Education course script generation system based on zepo large model fusion data processing

Abstract

This invention relates to the field of multi-source educational data fusion technology and discloses an educational course script generation system based on the fusion data processing of the Zhoubao large model. It employs a relative ranking method to determine the importance of teaching nodes and marks core knowledge points as key points for completing the script game. It establishes a user cognitive feature system and a manager's teaching style feature system, constructs a pre-mapping table of teacher and student dual-subject features and teaching preferences, and dynamically balances personalization and knowledge scope through feature overlap. It determines the starting and ending knowledge points of the knowledge graph through teaching objectives, selects the path containing the most highly important nodes as the optimal teaching path, and collaboratively generates text and script game videos under causal constraints. It collects classroom audio and video data, identifies dynamic classroom events based on continuous state change trends, and dynamically adjusts teaching content. It also dynamically adjusts the model iteration frequency based on teacher and student feedback trends. This invention achieves automatic generation of knowledge point-based, level-by-level teaching script game videos.

Description

Technical Field

[0001] This invention relates to the field of multi-source educational data fusion technology, specifically to an educational course script generation system based on the fusion data processing of the cosmic package big data model. Background Technology

[0002] Existing educational course script generation technologies based on large models have multi-dimensional systemic defects, making it difficult to meet the actual needs of high-quality teaching; In existing technologies, at the knowledge system construction level, most focus on simply listing explicit knowledge points of a single subject, failing to establish causal dependencies that conform to teaching principles, completely ignoring the implicit knowledge that determines teaching quality, and failing to effectively transfer the teaching experience of excellent administrators. Furthermore, they lack interdisciplinary knowledge association mechanisms and special education adaptation capabilities; interdisciplinary scripts are merely simple compilations of knowledge points, unable to cover the differentiated teaching needs of special needs students. Using absolute thresholds to determine teaching focus fails to adapt to the teaching requirements of different subjects and grade levels. At the personalized generation level, existing technologies make superficial adjustments to difficulty based on grade level and subject, failing to establish a personalized adaptation system involving both teachers and students, ignoring differences in administrators' teaching styles, resulting in generated scripts that do not match administrators' teaching habits; and they suffer from a difficult-to-solve cold start paradox. In the initial stage of system launch, personalized services could not be provided without student data. Overfitting occurred during the data accumulation phase, limiting students' knowledge horizons and thinking development. Existing technology lacks causal logic constraints for teaching, resulting in cognitive leaps and missing derivation processes. Multimodal content is merely a simple splicing of different modalities with low semantic alignment. It lacks the ability to generate knowledge point-based script game videos, failing to transform knowledge points into game completion points and generate teaching script game videos, making it difficult to enhance students' learning initiative through gamified interaction. The generated static scripts cannot respond to dynamic classroom events, relying on absolute thresholds to judge classroom status, resulting in poor recognition accuracy. It relies on cloud deployment, making it unusable in offline environments. The model iteration frequency is fixed, unable to be dynamically optimized based on teaching feedback, and the generated content lacks effective copyright traceability and protection mechanisms. Therefore, there is a need to provide an educational course script generation system based on the fusion data processing of the cosmic package large model. Summary of the Invention

[0003] The purpose of this invention is to provide an educational course script generation system based on the fusion data processing of the cosmic package model. Through multi-dimensional teaching causal knowledge graphs, teacher-student dual-subject learning pre-mapping, and trend-driven dynamic adjustment technologies, it achieves automatic generation of knowledge-point-based teaching script game videos. This system generates educational course scripts that combine teaching professionalism, dual-subject personalization, and dynamic classroom adaptability. To solve the aforementioned problems in the prior art, this invention is achieved through the following technical solutions: The first part, the educational course script generation system based on fusion data processing of the cosmic package large model provided in this embodiment of the invention, specifically includes the following steps: Adaptation Module: Construct a multi-dimensional hierarchical knowledge graph for teaching, and uniformly anchor implicit knowledge units, interdisciplinary integration points, and special education adaptation strategies to corresponding nodes and edges. Use relative sorting method to determine the importance of teaching nodes and mark core knowledge points as key points for passing the script game. Mapping Module: Combining multidimensional hierarchical knowledge graphs for teaching, we establish a user cognitive feature system and a manager teaching style feature system, train a meta-learning basic model, construct a pre-mapping table of teacher and student dual subject features and teaching preferences, and dynamically balance the degree of personalization and knowledge scope through the degree of feature overlap. The generation module combines the characteristics of both teachers and students with a pre-mapping table of teaching preferences. It determines the starting and ending knowledge points of the knowledge graph through teaching objectives, selects the path containing the most high-importance nodes as the optimal teaching path, and collaboratively generates text and script game videos under causal constraints. It generates a unique identifier for each content segment and stores it in a distributed ledger. Adjustment Module: For the generated text and script game video, collect classroom audio and video data, identify dynamic classroom events by combining the continuous change trend of the state, and dynamically adjust the teaching content. The model iteration frequency is dynamically adjusted according to the change trend of teacher and student feedback.

