CN120125003A - A large-scale model-based intelligent educational activity management system and method - Google Patents

Abstract

The present invention belongs to the field of information management technology, and discloses a large-model-based intelligent education activity management system and method. The system covers four layers: data collection and access, processing and storage, intelligent analysis and decision-making, and user interaction and application. Educational data is analyzed by integrating transfer learning with large models, activity processes are optimized based on semantic networks and large models, and personalized learning services are provided using emotional interaction and large models. The method is implemented through the steps of data collection, processing, model training and application. Experiments show that it can significantly improve student performance, improve the efficiency of organizing educational activities and student satisfaction, provide strong support for the intelligent management of educational activities, and has good application prospects.

Description

Intelligent education activity management system and method based on large model

Technical Field

The invention relates to the technical field of information management, in particular to an intelligent education activity management system and method based on a large model.

Background

With the continuous advancement of education informatization, the form and content of education activities are increasingly rich and diverse. Traditional educational activity management approaches rely primarily on manual operations and simple information systems, exposing problems in the face of massive educational data and complex educational activity needs. For example, it is difficult to comprehensively and accurately analyze the learning condition of students, personalized education service cannot be provided according to individual differences of the students, the organization and arrangement of education activities lack of scientificity and flexibility, waste of education resources is easily caused, communication interaction between teachers and students is not efficient enough, and improvement of teaching effects is affected.

The advent of large model technology has provided new ideas and methods for solving these problems. The large model has powerful language understanding, generating and knowledge reasoning capacity, and can deeply analyze and mine large-scale education data to provide powerful support for education activity management and decision. However, there are relatively few studies and practices currently applying large model techniques to educational activity management, and no mature systems and methods have been developed.

Disclosure of Invention

The present invention is directed to an educational activity management system based on a large model technique, comprising:

The data acquisition and access layer is used for deploying various data acquisition interfaces for interfacing with educational administration systems, student management systems, teaching resource libraries and external education data in schools, and adopting data encryption and desensitization technologies to ensure data safety and privacy;

the data processing and storage layer is used for cleaning, converting and integrating preprocessing operation on the acquired data, storing the data by adopting a distributed storage technology, establishing an index and carrying out multidimensional analysis and management by utilizing a data warehouse technology;

The intelligent analysis and decision layer is used for carrying out deep analysis and mining on the educational data based on an educational data analysis platform fused with the transfer learning and big model, a semantic network and an educational activity flow intelligent optimization algorithm of the big model and a personalized learning service system of emotion interaction and the big model, providing intelligent decision support and further comprising a model training and management module;

And the user interaction and application layer provides friendly man-machine interaction interfaces for teachers, students and education management staff, and meets the operation requirements of different users.

Further, the educational data analysis platform for fusion of the transfer learning and the large model comprises:

the migration learning module is used for constructing diversity and profession of the educational data, and utilizing a large model pre-trained on large-scale general data to migrate knowledge to the educational field through a migration learning method based on an attention mechanism;

The large model adaptation and optimization module selects a large model suitable for educational data processing, such as a pre-training language model based on a GPT architecture, and performs supervised fine-tuning training by utilizing a professional corpus in the educational field;

and (3) a data analysis flow, namely cleaning, labeling and characteristic engineering treatment on the educational data, inputting the educational data into a large model for local training, and optimizing the model by using a transfer learning technology for student learning condition assessment and learning progress prediction.

Further, the intelligent optimization algorithm for the educational activity flow of the semantic network and the large model comprises the following steps:

the education semantic network construction module is used for collecting and integrating knowledge in the education field, constructing structural semantic network data through entity extraction, relation extraction and semantic annotation technology, and storing the structural semantic network data in a graph database;

The activity flow modeling and optimizing module abstracts the education activity flow into a directed graph, analyzes a semantic network by using a large model, and digs bottlenecks and optimizing points in the flow;

And the intelligent decision support module searches information in the semantic network according to the education activity state and provides an activity organization scheme and decision suggestion by combining a large model intelligent recommendation function.

Further, the emotion interaction and large model personalized learning service system comprises:

The emotion interaction data acquisition and fusion module integrates voice emotion recognition, text emotion analysis and facial expression recognition, and acquires and fuses student emotion interaction data;

the large model driven personalized learning recommendation module is used for providing personalized learning resource recommendation and learning path planning by combining the large model with the learning condition, the hobbies and the emotion state of students;

And the learning effect evaluation and feedback module evaluates the learning effect of the students through multiple aspects of data and adjusts the personalized learning service strategy according to emotion feedback and learning experience.

