An automated assessment method and system for children's digital creativity in graphical programming

Abstract

This invention discloses an automatic assessment method and system for children's digital creativity in graphical programming, belonging to the fields of computer-aided education and educational data mining. The method includes: real-time acquisition and storage of multi-source heterogeneous raw data; extraction of a seven-dimensional standard feature vector, with creative feature sub-dimension values ​​truncated to convergence within an interval using an extreme value threshold; after unified representation of the features, generation of a comprehensive creativity representation vector through a multi-head self-attention mechanism and output of a single-dimensional evaluation result; and output of a comprehensive quantitative score and level using a multilayer perceptron joint prediction network. The system comprises a data acquisition layer, a feature extraction layer, a fusion evaluation layer, a cross-dimensional fusion layer, a comprehensive assessment layer, and a service output layer connected sequentially. This invention achieves automated, objective, and real-time comprehensive assessment of children's digital creativity, providing technical support for teaching evaluation and personalized training.

Description

Technical Field

[0001] This invention relates to the fields of computer-aided education and educational data mining, and in particular to an automatic assessment method and system for children's digital creativity in graphical programming. Background Technology

[0002] Children's graphical programming digital creativity refers to the creative ability demonstrated by children when using graphical programming tools to generate ideas, build programs, solve problems, and express their works. This ability is not only reflected in the novelty of ideas, but also in the digital realization and comprehensive construction of characters, scenes, interactive logic, and the content of the work.

[0003] Current assessments of digital creativity in children's graphical programming primarily rely on general creative strength scales and subjective human scoring, presenting three main challenges: First, the assessment is highly subjective; existing research has not fully integrated the improvement of children's digital literacy with the cultivation of creativity, lacking specific representational models and objective quantitative assessment methods for the core characteristics of digital creativity in graphical programming contexts. Second, the assessment is relatively time-consuming; evaluation methods are mostly based on phased, static, and outcome-oriented assessments, failing to provide dynamic evaluation in real time after learners complete programming tasks. Third, the assessment content is relatively singular; existing methods often rely on only a single type of evaluation evidence, lacking a synergistic and comprehensive analysis of the work's execution results, code construction process, task completion trajectory, and multimodal creative elements. Summary of the Invention

[0004] The purpose of this invention is to propose an automatic assessment method and system for children's digital creativity in graphical programming, which solves the problems of strong subjectivity, poor timeliness and limited content in existing assessment methods, and realizes a multi-dimensional, objective and real-time comprehensive assessment of children's digital creativity.

[0005] To achieve the above objectives, the present invention provides an automatic assessment method for children's digital creativity in graphical programming, comprising the following steps: Step S1: Collect multi-source heterogeneous raw data generated during children's graphical programming process in real time and store it in a preset database; Step S2: Extract features and perform semantic mapping on the heterogeneous raw data in the preset database, and transform it into a standard feature vector set that supports multi-dimensional quantitative evaluation of digital creativity; Step S3: After uniformly representing the standard feature vector set, the deep interaction relationship between features of each dimension is adaptively captured through a multi-head self-attention mechanism to generate a comprehensive creativity representation vector, and the evaluation results of each single dimension of creativity are output simultaneously. Step S4: Input the comprehensive creativity representation vector into the joint prediction network constructed by the multilayer perceptron, and output the comprehensive quantitative score and comprehensive evaluation level of the child's digital creativity.

[0006] Preferably, in step S1, the preset database includes the following data tables: The task order table includes fields such as task number, task description, number of function points, task requirements, and task code. The student programming operation flow table has fields including operation number, task number, student number, timestamp, event type, operation object, operation status, and duration. The project file table includes fields for project number, student number, task number, and project JSON text. The resource association table has fields including resource number, project number, resource name, resource type, and storage path.