[0004] The second part, the educational course script generation method based on fusion data processing of the large cosmic package model provided in this embodiment of the invention, specifically includes the following steps: Step 1: Construct a multi-dimensional hierarchical knowledge graph for teaching, and uniformly anchor implicit knowledge units, interdisciplinary integration points, and special education adaptation strategies to corresponding nodes and edges. Use the relative ranking method to determine the importance of teaching nodes, and mark core knowledge points as key points for passing the script game. Step 2: Combine multi-dimensional hierarchical knowledge graph of teaching to establish user cognitive feature system and administrator teaching style feature system, train meta-learning basic model, construct pre-mapping table of teacher and student dual subject features and teaching preferences, and dynamically balance the degree of personalization and knowledge vision through feature overlap. Step 3: Combining the pre-mapping table of teacher and student dual subject characteristics and teaching preferences, the starting and ending knowledge points of the knowledge graph are determined through teaching objectives. The path containing the most high-importance nodes is selected as the optimal teaching path. Under causal constraints, text and script game videos are collaboratively generated. A unique identifier is generated for each content segment and stored in a distributed ledger. Step 4: Collect classroom audio and video data for the generated text and script game video, identify dynamic classroom events by combining the continuous change trend of the state, and dynamically adjust the teaching content. Adjust the model iteration frequency dynamically based on the change trend of teacher and student feedback.

[0005] The beneficial effects of this invention are: 1. Construct a multi-dimensional hierarchical knowledge graph for teaching, comprising three sub-graphs: subject-specific causal relationships, interdisciplinary knowledge connections, and special education adaptation. This graph anchors implicit knowledge units, interdisciplinary integration points, and special education adaptation strategies to corresponding nodes and edges. A relative ranking method is used to determine the importance of teaching nodes, and core knowledge points are marked as key points for completing a scenario-based game. This addresses the shortcomings of constructing a single-subject explicit knowledge graph, lacking causal logic in teaching, completely ignoring the transmission of implicit knowledge, simply piecing together interdisciplinary content, lacking adaptation to special education scenarios, and using absolute thresholds to determine teaching focus. A dual-subject characteristic system for teachers and students and a meta-learning pre-mapping mechanism are established. A basic meta-learning model is trained on large-scale anonymous group data, and a pre-mapping table of teacher and student dual-subject characteristics and teaching preferences is constructed. The degree of personalization and knowledge scope is dynamically balanced by the overlap between individual and group characteristics. This addresses the shortcomings of students' superficial difficulty adjustment, completely ignoring differences in administrators' teaching styles, and failing to overcome the cold-start paradox of being unable to personalize without data and easily overfitting with data. 2. This paper proposes an optimal teaching path and gamification design method based on knowledge point-based progression. It supports administrators in directly inputting knowledge points to generate teaching scripts and game videos, setting core knowledge points as key progression points. Under strict causal constraints in teaching, it collaboratively generates teaching scripts, game videos, and multimodal resources. A distributed ledger generates a unique identifier for each content segment, enabling end-to-end copyright traceability. This addresses shortcomings such as the lack of scientific basis for teaching path selection, potential cognitive leaps and incomplete derivations, semantic disconnect between multimodal content, lack of copyright protection for generated content, and lack of gamification video generation capabilities. A dynamic event recognition and global adjustment mechanism based on continuous changes in classroom status and game progression data is adopted. Combined with a cloud-edge collaborative architecture, it supports offline operation and dynamically adjusts the model iteration frequency based on teacher and student feedback and progression rate trends. This solves the problems of static script generation failing to respond to classroom dynamics, reliance on absolute thresholds to judge classroom status, unavailability of offline environments, fixed model iteration frequency preventing timely optimization, and monotonous and uninteresting learning methods. Attached Figure Description