On the other hand, the invention also provides an educational activity management method based on the large model technology, which comprises the following steps:

deploying data acquisition interfaces in each business system of a school to acquire original education data;

Cleaning, preprocessing, labeling and characteristic engineering processing are carried out on the acquired data;

selecting a proper large model, performing fine tuning training by using local marking data, and setting training parameters;

After the training of each school node is completed, the model parameters are uploaded to a transfer learning server, and the server updates the global model by adopting a transfer learning algorithm based on an attention mechanism;

And issuing the updated global model to each school node for educational data analysis.

Further, the method also comprises the intelligent optimization step of the educational activity flow:

collecting knowledge in the education field, and extracting entities and relations by natural language processing;

Converting the extracted knowledge into a semantic network, storing the semantic network in a graph database, and constructing an educational activity flow semantic network;

abstracting the education activity flow into a directed graph, inputting the directed graph into a large model for analysis, and mining optimization points;

And inquiring information in a semantic network according to the education activity state, and providing an optimized activity organization scheme and operation suggestions by combining a large-model intelligent recommendation function.

Further, the method also comprises the step of personalized learning service:

Integrating voice emotion recognition, text emotion analysis and facial expression recognition multi-mode interaction components in a teaching platform and mobile application;

Carrying out multi-mode fusion processing on the input data of the students, and accurately understanding the demands and emotion states of the students;

inputting the student demands and the emotion states into a large model to generate personalized learning recommendation contents;

and feeding back recommended content to students, recording learning behaviors and emotion feedback, and optimizing large model parameters and personalized learning service strategies.

Further, the original education data includes basic information of students, learning achievements, course arrangements.

Further, the education field knowledge includes course standards, teaching outline, education cases.

Further, the multi-mode interaction component is used for collecting the voice, the characters and the expression information of the students.

The beneficial effects are that:

The high-efficiency education data analysis realizes the deep analysis and mining of the education data through fusion of the migration learning and the large model, breaks through the data barriers between schools, improves the data utilization efficiency, provides accurate data support for education decision, and helps teachers and education management staff to better know the learning condition and the demands of students.

The intelligent education activity flow optimization is based on the semantic network and the algorithm of the large model, logic and rules of the education activity can be deeply understood, optimization points in the flow are mined, intelligent management and optimization of the education activity are realized, the education work efficiency is improved, the education cost is reduced, and education resources are more reasonably configured.

The high-quality personalized learning service, namely the personalized learning service system of emotion interaction and a large model, provides convenient, efficient and personalized learning experience for students, focuses on emotion requirements and learning states of the students, enhances learning power and self-confidence of the students and promotes comprehensive development of the students.

The scientific education decision support provides a plurality of feasible schemes for education decision by utilizing the strong knowledge reasoning and data analysis capability of the large model, predicts and evaluates the effect of the scheme, assists education departments and schools to make scientific and reasonable education policies and teaching plans, and improves the scientificity and the accuracy of the education decision.

Drawings

FIG. 1 is a schematic flow diagram of a system;

Detailed Description

Embodiments of the present invention are described in further detail below with reference to the accompanying drawings and examples. The following examples are illustrative of the invention but are not intended to limit the scope of the invention.

Example 1

The invention aims to provide an educational activity management system and method based on large model technology,

The system comprises a data acquisition and access layer, wherein a plurality of data acquisition interfaces are deployed to support the docking with a educational administration system, a student management system, a teaching resource library and the like in a school, and meanwhile, external education data such as online education platform data, education data issued by an education research institution and the like can be acquired. And the safety and privacy of personal information and educational data of students in the process of acquisition and transmission are ensured by adopting data encryption, desensitization and other technologies. For example, data such as student results and attendance records are acquired through an API interface, and data transmission encryption is performed by utilizing an SSL/TLS protocol.

And the data processing and storage layer is used for carrying out preprocessing operations such as cleaning, conversion, integration and the like on the acquired data, removing noise and abnormal values and unifying data formats. And a distributed storage technology, such as a Ceph distributed storage system, is adopted to store massive educational data, and a data index is established, so that the data query efficiency is improved. Meanwhile, the data warehouse technology is utilized to carry out multidimensional analysis and management on the data, and data support is provided for subsequent intelligent analysis and decision.