[0007] Preferably, in step S2, the standard feature vector set includes feature vectors for problem-solving completeness, programming process fluency, programming logic novelty, character creativity, digital music creativity, scene creativity, and work creativity; wherein the sub-dimension values ​​of the character creativity, digital music creativity, scene creativity, and work creativity feature vectors are truncated using an extreme value threshold formula. converges to interval, These are the original calculated values ​​for the sub-dimensions. The preset maximum threshold for each sub-dimension is used as the basis for the specific extraction method of each feature vector: Problem Solving Integrity Feature Vector Extract the task requirement text set from the task sheet table. Extract the student logic script set from the project file table. Using a pre-trained text embedding model to and Mapped to the same 3D semantic feature space Calculate the feature vector of function points With engineering implementation feature vectors Cosine similarity between The maximum similarity score for each functional point in the student project is obtained, forming a maximum similarity set; the average matching degree feature is calculated based on the maximum similarity set. and minimum matching feature ,in The mean of the maximum similarity across all functionalities represents the overall task completion rate. The minimum value in the similarity set represents the degree of deficiency in problem-solving; ultimately, it generates... ; Programming process fluency feature vector Based on the timing and event data of the student programming operation flow table, count the number of deletion operations. Number of revisions and operation pause time ;in This represents the total number of records for which the event type is "deletion flag". This sets the frequency at which the same operation object experiences two or more consecutive "modify" events within a preset time window. This is the cumulative sum of the timeout durations exceeding a preset normal threshold among the timestamp differences between adjacent valid operations; ultimately generating... ; Programming logic novelty feature vector Parse the JSON text of the project to extract abstract structural information and obtain the characteristics of variable and list usage. , control structure usage characteristics Custom block encapsulation features Using features in complex algorithms ;in , , Obtain directly through parsing using preset rules. The system identifies complex logical processing methods such as counting, sorting, recursion, combination judgment, and multi-structure collaborative control through a pre-trained classification model, and verifies and corrects the results using rule-based validation; ultimately generating... ; Character Creative Feature Vector Extracting the uniqueness of character names Complexity of shape Uniqueness of voice Interaction characteristics between characters Environmental interaction characteristics Functional novelty ;in Subtract the maximum semantic similarity between the character name and the preset common name library and the existing set of character names from 1. The calculation is based on a weighted average of the number of times the character design was drawn, the frequency of modifications, and the degree of difference from the original materials. The calculation is based on a weighted average of the source of the character's voice material, editing behavior, and differences in audio characteristics. This is a normalized value representing the number of messages sent and received between roles. This is a normalized value representing the number of times a response is triggered by the character and the environment. Calculation based on the normalized value of the number of self-made building blocks and newly added modules; final generation ; Digital Music Creative Feature Vector Based on the MusicNet pre-trained model, the pitch, rhythm, timbre, and structural segmentation information of audio are extracted to calculate the melodic creativity. Rhythmic diversity timbre matching Work Structure ;in Calculations based on the number and direction of independent pitches Calculated based on the number of rhythmic patterns and changes in time value. Calculations based on timbre types and combinations Calculations based on the number of audio segments and their organizational hierarchy; ultimately generating... ; Scene creative feature vector Extracting Background Uniqueness Background artistry Value characteristics ;in Weighted synthesis through operational statistical normalization The weighted composition is analyzed using computer vision technology. Obtained through a semi-supervised decision model; finally generated. ; Creative Feature Vector of Works : Quantitatively extracting the creativity of storylines through a semi-supervised learning model Interaction and Gamification Educational value ; final generation .

[0008] Preferably, in step S3, the unified representation processing involves performing dimension alignment, projection mapping, or encoding transformation on standard feature vectors of different lengths and structures to obtain a structurally compatible unified feature representation. , , , , , , ; The results of the single-dimensional creativity assessment are generated through the following steps: Step S31: Single-dimensional weighted processing: Unified feature representation for all features with multi-component structures , , , , , , The weight vectors of each feature component are learned through an adaptive weighted network. According to the formula Element-wise weighting is performed to obtain a one-dimensional enhanced feature representation. ,in , For Hadamah accumulation; Step S32: Unidimensional scoring processing: Represent each unidimensional enhanced feature Input the corresponding multilayer perceptron scoring network According to the formula Output the creativity scores for each individual dimension. ,in All single-dimensional scoring results constitute a single-dimensional evaluation result set. .

[0009] Preferably, in step S3, the process of generating the comprehensive creativity representation vector is as follows: [The text abruptly shifts to a different topic] , , , , , , Concatenate the components in a preset order to construct an initial high-dimensional feature matrix. ;Will Input the multi-head self-attention layer, according to the formula Generate a comprehensive creativity representation vector ,in This is the multi-head self-attention calculation function.

[0010] Preferably, in step S4, the processing procedure of the joint prediction network is as follows: The integrated creativity representation vector... Input shared hidden layer To obtain high-level abstract feature representation ; predict branches through regression according to the formula Output mapping to Comprehensive quantitative score of the interval ,in This is a regression mapping function; the classification prediction branch is calculated according to the formula. Output the probability distribution of each preset evaluation level. ,in For classification mapping function, The activation function is used; the level corresponding to the highest probability in the probability distribution is selected as the comprehensive evaluation level. .

[0011] An automatic assessment system for digital creativity in children's graphical programming includes a data acquisition layer, a feature extraction layer, a fusion evaluation layer, a cross-dimensional fusion layer, a comprehensive evaluation layer, and a service output layer that are connected in sequence via communication. The data acquisition layer is configured to collect multi-source heterogeneous raw data generated during children's graphical programming process in real time and store it in a preset database. The feature extraction layer is configured to extract features and perform semantic mapping on heterogeneous raw data in a preset database, transforming them into a standard feature vector set that supports multi-dimensional quantitative evaluation of digital creativity. The integrated evaluation layer is configured to perform unified representation processing on the standard feature vector set and output the evaluation results of each single dimension of creativity. A cross-dimensional fusion layer is configured to adaptively capture deep interaction relationships between features of various dimensions through a multi-head self-attention mechanism, generating a comprehensive creativity representation vector. The comprehensive assessment layer is configured to input the comprehensive creativity representation vector into a joint prediction network constructed by a multilayer perceptron, and output a comprehensive quantitative score and comprehensive evaluation level of children's digital creativity. The service output layer is configured to output single-dimensional creativity evaluation results, comprehensive quantitative scores, and comprehensive evaluation levels.