[0006] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0007] Figure 1 This is a schematic diagram of the educational course script generation system based on cosmic package large model fusion data processing provided in Embodiment 1 of the present invention; Figure 2This is a flowchart of the steps in the educational course script generation method based on fusion data processing of the cosmic package large model provided in Embodiment 2 of the present invention; Figure 3 This is a flowchart of the steps for generating a level-based script game video using the educational course script generation method based on the fusion data processing of the cosmic package large model provided in Embodiment 1 of the present invention. Detailed Implementation

[0008] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0009] Example 1: As Figure 1 and Figure 3 As shown in the figure, the educational course script generation system based on the fusion data processing of the cosmic package large model provided in this embodiment of the invention specifically includes the following modules: Adaptation Module: Construct a multi-dimensional hierarchical knowledge graph for teaching, and uniformly anchor implicit knowledge units, interdisciplinary integration points, and special education adaptation strategies to corresponding nodes and edges. Use relative sorting method to determine the importance of teaching nodes and mark core knowledge points as key points for passing the script game. In a specific embodiment, a multi-dimensional hierarchical knowledge graph for teaching is constructed simultaneously. The multi-dimensional hierarchical knowledge graph for teaching includes: a subject teaching causal subgraph, an interdisciplinary knowledge association subgraph, and a special education adaptation subgraph. The subject teaching causal subgraph is constructed by extracting all explicit knowledge points from textbooks and curriculum standards and establishing the sequential dependency relationship between knowledge points according to the teaching order presented in the classroom recording of the super-level administrator. Each knowledge point node records its location and teaching requirements in different grade levels and different textbook versions, and each directed edge records the teaching transition logic between two knowledge points; Constructing interdisciplinary knowledge relationship sub-graphs: Discovering the inherent connections between knowledge points from different disciplines from interdisciplinary thematic learning guides and project-based learning cases; For example, we can establish connections between knowledge points about the Silk Road in history, the topography and climate of Central Asia in geography, frontier poetry in language, ancient weights and measures in mathematics, and Dunhuang murals in art. Each associated edge records the associated type and the applicable interdisciplinary topics. The associated types include: spatiotemporal association, principle association, application association, and aesthetic association. We constructed a special education adaptation sub-map, extracting teaching adjustment strategies for different types of special users from special education curriculum standards and classroom recordings of special education administrators; and added four special education adaptation tags for each knowledge point node, namely visual impairment, hearing impairment, learning disability and autism, with each tag corresponding to a set of exclusive teaching methods and content presentation methods. The teaching experience content that does not contain explicit knowledge points is extracted from all classroom recordings. Using the long contextual understanding ability of the Zhoubao model, it is transformed into standardized implicit knowledge units. Each unit contains three attributes: knowledge content, applicable teaching scenario, and applicable knowledge point range. The applicable teaching scenarios are further subdivided into: knowledge point introduction, concept explanation, difficulty breakthrough, example analysis, classroom questioning, error correction, and classroom summary. Anchor implicit knowledge units in teaching to the corresponding nodes and edges of the subject teaching causal subgraph: units applicable to the explanation of a single knowledge point are anchored to the knowledge point node, and units applicable to the transition between two knowledge points are anchored to the directed edge connecting the two knowledge points. Anchor the interdisciplinary integration points to the corresponding edges of the interdisciplinary knowledge association subgraph, and each integration point contains a specific interdisciplinary teaching activity design; Special education adaptation strategies are anchored to corresponding labels on special education adaptation submaps, with each strategy including specific content adjustment methods and teaching activity suggestions; Calculate the number of prerequisite dependencies and the total number of implicit knowledge units anchored for each knowledge point node. Sort all knowledge point nodes from highest to lowest according to the sum of the number of prerequisite dependencies and the number of implicit knowledge units. The ranking result is the relative importance of the teaching nodes. The system automatically marks the top 30% of knowledge points as core walkthrough points for the scenario game, marks the (30%) and (70%) points as side quest walkthrough points, and marks the bottom 30% as easter egg expansion points. Each walkthrough point is pre-associated with a corresponding gamified teaching resource template, including five basic templates: puzzle, adventure, experiment simulation, role-playing, and competitive challenge. It also supports administrators to customize templates. In the subsequent script generation process, priority is given to ensuring the allocation of teaching time and resources for knowledge points that are ranked first, and richer teaching activities and more detailed explanations are automatically designed for knowledge points with high importance; avoid using absolute thresholds to determine teaching focus, and adapt to the teaching requirements of different subjects and different grade levels; Mapping Module: Combining multidimensional hierarchical knowledge graphs for teaching, we establish a user cognitive feature system and a manager teaching style feature system, train a meta-learning basic model, construct a pre-mapping table of teacher and student dual subject features and teaching preferences, and dynamically balance the degree of personalization and knowledge scope through the degree of feature overlap. In a specific embodiment, the user cognitive feature system and the administrator teaching style feature system are constructed through the multi-dimensional teaching hierarchical knowledge graph obtained in step one. The meta-learning basic model is trained on large-scale anonymous group learning data to establish a pre-mapping relationship between the dual subject features of teachers and students and teaching preferences. Establish a user cognitive characteristic system, which includes six dimensions: educational stage, subject, grade, academic level, interest tendency, and learning habits; Each dimension is divided into a predetermined number of mutually exclusive categories. Academic level is divided into three levels: basic, intermediate, and excellent. Interest tendencies are divided into five categories: scientific inquiry, literary creation, artistic expression, sports, and history and humanities. Establish a teaching style characteristic system for administrators, which includes five dimensions: language style, teaching pace, interaction methods, blackboard design and classroom management; each dimension is divided into a predetermined number of typical style types; For example, language styles can be categorized as: meticulous and rigorous, humorous and witty, passionate and inspiring, and persuasive and guiding; interaction methods can be categorized as: manager-led, user-inquiry-based, and collaborative learning. Collect anonymous user group learning data and group them according to the six dimensions of the user cognitive characteristic system; each group contains learning process data of all users with the same combination of cognitive characteristics. The learning process data includes: average learning time, average homework accuracy, average number of classroom interactions and average knowledge mastery, average game completion time and average level