The intelligent analysis and decision layer is based on the educational data analysis platform, the semantic network and the intelligent optimization algorithm of the educational activity flow of the large model and the individualized learning service system of the emotion interaction and the large model, which are integrated with the transfer learning and the large model, and performs deep analysis and mining on the educational data to provide intelligent support for decision of educational activities. The layer also comprises a model training and management module which is responsible for training, optimizing and updating the large model so as to adapt to the continuously changing education requirements.

And a user interaction and application layer provides friendly man-machine interaction interfaces for teachers, students and education management staff. The teacher can make teaching plan making, teaching resource management, student learning condition analysis and other operations through the interface, students can interact with the system through websites, mobile applications and other modes to obtain learning resources, participate in educational activities, feed back learning problems and the like, and education management staff can make the work of organizing and arranging educational activities, resource allocation, data analysis and the like. The interface design follows the principle of simplicity and easiness in use, and the operation efficiency of the user is improved.

And constructing a migration learning framework aiming at the diversity and the specialty of the educational data. And the knowledge of the large model is migrated to the education field through a migration learning technology by utilizing the large model pre-trained on large-scale general data. For example, by adopting a migration learning method based on an attention mechanism, the method is quickly adapted to specific tasks and data characteristics in the education field on the basis of maintaining general knowledge of a large model.

Big model adaptation and optimization, namely selecting a big model suitable for educational data processing, such as a pre-training language model based on GPT architecture, and fine tuning according to the characteristics of the educational field. The large model is subjected to supervised fine tuning training by utilizing a professional corpus in the education field, including teaching materials, academic papers, teaching cases and the like, so that the large model can better understand and process education data.

And the data analysis flow is to firstly carry out cleaning, labeling and characteristic engineering treatment on the educational data, and then input the treated data into a large model for local training. After training is completed, the model is optimized on specific tasks in the education field by utilizing a transfer learning technology. Through analysis of student learning data, such as learning score, learning time, learning behavior and the like, support is provided for student learning condition assessment, learning progress prediction and the like.

Intelligent optimization algorithm for educational activity flow based on semantic network and large model

And constructing an educational semantic network, namely collecting and integrating various knowledge in the educational field, including information such as a course system, a teaching method, an educational policy and the like, and constructing a comprehensive educational semantic network. Through the technologies of entity extraction, relation extraction, semantic annotation and the like, educational knowledge is converted into structural semantic network data, and the structural semantic network data is stored in a graph database, so that knowledge inquiry and reasoning are facilitated.

Modeling and optimizing the activity flow, namely abstracting the education activity flow into a directed graph, wherein nodes represent activity links, and edges represent logical relations among the links. And analyzing the education semantic network by using the large model, and mining bottlenecks and optimization points in the flow. For example, by the reasoning capability of a large model, it is found that certain teaching links can employ more efficient teaching methods, or that certain educational resources can be distributed more reasonably, thereby improving the efficiency and quality of educational activities.

And the intelligent decision support is that when a teacher or an education manager organizes the education activities, the system searches related information in the semantic network according to the current activity state and combines the intelligent recommendation function of the large model to provide the optimal activity organization scheme and decision suggestion for the teacher or the education manager. For example, when scheduling a curriculum schedule, the system recommends an optimal curriculum schedule based on information such as teacher's teaching tasks, student's curriculum requirements, classroom resources, etc.

Individualized learning service system based on emotion interaction and large model

And the emotion interaction data acquisition and fusion are realized by integrating the technologies of voice emotion recognition, text emotion analysis, facial expression recognition and the like. For example, in the learning process of students, the system can collect voice information of the students through voice interaction equipment to analyze emotion tendencies in the voices of the students, collect facial expressions of the students through cameras to judge learning states and emotion changes of the students, and meanwhile, emotion analysis is carried out on text input of the students on a learning platform. And carrying out fusion processing on emotion data of different modes, and comprehensively knowing the emotion states of students.

And the large model driven personalized learning recommendation is to provide personalized learning resource recommendation and learning path planning for students by utilizing the powerful language understanding and generating capability of the large model and combining the learning condition, the hobbies and the emotion state of the students. The system firstly analyzes and semantically understands the input data of the students, then searches related knowledge in the large model, and generates recommended content according to personalized requirements of the students. For example, for students interested in mathematics but having difficulty at a certain knowledge point, the system may recommend related mathematics learning materials, online lessons, and solution skills.