[0012] Preferably, the data acquisition layer is configured to collect task sheet text, student programming operation flow data, project JSON files, audio resource files, and scene image data; the preset database includes a task sheet table, a student programming operation flow table, a project file table, and a resource association table; The feature extraction layer is configured to extract feature vectors for problem-solving integrity, programming process fluency, programming logic novelty, character creativity, digital music creativity, scene creativity, and work creativity. The complex algorithm for the programming logic novelty feature vector uses a combination mechanism of "pre-trained classification model + rule verification" to identify features. The digital music creativity feature vector is extracted based on the MusicNet pre-trained model. The scene creativity feature vector is generated by fusing operational statistics, computer vision, and semi-supervised judgment models. The work creativity feature vector is generated based on a semi-supervised learning judgment model.

[0013] Preferably, the fusion evaluation layer includes a feature unified representation module and a single-dimensional evaluation module; The unified feature representation module is configured to perform dimension alignment, projection mapping, or encoding transformation on standard feature vectors of different lengths and structures to obtain a structurally compatible unified feature representation. The single-dimensional evaluation module is configured to perform single-dimensional weighted processing and single-dimensional scoring processing on the unified feature representation, and output the creativity score results for each single dimension. The single-dimensional weighted processing learns the weights of each feature component through an adaptive weighted network and weights them element by element. The single-dimensional scoring processing establishes a non-linear mapping between the enhanced features and the scoring results through a multilayer perceptron. The cross-dimensional fusion layer configuration is used to concatenate the unified feature representations of each dimension into an initial high-dimensional feature matrix. The attention weights between features of different dimensions are calculated through a multi-head self-attention layer to mine cross-dimensional implicit interaction features and generate a comprehensive creativity representation vector.

[0014] Preferably, the joint prediction network of the comprehensive evaluation layer includes a shared hidden layer, a regression prediction branch, and a classification prediction branch; A shared hidden layer is configured to perform non-linear mapping and deep semantic decoding on the comprehensive creativity representation vector, outputting a high-level abstract feature representation. The regression prediction branch is configured to perform continuous value mapping on high-level abstract feature representations, and the output is mapped to... The comprehensive quantitative score of the interval; The classification prediction branch is configured to perform level discrimination on high-level abstract feature representations. It outputs the probability distribution of each preset evaluation level through the Softmax activation function and selects the level corresponding to the highest probability as the comprehensive evaluation level. The service output layer configuration is used to output the single-dimensional creativity score results, comprehensive quantitative score, comprehensive evaluation level and corresponding assessment report in the form of a visual interface display or API interface call.

[0015] Therefore, the present invention employs the above-described automatic assessment method and system for children's graphical programming digital creativity, which has the following advantages: (1) Objective evaluation: Based on multi-source heterogeneous data, quantifiable features are extracted to avoid the bias of subjective scoring by humans. Objective quantitative evaluation of each dimension is achieved through semantic similarity calculation, behavioral data statistics, deep learning model recognition and other technologies. (2) Real-time assessment: Real-time collection of children's operation flow, project files, resource data, etc. during the programming process. The assessment results can be automatically generated after the task is completed, supporting dynamic process evaluation and instant teaching feedback; (3) Comprehensive evaluation: It covers seven core dimensions: completeness of problem-solving, fluency of programming process, novelty of logic, creativity of characters, creativity of music, creativity of scene and creativity of work, and integrates multimodal data such as code, behavior, audio and images to achieve a comprehensive evaluation of digital creativity; (4) Intelligent evaluation: The multi-head self-attention mechanism is used to adaptively capture the deep interaction relationship between features in each dimension. The nonlinear mapping relationship is established through the multilayer perceptron, and the feature weights and evaluation criteria are automatically learned, which significantly improves the evaluation accuracy.

[0016] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0017] Figure 1This is an overall flowchart of an automatic assessment method for children's digital creativity in graphical programming, as described in an embodiment of the present invention. Figure 2 This is a module structure diagram of an automatic assessment system for children's digital creativity in graphical programming, as described in an embodiment of the present invention. Figure 3 This is a preset database in the embodiments of the present invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0019] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0020] Example like Figure 1-3 As shown in the figure, this embodiment proposes an automatic assessment method for children's digital creativity in graphical programming. The specific implementation steps are as follows: Step S1: Collect raw data generated during children's graphical programming in real time and store it in the database.

[0021] The database design is as follows: 1. TaskSheet: Contains fields: Task ID (Tid), Task Description (Description), Number of Functions (Function), Task Requirements (Requirements), and Task Code (Code); i.e., TaskSheet: {Tid;Description; Function; Requirements; Code}.

[0022] 2. Student Programming Operation Flow Table (StudentFlow): Contains fields: Operation ID (Fid), Task ID (Tid), Student ID (Sid), Timestamp (Timestamp), Event Type (Event), Operation Object (Object), Operation Status (Status), Duration (Duration); that is: StudentFlow: {Fid; Tid; Sid; Timestamp; Event; Object; Status; Duration}.

