suitability score (range [0, 100]) to the interval [0, 1], that is, , to ensure consistency with the dimension of the eigenvector.

[0120] The normalized suitability score Combined with the feature vector x (including multi-dimensional features such as topic classification, sentiment tendency, language complexity and value assessment), the fused input vector is obtained. .

[0121] (2) Classification prediction: fused input vector It is fed into the Softmax classifier to output the probability distribution of each age category. The specific process is as follows:

[0122] The Softmax classifier first classifies the input vector Perform a linear transformation and calculate the score (logits) for each age level: z=Wx′+b

[0123] Where W is the weight matrix, b is the bias vector, z is the score vector for each age level, and the dimension is equal to the number of age levels K (such as 0-3 years old, 3-6 years old, 7-12 years old, etc.).

[0124] Then, the score z is converted into a probability distribution through the Softmax function: , where P(y=k|x′) represents the probability that the input x′ is classified as the kth age category, and K is the total number of age categories.

[0125] (3) Prediction results: According to the probability distribution output by the Softmax classifier, the age grade with the highest probability is selected as the final prediction result: ,in, is the predicted age-appropriate classification label.

[0126] (4) Model training: The parameters of the Softmax classifier (W and b) are optimized through supervised learning. The training goal is to minimize the cross entropy loss function: , where: N is the number of training samples; is the true label of the i-th sample (using one-hot encoding, if the sample belongs to the k-th category, then , otherwise 0); is the probability predicted by the model. To prevent the model from overfitting, an L2 regularization term is added to the loss function: , in, is a hyperparameter of regularization strength, determined by cross-validation, is the L2 norm squared sum of the weights W. During training, the Adam optimizer is used to update the model parameters, and a learning rate decay strategy is introduced (for example, the initial learning rate is set to 0.001 and decays by 10% every few rounds) to improve the model convergence speed and stability.

[0127] (5) Reasoning stage: In actual application, for new children's text content, the system first generates a feature vector x and a suitability score Score through the text feature extraction module and the suitability scoring module, and then generates an input vector according to the above feature fusion steps. Through the trained Softmax classifier, the probability distribution of each age level is quickly calculated, and the optimal age-appropriate level is output. .

[0128] (6) Prediction process: The probability distribution of each age category is output through the Softmax classifier, and the age category with the highest probability is selected as the prediction result. The specific steps include: fusing the suitability score and the feature vector into a comprehensive feature vector, inputting it into the Softmax classifier, generating a probability distribution, and taking the maximum value.

[0129] Another embodiment is used to illustrate an artificial intelligence-based text rating system for children, see Figure 3 , the system 300 includes:

[0130] A preprocessing module 310 is used to preprocess the text content to be evaluated to generate a standardized text representation, wherein the preprocessing includes text segmentation, stop word filtering, and word form normalization;

[0131] The feature extraction module 320 is used to generate a semantic representation of the text using a pre-trained model for the standardized text representation, and extract a multi-dimensional feature vector based on the semantic representation, wherein the multi-dimensional feature vector includes topic classification, sentiment tendency, and language complexity features. The method of extracting the multi-dimensional feature vector from the standardized text representation using the pre-trained model specifically includes: based on the input of the standardized text representation, using the pre-trained BERT model to obtain a multi-dimensional vector of CLS tags and obtain a topic classification feature vector; based on the input of the standardized text representation, using the pre-trained BERT model for encoding, generating a multi-dimensional semantic representation for each token to obtain sentiment tendency confidence; based on the input of the standardized text representation, using the pre-trained BERT model for encoding, generating a multi-dimensional semantic for each token, adding a dual affine attention layer, predicting the dependency between each token, and quantifying the language complexity to obtain language complexity features.

[0132] A content suitability scoring module 330 is configured to input the multi-dimensional feature vector into a multi-layer perceptron, which processes the nonlinear mapping of the multi-dimensional feature vector using a Sigmoid function to generate a content suitability score ranging from 0 to 100;

[0133] The age grading module 340 is used to normalize the content suitability score, fuse it with the multi-dimensional feature vector, and input it into the Softmax classifier. Finally, the trained Softmax classifier outputs the age-appropriate grading label prediction result.

[0134] In addition to the above modules, the system 300 may also include other components. However, since these components are irrelevant to the content of the embodiment of the present disclosure, their illustration and description are omitted here.

[0135] For other specific working processes of the artificial intelligence-based text grading system 300 for children, please refer to the description of the embodiment of the artificial intelligence-based text grading method for children, which will not be repeated here.