And the learning effect evaluation and feedback is that the learning effect of the students is evaluated through various data such as the learning score, the work completion condition, the performance of the participation in the educational activity and the like of the students. Meanwhile, according to emotion feedback and learning experience of students, personalized learning service strategies are timely adjusted, and learning effects and satisfaction of the students are continuously improved. For example, if a student shows anxiety during learning, the system may adjust the learning progress and difficulty appropriately, providing more psychological counseling resources.

The data acquisition interface is deployed in each business system of the school, and is connected with the educational administration system, the student management system and the like to acquire the original educational data such as basic information, school achievements, course arrangement and the like of students.

And cleaning and preprocessing the acquired data, removing repeated data, correcting error data and filling missing values. For example, the data cleansing tool is used to process the error scores and repeated records in the student performance according to the data quality rules.

And marking and feature engineering processing are carried out on the preprocessed data, and key features in the data are extracted, such as quantized representation of the features of learning time, learning frequency and the like of students.

And selecting a proper large model, such as a GPT-4 model, and performing fine tuning training on a local node of a school by using local marked data. Training parameters, such as learning rate of 0.0001, training round number of 15 rounds, and optimizer choice Adagrad.

After the training of each school node is completed, the model parameters are uploaded to the migration learning server through the encryption channel. The server adopts a migration learning algorithm based on an attention mechanism to fuse and optimize parameters of each node and update the global model.

The migration learning server transmits the updated global model to all school nodes, and all nodes analyze the local education data by using the global model, such as predicting the learning score of students, evaluating the learning risk of the students and the like.

Various knowledge in the education field, including course standards, teaching outline, education cases, etc., is collected, and entity extraction and relation extraction are performed by using natural language processing technology. For example, entities such as course names, teaching targets, teaching contents, and the like, and association relations between the entities are extracted from course standards.

And converting the extracted knowledge into a semantic network and storing the semantic network in a Neo4j graph database. And constructing nodes and edges of the semantic network, namely, taking teaching links as nodes, taking the sequence among the links as edges, and constructing the educational activity flow semantic network.

The educational activity flow is abstracted into a directed graph, and is input into a large model for analysis. The large model discovers the optimization points in the educational activity flow through understanding and reasoning on the semantic network, and can be combined or adjusted in sequence in certain teaching links.

When a teacher or an education manager organizes the education activities, the system queries related information in the semantic network according to the current activity state and provides an optimized education activity organization scheme and operation advice for the teacher or the education manager by combining the intelligent recommendation function of the large model. For example, when organizing a teaching seminar, the system recommends appropriate conference flows and discussion topics based on the seminar theme, participants, and other information.

The multi-modal interaction components such as voice emotion recognition, text emotion analysis, facial expression recognition and the like are integrated in a teaching platform and mobile application of a school. For example, a student may input questions via voice, and the system uses voice emotion recognition techniques to determine the emotional state of the student.

When students input learning demands or feed back learning problems, the system firstly carries out multi-mode fusion processing on input data, integrates information such as voice, characters, expressions and the like, and accurately understands the demands and emotion states of the students. For example, a student uploads a homework photo and inquires about the thinking of solving the problem, and the system obtains homework contents through image recognition and performs comprehensive processing by combining text questions and emotion analysis.

The system inputs the student demands and the emotion states into a large model, and the large model generates personalized learning recommendation content according to the pre-trained knowledge and the finely tuned education field knowledge. For example, for a student with difficulty and low emotion in chinese writing, a large model may recommend relevant writing skill courses, excellent paradigms, and provide some encouragement and psycho-dispersion utterances.

The system feeds back recommended content to the students and records learning behaviors and emotion feedback of the students. By analyzing learning data and emotion feedback of students, parameters of a large model and personalized learning service strategies are continuously optimized, and service quality is improved. For example, if a student is not interested in a recommended content, the system analyzes the cause, adjusts the recommendation algorithm, and improves the recommended content.

Multiple schools in a certain area are selected as experimental objects, including primary schools, middle schools and universities, and the education activity management system provided by the invention is deployed in the schools.

The experimental schools are divided into an experimental group and a control group, wherein the experimental group adopts the system and the method of the invention, and the control group adopts a traditional education activity management mode.