[0023] 3. ProjectFile table: Contains fields: Project ID (Pid), Student ID (Sid), Task ID (Tid), Project JSON text (Pjson); that is: ProjectFile: {Pid; Sid; Tid; Pjson}.

[0024] 4. Resource Association Table (ProjectAssests): Contains fields: Resource ID (Aid), Project ID (Pid), Resource Name (Aname), Resource Type (Atype), and Storage Path (Furl); that is: ProjectAssests: {Aid; Pid; Aname; Atype; Furl}.

[0025] Step S2: By extracting and mapping features from the heterogeneous raw data in the database, the data is transformed into standard feature vectors that support quantitative evaluation of various dimensions of creativity.

[0026] 1. Problem-solving completeness: The system extracts the task requirement text sets from both the task sheet and the project file. Collection of student logic scripts And use a pre-trained text embedding model to map it to the same 3D semantic feature space In the middle. Then, by calculating the function point feature vector. With engineering implementation feature vectors Cosine similarity between The system obtains the maximum similarity set achievable by each functional point in the student project. Based on this similarity set, the system extracts features from two core dimensions: average matching degree feature. and minimum matching degree feature Regarding the specific extraction method, It is calculated from the mean of the maximum similarity of all functional points and represents the overall task completion rate; The minimum value in the similarity set is used to characterize the degree of weakness or deficiency in students' problem-solving process. Finally, the system generates a multi-dimensional feature vector to characterize the completeness of problem-solving. .

[0027] 2. Programming Process Fluency: By analyzing the raw student programming behavior sequence data, process feature information to characterize the fluency of the programming process is extracted. Based on the time series and event data in StudentFlow, the system extracts process features reflecting students' cognitive load and trial-and-error costs through rule-based statistical methods. Specifically, the system extracts three key indicators: the number of deletion operations... (Directly count the total number of records where the Event field has the value "deletion flag") Number of repeated modifications (Use Timestamp to define a time window and count the frequency of "modify" events occurring twice or more consecutively on the same object), as well as the operation pause time. (Calculate the timestamp difference between adjacent valid operations, and sum the abnormal duration of no operations exceeding the normal threshold). The final constructed fluency feature vector. The number of deletion operations, the number of repeated modifications, and the duration of pauses are used to characterize the lag, trial and error, and interruption characteristics in the student's programming process, and serve as input for the subsequent fluency scoring module.

[0028] 3. Novelty of Programming Logic: Regarding the novelty of the programming logic, the system parses the abstract structure information of the program based on the JSON text (Pjson) of the student project files, extracting features from four core dimensions: variable and list usage characteristics. , control structure usage characteristics Custom block encapsulation features and features used in complex algorithms Regarding the specific extraction method, The three shallow structural features are obtained directly through pre-defined rule parsing; while the deep, complex algorithmic features... The system employs a combined mechanism of "artificial intelligence model + rule verification." Specifically, structural nodes and their combined relationships are input into a pre-trained classification model or supervised learning model to identify whether they constitute a preset complex logical processing method. This complex logical processing method includes, but is not limited to, counting, sorting, recursion, combination judgment, or multi-structure collaborative control. Subsequently, the identification results are compared with the rule parsing results to achieve supplementary identification, result verification, or conflict correction. Finally, the system generates feature vectors to characterize the novelty of the programming logic. .

[0029] 4. Character Creativity: The system extracts multimodal data of characters from project files and operation sequences, and performs quantitative extraction in six sub-dimensions: Based on the semantic similarity between the character name and a preset common name library or existing character name set, the system calculates the uniqueness features of the character name. The lower the similarity, the higher the uniqueness of the name; based on the number of times the character design was drawn, the frequency of modification, and the degree of difference between the character design and the original materials, the complexity features of the design were extracted after normalization and weighting. Based on the differences in the source of the character's voice materials, editing behavior, and audio characteristics, unique voice features were extracted after normalization and weighting. The number of messages sent and received between roles was counted and normalized to extract interaction features between roles. ; Statistically analyze and normalize the number of times a character triggers and responds to the background, UI controls, or environmental objects in order to extract environmental interaction features. Finally, the novelty of the character-specific functions was calculated by combining the normalized index of the number of self-made building blocks and newly added modules. The calculation of all the above indicators incorporates a maximum value threshold truncation. With the weighting synthesis mechanism, it is ensured that each sub-eigenvalue strictly converges to... The intervals are used to ultimately construct feature vectors that represent the overall creativity of the character. .

[0030] 5. Creativity in Digital Music: The system inputs collected engineering audio files and operation sequences into a music feature analysis model pre-trained on MusicNet to extract low-level information such as pitch, rhythm, timbre, and structural segmentation. Subsequently, extreme value thresholding is used for truncation. With the weighting allocation mechanism, the melodic creativity composed of the number of independent pitches and the changes in direction is calculated respectively. Rhythmic diversity is constituted by the number of rhythmic patterns and changes in time value. The combination of timbre types and combinations And the structure of the work, which is composed of the number of fragments and the organizational hierarchy. Ultimately, each sub-feature converges strictly to... Intervals for constructing creative feature vectors .