[0136] Another embodiment is used to illustrate that the system of the present invention can also be used with the help of Figure 4 The architecture of the computing device shown is implemented. Figure 4 The architecture of the computing device is shown in FIG. Figure 4 As shown, a computer system 410, a system bus 430, one or more CPUs 440, an input / output 420, a memory 450, etc. The memory 450 can store various data or files used for computer processing and / or communication, as well as program instructions executed by the CPU including the embodiment of the artificial intelligence-based text grading method for children. Figure 4 The architecture shown is only exemplary and may be adjusted based on actual needs when implementing different devices. Figure 4 One or more components in. The memory 450, as a computer-readable storage medium, can be used to store software programs, computer executable programs and modules, such as the program instructions / modules corresponding to the artificial intelligence-based child-oriented text grading method in the embodiment of the present invention (for example, the preprocessing module 310, feature extraction module 320, content suitability scoring module 330 and age grading module 340 in the artificial intelligence-based child-oriented text grading system 300). One or more CPUs 440 execute various functional applications and data processing of the system of the present invention by running the software programs, instructions and modules stored in the memory 450, that is, to implement the above-mentioned artificial intelligence-based child-oriented text grading method, which includes the following steps:

[0137] Preprocessing the text content to be evaluated to generate a standardized text representation, wherein the preprocessing includes text segmentation, stop word filtering, and word form normalization;

[0138] The standardized text representation is represented by using a pre-trained model to generate a semantic representation of the text, and a multi-dimensional feature vector is extracted based on the semantic representation, wherein the multi-dimensional feature vector includes topic classification, sentiment tendency and language complexity features; the standardized text representation is represented by using a pre-trained model to extract a multi-dimensional feature vector, specifically comprising: based on the input of the standardized text representation, using a pre-trained BERT model to obtain a multi-dimensional vector of CLS tags and obtain a topic classification feature vector; based on the input of the standardized text representation, using a pre-trained BERT model for encoding, generating a multi-dimensional semantic representation for each token to obtain sentiment tendency confidence; based on the input of the standardized text representation, using a pre-trained BERT model for encoding, generating a multi-dimensional semantic for each token, adding a dual affine attention layer, predicting the dependency between each token, quantifying the language complexity to obtain language complexity features;

[0139] Inputting the multi-dimensional feature vector into a multi-layer perceptron, wherein the multi-layer perceptron processes the nonlinear mapping of the multi-dimensional feature vector through a Sigmoid function to generate a content suitability score from 0 to 100;

[0140] The content suitability score is normalized, fused with the multi-dimensional feature vector, and then input into the Softmax classifier. Finally, the trained Softmax classifier outputs the age-appropriate classification label prediction result.

[0141] Of course, the processor of the server provided in the embodiment of the present invention is not limited to executing the method operations described above, but can also execute relevant operations in the artificial intelligence-based text grading method for children provided in any embodiment of the present invention.

[0142] The memory 450 may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system and applications required for at least one function; the data storage area may store data created based on the use of the terminal, etc. In addition, the memory 450 may include a high-speed random access memory and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device. In some instances, the memory 450 may further include a memory remotely located relative to one or more CPUs 440, and these remote memories may be connected to the device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0143] The input / output 420 may be used to receive input digital or character information and generate key signal input related to user settings and function control of the device. The input / output 420 may also include a display device such as a display screen.

[0144] Embodiments of the present invention also provide a non-transitory computer-readable storage medium storing a computer program that, when executed by a processor, implements the artificial intelligence-based text grading method for children described in the above embodiments. The computer-readable storage medium of the embodiments of the present invention may be any combination of one or more computer-readable media. The computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples (a non-exhaustive list) of computer-readable storage media include: an electrical connection having one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In this document, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0145] A computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, which carries computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium that can transmit, propagate, or transport a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0146] The program code contained on the storage medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.

[0147] In addition, other specific working processes of a non-transitory computer-readable storage medium refer to the description of the above-mentioned embodiment of the text grading method for children based on artificial intelligence, and will not be repeated here.

[0148] To verify the effectiveness of the method, system, electronic device, and storage medium described in this paper, multiple rounds of experiments were conducted on a diverse, annotated dataset consisting of 500 publicly available children's books and 500 online-generated content. The dataset covers a variety of topics, including education, entertainment, adventure, science fiction, history, and literature, ensuring the model's robustness across diverse content types. The dataset was annotated by child education experts based on Piagetian theory of cognitive development. The annotations include manual suitability scores (0-100) and age-appropriate classifications (0-3 years, 3-6 years, 7-12 years, etc.). The experiments employed 5-fold cross-validation and introduced multiple evaluation metrics and baseline methods for comparison to enhance the scientific and persuasive nature of the validation.

[0149] The experimental setup includes:

[0150] (1) Dataset division: The dataset is divided into training set, validation set and test set in a ratio of 8:1:1.