And determining experimental indexes including student learning score improvement rate, education activity organization efficiency, student satisfaction and the like. The improvement rate of the student learning score is measured by comparing examination scores of students before and after an experiment, the education activity organization efficiency is evaluated by counting indexes such as activity preparation time, resource utilization rate and the like, and the student satisfaction is collected through questionnaire investigation and online evaluation.

The average improvement rate of the learning score of the students in the experimental group reaches 20%, and the learning score is particularly obvious in main subjects such as mathematics, chinese and the like. For example, in the mathematics subjects, the average ratio of students in the experimental group is improved by 10 points compared with that in the control group, and through personalized learning service, the students can learn in a targeted way aiming at own weak links, so that the learning effect is improved.

Educational activity organization efficiency, namely the educational activity setup time of the experimental group is shortened by 30 percent on average, and the resource utilization rate is improved by 25 percent. For example, when organizing a campus culture activity, the activity links and the resource allocation are reasonably arranged through intelligent optimization of the educational activity flow, so that the activity preparation time is shortened from the original one week to four days, and the resource waste is reduced.

Student satisfaction, namely, through questionnaire investigation and online evaluation, the student satisfaction of the experimental group reaches more than 90%, and the control group is only about 70%. The personalized learning service function of the student feedback system can meet the learning requirement of students, the enthusiasm and initiative of learning are improved, meanwhile, the intelligent interaction function is convenient and quick, and the communication interaction between the students and the teacher and the system is enhanced.

Experimental results show that the education activity management system and method based on the large model technology can remarkably improve the learning score of students, improve the organization efficiency of the education activity, enhance the satisfaction of the students, have good application prospect and popularization value, and provide powerful technical support for intelligent management of the education activity.

Example 2

And collecting and preprocessing data such as past course selection records, school achievements, interest and hobby investigation results of students. And cleaning the data to remove missing values and error data, and then carrying out normalization processing to unify different types of data into the same numerical range for subsequent analysis. For example, the score is converted into a value between 0 and 1, and the interest is digitally represented by means of single-heat coding and the like.

And constructing a student status space, namely combining various characteristics of students into a status vector as the input of the reinforcement learning intelligent agent. The status vector includes information of the student's selected course, current learning progress, score ranking, interest preference, etc. For example, the state vector is represented using a multi-dimensional array, one feature for each dimension.

Defining an action space, wherein the action space is a course set which can be selected by students. Each course acts as an action, and the agent affects the learning state of the student by selecting different course combinations.

And designing a reward function, wherein the reward function aims at measuring learning benefits of students after selecting courses. For example, rewards are comprehensively calculated according to indexes such as subsequent score improvement, course completion rate, interest satisfaction and the like of students. The larger the score improvement amplitude, the higher the course completion rate, the higher the matching degree of the selected course and the interests, and the higher the rewarding value.

Model training, namely training by adopting reinforcement learning algorithms such as Deep Q Network (DQN) and the like. In the training process, the intelligent agent selects actions (courses) according to the current student status, observes the new status after the actions are executed and obtains rewards, and continuously updates Q value tables or neural network parameters to optimize strategies and maximize long-term accumulated rewards. And obtaining a strategy model capable of recommending optimal course combinations according to student states through repeated iterative training.

Blockchain network building, namely selecting a proper blockchain platform, such as an Ethernet or a alliance chain. Each school serves as a blockchain node, deploys a blockchain client, and sets public and private key pairs of the nodes for data encryption and signature verification.

And (3) encrypting and uploading data, namely encrypting the data such as student score, comprehensive quality evaluation and the like, for example, adopting an AES symmetric encryption algorithm. The encrypted data is combined with metadata (such as the school to which the data belongs, the data type, the timestamp, etc.) into a transaction record. Each school node signs the transaction record and then broadcasts it into the blockchain network.

And (3) intelligent contract writing and deployment, namely writing data access authority management intelligent contracts by using intelligent contract programming languages such as Solidity. The intelligent contracts define access authority rules of different roles (teachers, education management staff, students and the like) on data, for example, the teacher can check the performance of students in a class of the teacher, and the education management staff can check comprehensive data of students in a whole school. And deploying the intelligent contract on the blockchain to acquire the contract address.

And (3) data access and verification, namely when the data need to be accessed, the user sends an access request to the block link, wherein the access request comprises user identity information and identification of the required data. The node verifies the user identity and then invokes the smart contract to query access rights. If the authority allows, the node acquires the encrypted data from the blockchain, decrypts the data by using the corresponding private key and returns the encrypted data to the user.