[0031] 6. Scene Creativity: The system integrates operational statistics, visual analysis, and deep learning models for evaluation. First, it normalizes and weights the uniqueness of the synthesized background. Secondly, the background artistry is analyzed and weighted using computer vision technology. Finally, a semi-supervised decision model is used to obtain value features. Ultimately, a scene creativity feature vector is constructed. .

[0032] 7. Creativity of the work: The system deeply analyzes the global semantics of the project files and uses a semi-supervised learning-based judgment model to quantify and extract the creativity of the storyline. Interaction and Gamification and educational value Three normalized features. Then, a creative feature vector of the work is constructed. .

[0033] Step S3: Using a multi-head self-attention mechanism, the extracted seven-dimensional feature vectors are concatenated and input into the multi-head self-attention layer to adaptively capture deep interaction features across dimensions, constructing a comprehensive creativity representation vector with global perception capabilities. And simultaneously output the unit evaluation results.

[0034] Obtain the normalized features of the seven dimensions extracted in step S2, including: problem-solving completeness score. Flow vector Logical novelty vector Character creative vector Musical creative vector Scene creative vector and the creative vector of the work Since the aforementioned features differ in data source, vector length, representation structure, and representation granularity, they are considered heterogeneous features. Therefore, a unified representation processing is first performed on the features of each dimension. This unified representation processing includes: performing dimension alignment, projection mapping, or encoding transformation on normalized features of different lengths and structures to unify the features of each dimension into a structurally compatible standard representation, thereby meeting the requirements of the subsequent single-dimensional scoring module and cross-dimensional fusion module for consistency in input format. After the unified representation processing, the problem-solving completeness feature representations are obtained respectively. Programming process fluency feature vector Programming logic novelty feature vector Character creative feature vector Digital music creative feature vector Scene creative feature vector and the creative characteristic vector of the work .

[0035] After completing the unified representation processing, step S3 includes module A, module B, and module C.

[0036] Module A: Single-dimensional weighted module: The system inputs the features of each dimension after unified representation processing into the single-dimensional weighting module. For multi-component features such as programming process fluency, programming logic novelty, character creativity, digital music creativity, scene creativity, and work creativity, the system uses an artificial intelligence model to automatically learn the weights of each feature component based on training samples, and then performs element-wise weighting on the feature vectors of each dimension to obtain the corresponding single-dimensional enhanced feature representation. The weighting process can be uniformly represented as follows: The weight vector It is learned from the corresponding adaptive weighted network. For problem-solving completeness, due to its feature representation... Since it is a single scalar, no internal weighting is performed; instead, it is directly used as the enhanced feature representation for that dimension, denoted as... Finally, the system outputs the single-dimensional enhanced feature representations corresponding to each dimension. This is used for subsequent unidimensional scoring and cross-dimensional fusion processing.

[0037] Module B: Unidimensional Scoring Module The system inputs the enhanced feature representations of each dimension output by the single-dimensional weighted module into the corresponding multilayer perceptron scoring module, establishes a non-linear mapping relationship between the single-dimensional enhanced features and the single-dimensional creativity scoring results, and outputs the creativity scoring results for each dimension. Specifically, the problem-solving completeness enhanced feature representation... Enhanced fluency features in the programming process Novelty Enhancement Feature Representation of Programming Logic Character creativity enhancement features Digital music creative enhancement features Scene Creativity Enhancement Feature Representation and the characteristics of enhanced creativity in the work. The corresponding multilayer perceptron scoring network is input into each network, and the results of the creativity scores for each single dimension are output. , , , , , , The scoring process can be represented as follows: Ultimately, the system obtains a set of one-dimensional evaluation results. This is used to characterize children's specific performance levels in dimensions such as problem-solving completeness, programming process fluency, programming logic novelty, character creativity, digital music creativity, scene creativity, and work creativity. Preferably, the creativity scores for each single dimension are continuous scores; furthermore, the continuous scores can be mapped to the corresponding single-dimensional creativity level results according to a preset piecewise function or a threshold division rule learned from training samples.

[0038] Module C: Cross-dimensional Fusion Module The system will complete the unified representation of seven dimensional features: problem-solving completeness feature vector, programming process fluency feature vector, programming logic novelty feature vector, character creativity feature vector, digital music creativity feature vector, scene creativity feature vector, and work creativity feature vector. These features will be concatenated in a preset order to construct an initial high-dimensional feature matrix representing the multidimensional creativity state of a child's single programming project. Its construction process can be represented as: Considering the complex nonlinear synergistic relationships among the various dimensions of digital creativity, the system will use the initial high-dimensional feature matrix... By inputting a multi-head self-attention layer, attention weights are calculated between features of different dimensions. The interaction relationships between dimensions are adaptively learned, and implicit cross-dimensional interaction features are mined to directly generate a comprehensive creativity representation vector. It is used to characterize the overall status and potential correlation features of children's single programming projects across multiple dimensions of digital creativity, and serves as the input basis for subsequent comprehensive feature enhancement processing and overall creativity evaluation.