retry rate. Statistical analysis was performed on each set of data to calculate the average teaching needs of each user group for each knowledge point. The average teaching needs included: suggested teaching time, suggested teaching difficulty, suggested interaction frequency, suggested level difficulty gradient, and suggested plot complexity. We collected teaching style data from anonymous managers, grouped them according to the five dimensions of the manager teaching style characteristic system, and calculated the typical teaching preferences of each group of managers in each teaching segment. Typical teaching preferences include: introduction method, questioning skills, explanation style and summary method. The statistically obtained teaching preference data of user groups and teaching preference data of managers are input into the Zhoubao large model for meta-learning training. The Zhoubao model is trained through meta-learning to predict the optimal teaching preference based on any given combination of user cognitive features and manager teaching style. In the initial stage of launch, if there is no personal data of any individual user or administrator, the corresponding optimal teaching preference data will be directly retrieved from the pre-mapping table based on the user's basic information input by the administrator and the administrator's own teaching style, and a personalized teaching script that is suitable for both the user's cognitive level and the administrator's teaching style will be generated. If a small amount of personal usage data from a single user or administrator is collected, the corresponding personal characteristics are compared with the group characteristics in the pre-mapping table, and the degree of overlap between the two in each dimension is calculated. If the degree of overlap is high, it means that the characteristics of the corresponding user or manager match those of the corresponding group, and a higher degree of personalization should be adopted. If the overlap is low, it indicates that the corresponding user or manager has unique characteristics. Therefore, the personalization level of the corresponding user or manager should be reduced, while the expanded content should be increased. The generation module combines the characteristics of both teachers and students with a pre-mapping table of teaching preferences. It determines the starting and ending knowledge points of the knowledge graph through teaching objectives, selects the path containing the most high-importance nodes as the optimal teaching path, and collaboratively generates text and script game videos under causal constraints. It generates a unique identifier for each content segment and stores it in a distributed ledger. In a specific embodiment, the administrator inputs the teaching objectives, determines the starting and ending knowledge points of the teaching from the multi-dimensional teaching knowledge graph based on the teaching objectives, and searches all teaching paths from the starting knowledge point to the ending knowledge point. Each path contains a series of continuous knowledge point nodes and directed edges. The teaching objectives include: subject, grade level, chapter, duration, user information, administrator's teaching style, whether interdisciplinary integration is needed, whether special education adaptation is required, whether scripted game videos are generated, and game type selection. The starting and ending knowledge points of the dynamic derivation teaching are identified, and all teaching paths from the starting knowledge point to the ending knowledge point are searched. Each path contains a series of continuous knowledge point nodes and directed edges. Based on the relative importance ranking of the teaching nodes, the path containing the most important knowledge points is selected as the optimal teaching path. If the administrator selects interdisciplinary integration, interdisciplinary integration points related to the teaching content are automatically inserted into the optimal teaching path. If the administrator selects special education adaptation, a corresponding special education adaptation strategy is automatically added to each knowledge point in the optimal teaching path. Based on the optimal teaching path and the teaching preferences of teachers and students, the teaching process is divided into a predetermined number of teaching segments, and a reasonable teaching time is allocated to each teaching segment. At the same time, the teaching segments are transformed into game level structures: the initial knowledge points correspond to the game's introductory level, the core level completion points correspond to independent core levels, the side level completion points correspond to sub-tasks in the main level, and the easter egg expansion points correspond to hidden levels. For each teaching segment, core text teaching content is generated, which includes: explanation of knowledge points, analysis of examples, and classroom questions; in the generation process, the cause-and-effect logic of teaching is strictly followed, the explanation of each knowledge point is combined with the corresponding prerequisite knowledge points, and the conclusion of each conclusion includes a complete derivation process; Based on the needs of the teaching content, the system automatically generates corresponding script game videos, which include: mathematical formulas, geometric figures, experimental diagrams, animation scripts, and audio narration; and utilizes the multimodal unified understanding capabilities of the Zhoubao large model to ensure that all multimodal resources are semantically fully aligned with the text content. For example, the generated geometry accurately corresponds to the geometric relationships described in the text, and the generated animation script accurately demonstrates the physical processes described in the text; The implicit knowledge units anchored to the corresponding knowledge points and edges are naturally integrated into the teaching content, ensuring that the generated scripts contain the teaching wisdom and experience of excellent managers. A unique content identifier is generated for each generated teaching content segment, and the generation time, generation model version, source of training data used, and information on publicly available teaching resources referenced are recorded. Administrators upload original teaching content to the platform, generate a unique copyright certificate for the original content, and record it in the distributed ledger; Adjustment module: For the generated text and script game video, collect classroom audio and video data, identify dynamic classroom events by combining the continuous change trend of the state and dynamically adjust the teaching content, and dynamically adjust the model iteration frequency according to the change trend of teacher and student feedback; In a specific embodiment, speech recognition technology is used to analyze the dialogue between users and administrators in the classroom to extract the number of times users spoke, the content of their questions, and the accuracy of their answers; computer vision technology is used to analyze users' facial expressions and body movements to extract information on changes in users' attention and emotions. Collect students' playthrough data for the script game, including: completion time for each level, number of retries for each level, distribution of incorrect answers, frequency of clicks on interactive buttons, and trigger rate of hidden levels; A multi-dimensional quantitative indicator system for clearance effectiveness is constructed, covering three core dimensions: knowledge mastery, process participation, and thinking development. The knowledge mastery dimension includes single-level clearance time, accuracy rate of answering core knowledge points, accuracy rate of applying prerequisite knowledge points, and the concentration of incorrect answers. The process participation dimension includes the frequency of effective interactive button clicks, the trigger rate of hidden levels, and the contribution to tasks in multi-person collaborative mode. The thinking development dimension includes the logical score of answering open-ended questions, the score of applying interdisciplinary knowledge points, and the recognition rate of non-standard innovative solutions. The system performs real-time evaluation at each level and comprehensive evaluation of the whole system. After a student completes a single game level, all indicator data