Teacher models, which are complex large models trained on large-scale educational data, such as pre-trained language models based on a transducer architecture, and student models are selected. The student model selects a lightweight model with simple structure and less parameters, such as MobileNet and other neural network structures.

The distillation loss function is defined, and consists of two parts, namely, soft label loss between the student model and the teacher model output and hard label loss between the student model output and the real label. Soft tag loss is measured by calculating KL divergence of the student model and teacher model output probability distributions, and hard tag loss employs a cross entropy loss function. For example, distillation loss = α soft tag loss + (1- α) hard tag loss, where α is a super parameter used to balance the weights of the two losses.

Knowledge distillation training, namely, when training a student model, inputting the same input data into a teacher model and the student model simultaneously. The teacher model outputs soft labels (probability distributions) and the student model trains based on the hard labels (true labels) and the soft labels of the teacher model. Through a back propagation algorithm, parameters of the student model are continuously adjusted, so that the output of the student model is as close to that of a teacher model as possible, and meanwhile, the prediction capability of the real label is kept. Through multiple rounds of training, the student model gradually learns the key knowledge of the teacher model, and becomes a lightweight model capable of efficiently running on low-configuration equipment.

And (3) agent modeling, namely defining a plurality of agents according to different links of educational activities. For example, a venue arrangement agent is responsible for planning an active venue layout and a programming agent is responsible for determining an active program order and time. Each agent has its own targets, knowledge base and decision mechanism. The knowledge base of the intelligent agent contains knowledge related to the task of the intelligent agent, such as information of different site sizes, equipment placement rules and the like stored in the knowledge base of the site arrangement intelligent agent.

The communication mechanism is established by designing a communication protocol between the agents, for example using a message queue or a publish-subscribe mode. The agents share information and coordinate actions by exchanging messages between them. For example, after determining the program time, the programming agent sends a message to the venue operator informing the program duration and stage equipment requirements, and the venue operator adjusts the venue layout scheme based on the information.

Task allocation and collaboration, namely analyzing the whole flow of the educational activity by the large model, and making an overall task plan. The tasks are then broken down and distributed to the individual agents. Each agent autonomously decides and executes tasks according to the assigned tasks and the received information of other agents. In the execution process, the intelligent agent continuously adjusts own actions according to the new information so as to realize efficient collaborative organization of educational activities. For example, when a certain program time is temporarily adjusted, the program arrangement agent timely notifies the venue arrangement agent and other related agents, each of which correspondingly adjusts its own schedule.

Fifth embodiment is virtual teaching resource generation based on countermeasure generation network

Data collection and preprocessing, that is, collecting a large amount of art data including image data of pictorial representations, audio data of musical representations, and the like. The method comprises the steps of carrying out normalization processing on image data, adjusting pixel values to be 0-1, carrying out preprocessing operations such as sampling, framing and the like on audio data, and extracting audio characteristics such as Mel Frequency Cepstrum Coefficient (MFCC).

The generator and the discriminator are designed in such a way that the generator adopts a Convolutional Neural Network (CNN) or a cyclic neural network (RNN) structure, and virtual works of art are generated according to information such as styles, subjects and the like of works of art obtained by large model analysis. For example, for paint generation, the generator inputs a random noise vector and a style, theme-encoded vector, outputting a virtual paint image. The discriminator also adopts a CNN or RNN structure for judging the authenticity of the generated virtual works and real works.

Countermeasure training the generator and the discriminator perform countermeasure training. The generator attempts to generate a realistic virtual work to fool the discriminant, which in turn attempts to distinguish the real work from the generated virtual work. During training, the parameters of the generator and the arbiter are updated alternately. The generator adjusts the generation strategy according to the feedback of the discriminator to make the generated works more vivid, and the discriminator continuously improves the discrimination capability according to the difference between the real works and the generated works. Through multiple rounds of countermeasure training, the generator can generate high-quality virtual teaching resources, and the requirements of art education are met.