[0039] Step S4: Use a multilayer perceptron to conduct a comprehensive rating and quantitative score of digital creativity.

[0040] The system will output the comprehensive creativity representation vector from step S3. The input is a joint prediction network built on a multilayer perceptron. The comprehensive feature vector is nonlinearly mapped and deeply semantically decoded through several hidden layers to obtain a high-level abstract feature representation of the shared hidden layer output. Subsequently, the system performs regression prediction and classification prediction based on the high-level abstract feature representation: on the one hand, it utilizes the regression prediction branch to... Perform continuous value mapping to output a comprehensive quantitative score of children's numerical creativity. The process of mapping to a continuous interval from 0 to 100 can be represented as follows: On the other hand, using classification prediction branches To determine the level, first, the response for each level is obtained through classification mapping, and then the probability distribution of children at each preset digital creativity assessment level is output through the Softmax activation function. The process can be represented as Based on this, the system selects a probability distribution. The grade corresponding to the item with the highest probability will be used as the final comprehensive evaluation grade. , This allows for a comprehensive quantitative scoring and rating of children's digital creativity levels.

[0041] The symbol mapping tables for the various dimensions of creativity and the symbol mapping tables for system processing and output are shown in Tables 1 and 2 below: Table 1. Comparison of Symbols for Each Dimension of Creativity

[0042] Table 2 System Processing and Output Symbol Comparison Table

[0043] like Figure 2 As shown, an automatic evaluation system for digital creativity in children's graphical programming is used to implement an automatic evaluation method for digital creativity in children's graphical programming, including: a data acquisition layer, a feature extraction layer, a fusion evaluation layer, a cross-dimensional fusion layer, a comprehensive evaluation layer, and a service output layer.

[0044] The data acquisition layer is used to acquire raw data related to children's graphical programming works; the feature extraction layer is used to extract feature information representing various dimensions of digital creativity from the raw data; the fusion evaluation layer is used to perform unified representation processing on heterogeneous features and complete single-dimensional creativity evaluation; the cross-dimensional fusion layer is used to splice features from various dimensions and learn the correlation between different dimensions through cross-dimensional interactive modeling to generate a comprehensive creativity representation vector; the comprehensive evaluation layer is used to output a comprehensive quantitative score and level evaluation result based on the comprehensive creativity representation result; and the service output layer is used to output the single-dimensional evaluation result, comprehensive quantitative score, and comprehensive evaluation level of children's digital creativity.

[0045] The data acquisition module (data acquisition layer) is configured to acquire raw data related to children's graphical programming works from multi-source heterogeneous data such as task sheet text, project files, programming behavior sequences, audio files and scene images; A multidimensional feature extraction module (feature extraction layer) is configured to extract multidimensional features representing children's digital creativity based on the original data. The multidimensional features include problem-solving completeness features, programming process fluency features, programming logic novelty features, character creativity features, digital music creativity features, scene creativity features, and work creativity features. The unified feature representation module (fusion evaluation layer) is configured to perform dimensional alignment, projection mapping, or encoding transformation on heterogeneous features from different sources, with different lengths, and with different structures, in order to obtain a structurally compatible unified feature representation. The single-dimensional evaluation module (fusion evaluation layer) is configured to perform single-dimensional enhancement and single-dimensional scoring on the unified feature representation of each dimension, and output the creativity evaluation results of each single dimension, such as problem-solving completeness, programming process fluency, programming logic novelty, character creativity, digital music creativity, scene creativity, and work creativity. The cross-dimensional fusion module (cross-dimensional fusion layer) is configured to concatenate unified feature representations of various dimensions and learn the interaction relationship between features of different dimensions based on a multi-head self-attention mechanism to generate a comprehensive creativity representation vector. The comprehensive assessment module (comprehensive assessment layer) is configured to perform comprehensive quantitative scoring and level assessment based on the comprehensive creativity representation vector through a joint prediction network, and output the comprehensive score and comprehensive evaluation level of the child's digital creativity. The results output module (service output layer) is configured to output, display, or call interfaces for single-dimensional evaluation results, comprehensive quantitative scores, and comprehensive evaluation levels.

[0046] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. An automatic assessment method for children's digital creativity in graphical programming, characterized in that, Includes the following steps: Step S1: Collect multi-source heterogeneous raw data generated during children's graphical programming process in real time and store it in a preset database; Step S2: Extract features and perform semantic mapping on the heterogeneous raw data in the preset database, and transform it into a standard feature vector set that supports multi-dimensional quantitative evaluation of digital creativity; Step S3: After uniformly representing the standard feature vector set, the deep interaction relationship between features of each dimension is adaptively captured through a multi-head self-attention mechanism to generate a comprehensive creativity representation vector, and the evaluation results of each single dimension of creativity are output simultaneously. Step S4: Input the comprehensive creativity representation vector into the joint prediction network constructed by the multilayer perceptron, and output the comprehensive quantitative score and comprehensive evaluation level of the child's digital creativity.