for that level are collected immediately and compared with the benchmark data of the same cognitive characteristics group in the pre-mapping table in a standardized manner to calculate the knowledge mastery score for that level. After all teaching levels are completed, the overall pass score is calculated by weighting the weights of each level and generating a heatmap of the mastery of knowledge points at a granular level. The system implements a tiered assessment of knowledge mastery and identifies deficiencies, dividing overall pass rates into four levels: 90 points and above for complete mastery, 70-89 points for basic mastery, 50-69 points for partial mastery, and below 50 points for no mastery. A separate mastery report is generated for each core knowledge point, identifying the knowledge points with deficiencies and their upstream dependencies, and marking the error type. Establish a closed-loop linkage mechanism for evaluation results, and use the automatic evaluation results as the core basis for identifying dynamic events in the classroom. If the mastery of a single core knowledge point is below 60 points, a targeted supplementary explanation process will be directly triggered. If the overall pass rate is below 70 points, a post-class reinforcement script and exclusive game level will be automatically generated to match the student's weaknesses. The evaluation data will be integrated with the classroom audio and video feedback data to jointly drive the fine-tuning of the meta-learning model parameters and the updating of the knowledge graph content. By analyzing the continuous changing trends of classroom status data, dynamic events in the classroom can be identified. If the number of times a user speaks continues to decrease in three consecutive teaching sessions, or if the user's concentration shows a significant downward trend, it is judged that the user is experiencing learning fatigue. If, after any knowledge point is explained, the user asks multiple related questions, or if the user's correct answer rate is significantly lower than the average level, it is judged that the user has not mastered the corresponding knowledge point firmly. If a dynamic event in the classroom is detected, the current teaching script is paused, and the adjusted teaching content and process are generated in real time based on the event type and the trend of changes in the classroom status. After identifying dynamic events in the classroom, immediately locate the precise node position of the current teaching progress in the subject teaching causal subgraph, and simultaneously retrieve the node's preceding dependency chain, subsequent teaching path, and node relative importance ranking data; By combining the teaching preference data corresponding to the dual subject characteristics of teachers and students in the meta-learning basic model, the influence range of events on the overall causal logic of teaching is calculated, the teaching time of high-importance nodes is reserved first, the order and time allocation of the remaining teaching links are re-planned, and the adjusted process is ensured to not destroy the causal dependence between knowledge points. At the same time, the backup teaching implicit knowledge units anchored on relevant nodes are automatically associated. If a user experiences learning fatigue, a fun interactive activity or short story related to the teaching content will be inserted, and the duration of the activity will be dynamically adjusted according to the changing trend of the user's fatigue level. When a continuous trend of user learning fatigue is detected, content that is related to the principle or application of the current knowledge point is retrieved from the interdisciplinary knowledge association sub-graph. At the same time, interesting introduction and interactive experience teaching implicit knowledge units anchored on the current knowledge point and adjacent low importance nodes are filtered. By combining the interest tendency dimension in the user's cognitive feature system and the language style dimension in the manager's teaching style feature system, the most matching fun interactive activity or short story script is generated; the activity duration is dynamically determined based on the slope of the fatigue trend and the remaining teaching time of subsequent high-importance nodes. If there are three or more subsequent high-importance nodes, the activity duration is automatically compressed to the preset minimum value. If a user does not have a grasp of a certain knowledge point, supplementary explanations for that knowledge point will be generated. The level of detail in the supplementary explanations will be dynamically adjusted based on the number and depth of questions asked by the user. When it is determined that a user has not mastered the corresponding knowledge point, the corresponding knowledge point is traced back to all the prerequisite dependency nodes in the teaching cause-effect subgraph. Combined with the user group's historical learning data, the prerequisite knowledge points with mastery deficiencies are located. From the implicit knowledge units of error correction and difficulty breakthrough teaching anchored on the current knowledge points and the previous knowledge points of deficiencies, select content that matches the user's cognitive level and organize the supplementary explanation logic according to the causal order from the previous knowledge points to the current knowledge points. The level of detail in the supplementary explanations is positively correlated with the depth of knowledge points involved in the user's questions. At the same time, the depth of teaching content for subsequent less important nodes is automatically adjusted to ensure the integrity of the core causal chain. If the actual time spent on any teaching segment significantly exceeds the planned time, then the teaching time for subsequent knowledge points of lower importance will be reduced. If a user raises a question that is beyond the scope of the lesson but related to the teaching content, a brief extended explanation will be generated, or the administrator will be advised to conduct an in-depth discussion after class. The system adopts a cloud-edge collaborative architecture. The cloud deploys a complete cosmic package model and a multi-dimensional teaching knowledge graph, which is responsible for handling complex script generation tasks and model iteration optimization. The edge deploys a lightweight cosmic package model and a cache of commonly used teaching content, which is responsible for handling real-time classroom perception and dynamic adjustment tasks. The edge deployment includes a lightweight cosmic package model, cached common teaching content, and common game level templates, responsible for handling real-time classroom perception, local playback of game videos, interactive data collection, and dynamic adjustment tasks. When a network connection is available, the cloud synchronizes the teaching scripts and resources to the edge device in advance according to the administrator's teaching plan; when there is no network or the network bandwidth is low, the edge device runs independently, providing core script generation and dynamic adjustment functions to ensure the normal conduct of teaching activities. After the network is restored, the edge device will synchronize classroom feedback data and administrator's modification suggestions to the cloud for iterative optimization; Collect managers' feedback on the generated script and users' classroom learning feedback data; regularly conduct statistical analysis on the classroom learning feedback data to identify knowledge points and teaching segments where the model's generated quality is below the preset quality threshold; Based on the statistical results of classroom learning feedback data, we update the implicit knowledge units, interdisciplinary integration points, and special education adaptation strategies in the multi-dimensional teaching knowledge graph, while fine-tuning the parameters of the meta-learning basic model to improve the quality and personalization of script generation. The iteration frequency of the model is dynamically adjusted based on the changing trends of classroom learning feedback data. When classroom learning feedback data shows a significant decline in the quality of model generation, the iteration frequency is increased; when classroom learning feedback data shows that the quality of model generation is stable at a high level, the iteration frequency is decreased.