[0151] (2) The model includes: topic classification, sentiment analysis, suitability scoring, and age-appropriate grading.

[0152] (3) Baseline: The performance of human experts is used as a benchmark.

[0153] Evaluation metrics include:

[0154] (1) Topic classification: accuracy;

[0155] (2) Sentiment analysis: F1 score;

[0156] (3) Suitability score: Pearson correlation coefficient;

[0157] (4) Age-appropriate classification: Accuracy;

[0158] Experimental results:

[0159] Table 1 shows the comparison data (accuracy comparison) between this method and the manual benchmark. After adjustment, the performance of this method is slightly higher than the manual benchmark.

[0160] Table 1 Comparison data between this method and manual benchmark

[0161]

[0162] Experimental results show that on a dataset of 1,000 articles, this method outperforms the baseline method in topic classification, sentiment analysis, suitability scoring, and age-appropriate grading tasks, validating its technical advantages and practicality.

[0163] This invention achieves comprehensive content evaluation and age-appropriate grading of children's texts through multi-dimensional feature extraction, dynamic fusion of deep neural networks, and refined grading using probabilistic models. Its main technical innovations include:

[0164] Multi-level feature fusion: Combining statistical, semantic, and knowledge-driven methods to capture the deep characteristics of text.

[0165] Dynamic scoring mechanism: Utilizes the self-attention mechanism to adapt to the complex interactive relationships between different texts.

[0166] Cognitive science constraints: Incorporating the laws of children's development to improve the scientific nature of grading.

[0167] The above content is only one of the specific implementation methods of the technical solution of the present invention. Its detailed description is intended to explain the technical principles, algorithm architecture and application scenarios of the present invention, rather than to fully limit the scope of protection of the present invention. After mastering the technical ideas and method framework disclosed in the present invention, those skilled in the art can, without deviating from the core spirit and technical purpose of the present invention, combine their own professional knowledge and application needs to make various adjustments, optimizations or equivalent replacements to the technical details, parameter configurations, algorithm combinations or processing procedures in the above implementation methods. Any variation derived from the technical solution of the present invention, including but not limited to the improvement of feature extraction strategies, variant designs of suitability scoring models, and alternative implementations of age grading algorithms, should be regarded as a natural extension and coverage of the scope of protection of the present invention. Therefore, the actual scope of protection of the present invention should be based on the technical features and their equivalent scope clearly defined in the claims, rather than being limited to the specific embodiments described herein. The present invention provides an innovative solution for text content evaluation and grading for children through the complex fusion of multi-dimensional feature extraction and deep learning technology. Its technical value and application potential have significant scalability and universality in this field. It can be widely used in children's digital reading materials, online education platforms, game content review and other fields, providing strong technical support for protecting children from the potential harm of inappropriate content.

[0168] In this document, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a step or method that comprises a series of elements includes not only those elements, but also includes other elements not expressly listed, or also includes elements inherent to such step or method.

[0169] The above is a further detailed description of the present invention in conjunction with specific preferred embodiments, and the specific implementation of the present invention should not be considered to be limited to these descriptions. For those skilled in the art of the present invention, without departing from the concept of the present invention, several simple deductions or substitutions can be made, which should be considered to fall within the scope of protection of the present invention.

Claims

1. A text grading method for children based on artificial intelligence, characterized in that: The method comprises the following steps: Preprocessing the text content to be evaluated to generate a standardized text representation, wherein the preprocessing includes text segmentation, stop word filtering, and word form normalization; The standardized text representation is represented by using a pre-trained model to generate a semantic representation of the text, and a multi-dimensional feature vector is extracted based on the semantic representation, wherein the multi-dimensional feature vector includes topic classification, sentiment tendency and language complexity features; the standardized text representation is represented by using a pre-trained model to extract a multi-dimensional feature vector, specifically comprising: based on the input of the standardized text representation, using a pre-trained BERT model to obtain a multi-dimensional vector of CLS tags and obtain a topic classification feature vector; based on the input of the standardized text representation, using a pre-trained BERT model for encoding, generating a multi-dimensional semantic representation for each token to obtain sentiment tendency confidence; based on the input of the standardized text representation, using a pre-trained BERT model for encoding, generating a multi-dimensional semantic for each token, adding a dual affine attention layer, predicting the dependency between each token, quantifying the language complexity to obtain language complexity features; Inputting the multi-dimensional feature vector into a multi-layer perceptron, wherein the multi-layer perceptron processes the nonlinear mapping of the multi-dimensional feature vector through a Sigmoid function to generate a content suitability score from 0 to 100; The content suitability score is normalized, fused with the multi-dimensional feature vector, and then input into the Softmax classifier. Finally, the trained Softmax classifier outputs the age-appropriate classification label prediction result; The content suitability score is generated by a multi-layer perceptron based on the multi-dimensional feature vector. The specific expression is: ， in represents the content suitability score, represents the Sigmoid function, is the weight matrix of the neural network, x represents the multi-dimensional feature vector, Represents the bias vector, and ReLU represents the rectified linear unit function.