2. The automatic assessment method for digital creativity in children's graphical programming according to claim 1, characterized in that: In step S1, the preset database contains the following data tables: The task order table includes fields such as task number, task description, number of function points, task requirements, and task code. The student programming operation flow table has fields including operation number, task number, student number, timestamp, event type, operation object, operation status, and duration. The project file table includes fields for project number, student number, task number, and project JSON text. The resource association table has fields including resource number, project number, resource name, resource type, and storage path.

3. The automatic assessment method for digital creativity in children's graphical programming according to claim 1, characterized in that: In step S2, the standard feature vector set includes feature vectors for problem-solving completeness, programming process fluency, programming logic novelty, character creativity, digital music creativity, scene creativity, and work creativity; among them, the sub-dimension values ​​of the character creativity, digital music creativity, scene creativity, and work creativity feature vectors are truncated using an extreme value threshold formula. converges to interval, These are the original calculated values ​​for the sub-dimensions. The preset maximum threshold for each sub-dimension is used as the basis for the specific extraction method of each feature vector: Problem Solving Integrity Feature Vector Extract the task requirement text set from the task sheet table. Extract the student logic script set from the project file table. Using a pre-trained text embedding model to and Mapped to the same 3D semantic feature space Calculate the feature vector of function points With engineering implementation feature vectors Cosine similarity between The maximum similarity score for each functional point in the student project is obtained, forming a maximum similarity set; the average matching degree feature is calculated based on the maximum similarity set. and minimum matching feature ,in The mean of the maximum similarity across all functionalities represents the overall task completion rate. The minimum value in the similarity set represents the degree of deficiency in problem-solving; ultimately, it generates... ; Programming process fluency feature vector Based on the timing and event data of the student programming operation flow table, count the number of deletion operations. Number of revisions and operation pause time ;in This represents the total number of records for which the event type is "deletion flag". This sets the frequency at which the same operation object experiences two or more consecutive "modify" events within a preset time window. This is the cumulative sum of the timeout durations exceeding a preset normal threshold among the timestamp differences between adjacent valid operations; ultimately generating... ; Programming logic novelty feature vector Parse the JSON text of the project to extract abstract structural information and obtain the characteristics of variable and list usage. Control structure usage characteristics Custom block encapsulation features Using features in complex algorithms ;in , , Obtain directly through parsing using preset rules. The system identifies complex logical processing methods such as counting, sorting, recursion, combination judgment, and multi-structure collaborative control through a pre-trained classification model, and verifies and corrects the results using rule-based validation; ultimately generating... ; Character Creative Feature Vector Extracting the uniqueness of character names Complexity of shape Uniqueness of voice Interaction characteristics between characters Environmental interaction characteristics Functional novelty ;in Subtract the maximum semantic similarity between the character name and the preset common name library and the existing set of character names from 1. The calculation is based on a weighted average of the number of times the character design was drawn, the frequency of modifications, and the degree of difference from the original materials. The calculation is based on a weighted average of the source of the character's voice material, editing behavior, and differences in audio characteristics. This is a normalized value representing the number of messages sent and received between roles. This is a normalized value representing the number of times a response is triggered by the character and the environment. Calculation based on the normalized value of the number of self-made building blocks and newly added modules; final generation ; Digital Music Creative Feature Vector Based on the MusicNet pre-trained model, the pitch, rhythm, timbre, and structural segmentation information of audio are extracted to calculate the melodic creativity. Rhythmic diversity timbre matching Work Structure ;in Calculations based on the number and direction of independent pitches. Calculated based on the number of rhythmic patterns and changes in time value. Calculations based on timbre types and combinations Calculations based on the number of audio segments and their organizational hierarchy; ultimately generating... ; Scene creative feature vector Extracting Background Uniqueness Background artistry Value characteristics ;in Weighted synthesis through operational statistical normalization The weighted composition is analyzed using computer vision technology. Obtained through a semi-supervised decision model; finally generated. ; Creative Feature Vector of Works : Quantitatively extracting the creativity of storylines through a semi-supervised learning model Interaction and Gamification Educational value ; final generation .

4. The automatic assessment method for children's digital creativity in graphical programming according to claim 1, characterized in that: In step S3, the unified representation processing involves performing dimension alignment, projection mapping, or encoding transformation on standard feature vectors of different lengths and structures to obtain a structurally compatible unified feature representation. , , , , , , ; The results of the single-dimensional creativity assessment are generated through the following steps: Step S31: Single-dimensional weighted processing: Unified feature representation for all features with multi-component structures , , , , , , The weight vectors of each feature component are learned through an adaptive weighted network. According to the formula Element-wise weighting is performed to obtain a one-dimensional enhanced feature representation. ,in , For Hadamah accumulation; Step S32: Unidimensional scoring processing: Represent each unidimensional enhanced feature Input the corresponding multilayer perceptron scoring network According to the formula Output the creativity scores for each individual dimension. ,in All single-dimensional scoring results constitute a single-dimensional evaluation result set. .