[0010] Example 2: As Figure 2 As shown in the figure, the educational course script generation method based on the fusion data processing of the cosmic package large model provided in this embodiment of the invention specifically includes the following steps: Step 1: Construct a multi-dimensional hierarchical knowledge graph for teaching, and uniformly anchor implicit knowledge units, interdisciplinary integration points, and special education adaptation strategies to corresponding nodes and edges. Use the relative ranking method to determine the importance of teaching nodes, and mark core knowledge points as key points for passing the script game. Step 2: Combine multi-dimensional hierarchical knowledge graph of teaching to establish user cognitive feature system and administrator teaching style feature system, train meta-learning basic model, construct pre-mapping table of teacher and student dual subject features and teaching preferences, and dynamically balance the degree of personalization and knowledge vision through feature overlap. Step 3: Combining the pre-mapping table of teacher and student dual subject characteristics and teaching preferences, the starting and ending knowledge points of the knowledge graph are determined through teaching objectives. The path containing the most high-importance nodes is selected as the optimal teaching path. Under causal constraints, text and script game videos are collaboratively generated. A unique identifier is generated for each content segment and stored in a distributed ledger. Step 4: Collect classroom audio and video data for the generated text and script game video, identify dynamic classroom events by combining the continuous change trend of the state, and dynamically adjust the teaching content. Adjust the model iteration frequency dynamically based on the change trend of teacher and student feedback.