2. The artificial intelligence-based text grading method for children according to claim 1, characterized in that: The pre-trained model is used to extract multi-dimensional feature vectors from the standardized text representation, including: Based on the input of the standardized text representation, a pre-trained BERT model is used to obtain a 768-dimensional vector of CLS tags, which is used as the feature vector for topic classification and input into a support vector machine. The support vector machine performs topic classification based on the radial basis function. Based on the input of the standardized text representation, the pre-trained BERT model is used for encoding to generate a 768-dimensional semantic representation for each token. A fully connected layer is added to map the 768-dimensional semantic representation to the probability distribution of the sentiment label and output the confidence score through the softmax function. Based on the input of standardized text representation, the pre-trained BERT model is used for encoding, generating a 768-dimensional semantic representation for each token. A dual affine attention layer is added to predict the dependencies between tokens and generate a syntactic tree. The language complexity is quantified based on the topological analysis and information entropy calculation of the syntactic tree. The extracted feature vectors are normalized to ensure that the dimensions of features in different dimensions are consistent.

3. The artificial intelligence-based text grading method for children according to claim 1, characterized in that: The multi-dimensional feature vector also includes social value features, which are extracted through graph embedding technology of knowledge graph.

4. The artificial intelligence-based text grading method for children according to claim 1, characterized in that: A classification model is used to predict age classification based on suitability scores and multi-dimensional feature vectors, including: Fusion of suitability score and feature vector into input vector; Output the probability distribution of each age level through the Softmax classifier; The age category with the highest probability is selected as the prediction result.

5. The artificial intelligence-based text grading method for children according to claim 1, characterized in that: The content suitability scoring process also includes: A multilayer perceptron is pre-trained and its parameters are optimized using a labeled dataset that includes both appropriate and inappropriate text samples for children.

6. The artificial intelligence-based text grading method for children according to claim 4, characterized in that: The age ratings also include: Predefine the cognitive and psychological characteristics of children of different age groups, including vocabulary comprehension and emotional receptivity; Train the Softmax classifier using historical data.

7. A text rating system for children based on artificial intelligence, characterized by: The system comprises: A preprocessing module, configured to preprocess the text content to be evaluated and generate a standardized text representation, wherein the preprocessing includes text segmentation, stop word filtering, and word form normalization; A feature extraction module is used to generate a semantic representation of the text using a pre-trained model for the standardized text representation, and extract a multi-dimensional feature vector based on the semantic representation, wherein the multi-dimensional feature vector includes topic classification, sentiment tendency and language complexity features; extracting a multi-dimensional feature vector from the standardized text representation using a pre-trained model specifically includes: based on the input of the standardized text representation, using a pre-trained BERT model to obtain a multi-dimensional vector of CLS tags and obtain a topic classification feature vector; based on the input of the standardized text representation, using a pre-trained BERT model for encoding, generating a multi-dimensional semantic representation for each token to obtain sentiment tendency confidence; based on the input of the standardized text representation, using a pre-trained BERT model for encoding, generating a multi-dimensional semantic for each token, adding a dual affine attention layer, predicting the dependency between each token, quantifying the language complexity to obtain language complexity features; A content suitability scoring module, configured to input the multi-dimensional feature vector into a multi-layer perceptron, wherein the multi-layer perceptron processes the nonlinear mapping of the multi-dimensional feature vector using a Sigmoid function to generate a content suitability score ranging from 0 to 100; The age rating module is used to normalize the content suitability score, fuse it with the multi-dimensional feature vector, and input it into the Softmax classifier. The trained Softmax classifier finally outputs the predicted age-appropriate rating label. The content suitability score is generated by a multi-layer perceptron based on the multi-dimensional feature vector. The specific expression is: ， in represents the content suitability score, represents the Sigmoid function, is the weight matrix of the neural network, x represents the multi-dimensional feature vector, Represents the bias vector, and ReLU represents the rectified linear unit function.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the steps of the artificial intelligence-based text grading method for children as described in any one of claims 1 to 6 are implemented.

9. A non-transitory computer-readable storage medium having computer instructions stored thereon, characterized in that: When the instructions are executed by the processor, the steps of the artificial intelligence-based text grading method for children as described in any one of claims 1 to 6 are implemented.

Images