Claims (10)

1. A large-scale intelligent educational activity management system, characterized by comprising: The data collection and access layer deploys a variety of data collection interfaces to connect with the school's internal teaching system, student management system, teaching resource library, and external education data, and uses data encryption and desensitization technology to ensure data security and privacy; The data processing and storage layer cleans, converts, and integrates the collected data for preprocessing operations, uses distributed storage technology to store data and create indexes, and uses data warehouse technology for multi-dimensional analysis and management; Intelligent analysis and decision-making layer, based on the education data analysis platform integrating transfer learning and big models, the intelligent optimization algorithm of education activity process based on semantic network and big models, and the personalized learning service system based on emotional interaction and big models, it conducts in-depth analysis and mining of education data, provides intelligent decision-making support, and also includes model training and management modules; The user interaction and application layer provides a friendly human-computer interaction interface for teachers, students and educational administrators to meet the operational needs of different users. 2. The educational activity management system according to claim 1, characterized in that the educational data analysis platform integrating transfer learning and large models comprises: The transfer learning module is built to target the diversity and specialization of educational data. It uses a large model pre-trained on large-scale general data to transfer knowledge to the education field through a transfer learning method based on the attention mechanism. Large model adaptation and optimization module, which selects large models suitable for educational data processing, pre-trained language models based on the GPT architecture, and uses professional corpora in the education field for supervised fine-tuning training; The data analysis process involves cleaning, labeling, and feature engineering the educational data before inputting it into a large model for local training. The model is then optimized using transfer learning technology for use in evaluating student learning situations and predicting learning progress. 3. The educational activity management system according to claim 1, characterized in that the educational activity process intelligent optimization algorithm of the semantic network and the large model includes: The educational semantic network construction module collects and integrates knowledge in the educational field, constructs structured semantic network data through entity extraction, relationship extraction and semantic annotation technology, and stores it in a graph database; The activity process modeling and optimization module abstracts the educational activity process into a directed graph, uses a large model to analyze the semantic network, and discovers bottlenecks and optimization points in the process; The intelligent decision support module searches for information in the semantic network according to the status of educational activities, and provides activity organization plans and decision suggestions in combination with the large model intelligent recommendation function. 4. The educational activity management system according to claim 1, characterized in that the emotional interaction and large model personalized learning service system comprises: Emotional interaction data collection and fusion module, which integrates speech emotion recognition, text emotion analysis, and facial expression recognition, and collects and integrates students' emotional interaction data; The personalized learning recommendation module driven by the big model uses the big model to combine students' learning situation, interests, hobbies and emotional state to provide personalized learning resource recommendations and learning path planning; The learning effect evaluation and feedback module evaluates students' learning effects through multi-faceted data and adjusts personalized learning service strategies based on emotional feedback and learning experience. 5. A method for managing educational activities based on a large model, characterized in that it includes the following steps: Deploy data collection interfaces in various business systems of the school to obtain original education data; Clean, preprocess, annotate and feature engineer the collected data; Select a suitable large model, use local annotated data for fine-tuning training, and set training parameters; After the training of each school node is completed, the model parameters are uploaded to the transfer learning server, and the server uses the transfer learning algorithm based on the attention mechanism to update the global model; The updated global model will be sent to each school node for educational data analysis. 6. The educational activity management method according to claim 5, characterized in that it also includes the step of intelligent optimization of the educational activity process: Collect knowledge in the field of education and use natural language processing for entity extraction and relationship extraction; The extracted knowledge is converted into a semantic network and stored in a graph database to construct a semantic network of educational activity processes; Abstract the educational activity process into a directed graph and input it into the big model for analysis to discover optimization points; Query information in the semantic network according to the status of educational activities, and provide optimized activity organization plans and operation suggestions in combination with the big model intelligent recommendation function. 7. The educational activity management method according to claim 5, further comprising a personalized learning service step: Integrate speech emotion recognition, text emotion analysis, and facial expression recognition multimodal interaction components into teaching platforms and mobile applications; Perform multimodal fusion processing on student input data to accurately understand students' needs and emotional states; Input students’ needs and emotional states into the big model to generate personalized learning recommendations; Feedback recommended content to students, record learning behaviors and emotional feedback, and optimize large model parameters and personalized learning service strategies. 8. The educational activity management method according to claim 5 is characterized in that the original educational data includes students' basic information, academic performance, and course schedule. 9. The educational activity management method according to claim 6 is characterized in that the educational field knowledge includes curriculum standards, teaching syllabuses, and educational cases. 10. The educational activity management method according to claim 7 is characterized in that the multimodal interaction component is used to collect students' voice, text, and expression information.