5. The automatic assessment method for digital creativity in children's graphical programming according to claim 1, characterized in that: In step S3, the process of generating the comprehensive creativity representation vector is as follows: The unified feature representation... , , , , , , Concatenate the components in a preset order to construct an initial high-dimensional feature matrix. ;Will Input the multi-head self-attention layer, according to the formula Generate a comprehensive creativity representation vector ,in This is the multi-head self-attention calculation function.

6. The automatic assessment method for children's digital creativity in graphical programming according to claim 1, characterized in that: In step S4, the joint prediction network processes the following steps: It integrates the creativity representation vector... Input shared hidden layer To obtain high-level abstract feature representation ; predict branches through regression according to the formula Output mapping to Comprehensive quantitative score of the interval ,in This is a regression mapping function; the classification prediction branch is calculated according to the formula. Output the probability distribution of each preset evaluation level. ,in For classification mapping function, For activation functions; The level corresponding to the highest probability in the probability distribution is selected as the comprehensive evaluation level. .

7. An automatic assessment system for digital creativity in children's graphical programming, used to implement the automatic assessment method for digital creativity in children's graphical programming as described in any one of claims 1-6, characterized in that, It includes a data acquisition layer, a feature extraction layer, a fusion evaluation layer, a cross-dimensional fusion layer, a comprehensive evaluation layer, and a service output layer, which are connected in sequence. The data acquisition layer is configured to collect multi-source heterogeneous raw data generated during children's graphical programming process in real time and store it in a preset database. The feature extraction layer is configured to extract features and perform semantic mapping on heterogeneous raw data in a preset database, transforming them into a standard feature vector set that supports multi-dimensional quantitative evaluation of digital creativity. The integrated evaluation layer is configured to perform unified representation processing on the standard feature vector set and output the evaluation results of each single dimension of creativity. A cross-dimensional fusion layer is configured to adaptively capture deep interaction relationships between features of various dimensions through a multi-head self-attention mechanism, generating a comprehensive creativity representation vector. The comprehensive assessment layer is configured to input the comprehensive creativity representation vector into a joint prediction network constructed by a multilayer perceptron, and output a comprehensive quantitative score and comprehensive evaluation level of children's digital creativity. The service output layer is configured to output single-dimensional creativity evaluation results, comprehensive quantitative scores, and comprehensive evaluation levels.

8. The automatic assessment system for digital creativity in children's graphical programming according to claim 7, characterized in that: The data acquisition layer is configured to collect task sheet text, student programming operation flow data, project JSON files, audio resource files, and scene image data; the preset database includes a task sheet table, a student programming operation flow table, a project file table, and a resource association table; The feature extraction layer is configured to extract feature vectors for problem-solving integrity, programming process fluency, programming logic novelty, character creativity, digital music creativity, scene creativity, and work creativity. The complex algorithm for the programming logic novelty feature vector uses a combination mechanism of "pre-trained classification model + rule verification" to identify features. The digital music creativity feature vector is extracted based on the MusicNet pre-trained model. The scene creativity feature vector is generated by fusing operation statistics, computer vision, and semi-supervised judgment model. The work creativity feature vector is generated based on a semi-supervised learning judgment model.

9. The automatic assessment system for digital creativity in children's graphical programming according to claim 7, characterized in that: The fusion evaluation layer includes a unified feature representation module and a single-dimensional evaluation module; The unified feature representation module is configured to perform dimension alignment, projection mapping, or encoding transformation on standard feature vectors of different lengths and structures to obtain a structurally compatible unified feature representation. The single-dimensional evaluation module is configured to perform single-dimensional weighted processing and single-dimensional scoring processing on the unified feature representation, and output the creativity score results for each single dimension. The single-dimensional weighted processing learns the weights of each feature component through an adaptive weighted network and weights them element by element. The single-dimensional scoring processing establishes a non-linear mapping between the enhanced features and the scoring results through a multilayer perceptron. The cross-dimensional fusion layer configuration is used to concatenate the unified feature representations of each dimension into an initial high-dimensional feature matrix. The attention weights between features of different dimensions are calculated through a multi-head self-attention layer to mine cross-dimensional implicit interaction features and generate a comprehensive creativity representation vector.

10. An automatic assessment system for digital creativity in children's graphical programming according to claim 7, characterized in that: The joint prediction network of the comprehensive evaluation layer includes a shared hidden layer, a regression prediction branch, and a classification prediction branch; A shared hidden layer is configured to perform non-linear mapping and deep semantic decoding on the comprehensive creativity representation vector, outputting a high-level abstract feature representation. The regression prediction branch is configured to perform continuous value mapping on high-level abstract feature representations, and the output is mapped to... The comprehensive quantitative score of the interval; The classification prediction branch is configured to perform level discrimination on high-level abstract feature representations. It outputs the probability distribution of each preset evaluation level through the Softmax activation function and selects the level corresponding to the highest probability as the comprehensive evaluation level. The service output layer configuration is used to output the single-dimensional creativity score results, comprehensive quantitative score, comprehensive evaluation level and corresponding assessment report in the form of a visual interface display or API interface call.

Images