[0011] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. An educational course script generation system based on fusion data processing of the cosmic package large model, characterized in that: Includes the following modules: Adaptation Module: Construct a multi-dimensional hierarchical knowledge graph for teaching, and uniformly anchor implicit knowledge units, interdisciplinary integration points, and special education adaptation strategies to corresponding nodes and edges. Use relative sorting method to determine the importance of teaching nodes and mark core knowledge points as key points for passing the script game. Mapping Module: Combining multidimensional hierarchical knowledge graphs for teaching, we establish a user cognitive feature system and a manager teaching style feature system, train a meta-learning basic model, construct a pre-mapping table of teacher and student dual subject features and teaching preferences, and dynamically balance the degree of personalization and knowledge scope through the degree of feature overlap. The generation module combines the characteristics of both teachers and students with a pre-mapping table of teaching preferences. It determines the starting and ending knowledge points of the knowledge graph through teaching objectives, selects the path containing the most high-importance nodes as the optimal teaching path, and collaboratively generates text and script game videos under causal constraints. It generates a unique identifier for each content segment and stores it in a distributed ledger. Adjustment Module: For the generated text and script game video, collect classroom audio and video data, identify dynamic classroom events by combining the continuous change trend of the state, and dynamically adjust the teaching content. The model iteration frequency is dynamically adjusted according to the change trend of teacher and student feedback.

2. The educational course script generation system based on cosmic package large model fusion data processing as described in claim 1, characterized in that, The method for constructing a multidimensional hierarchical knowledge graph for instruction is as follows: Construct a subject teaching causal sub-graph, an interdisciplinary knowledge association sub-graph, and a special education adaptation sub-graph; when constructing the subject teaching causal sub-graph, extract all explicit knowledge points from the textbook and curriculum standards, and establish the sequential dependency relationship between knowledge points according to the teaching order presented in the classroom recordings of top-level administrators; When constructing an interdisciplinary knowledge association subgraph, we explore the inherent connections between knowledge points from different disciplines from interdisciplinary topic learning guides and project-based learning cases. Each association edge records the association type and the applicable interdisciplinary topic. When constructing the special education adaptation submap, teaching adjustment strategies for different types of special users are extracted from the special education curriculum standards and classroom recordings of special education administrators, and four types of special education adaptation tags are added to each knowledge point node.

3. The educational course script generation system based on the fusion data processing of the large cosmic package model according to claim 1, characterized in that, The method for uniformly anchoring to the corresponding node and edge is as follows: The teaching experience content that does not contain explicit knowledge points is extracted from all classroom recordings, and then transformed into standardized implicit knowledge units by utilizing the long contextual understanding ability of the Zhoubao model. Anchoring implicit knowledge units in teaching to the corresponding nodes and edges of the subject teaching causal subgraph is applicable to anchoring units for explaining a single knowledge point to the knowledge point node; Anchor interdisciplinary integration points to the corresponding edges of the interdisciplinary knowledge association subgraph, with each integration point containing a specific interdisciplinary teaching activity design; anchor special education adaptation strategies to the corresponding labels of the special education adaptation subgraph.

4. The educational course script generation system based on cosmic package large model fusion data processing according to claim 1, characterized in that, The method for determining the importance of teaching nodes is as follows: Calculate the number of prerequisite dependencies and the total number of implicit knowledge units anchored for each knowledge point node; sort all knowledge point nodes from highest to lowest according to the sum of the number of prerequisite dependencies and the number of implicit knowledge units, and the sorting result is the relative importance of the teaching node. In the subsequent script generation process, priority is given to ensuring the allocation of teaching time and resources for knowledge points that are ranked higher, and richer teaching activities and more detailed explanations are automatically designed for knowledge points that are of high importance.

5. The educational course script generation system based on cosmic package large model fusion data processing according to claim 1, characterized in that, The method for obtaining the pre-mapping table of teaching preferences is as follows: Establish a teaching style characteristic system for administrators, which includes five dimensions: language style, teaching pace, interaction method, blackboard design and classroom management. Each dimension is divided into a predetermined number of typical style types. Collect anonymous user group learning data, group them according to the six dimensions of the user cognitive characteristic system, perform statistical analysis on each group of data, and calculate the average teaching needs of each group of users on each knowledge point. We collected teaching style data from anonymous managers, grouped them according to the five dimensions of the manager teaching style characteristic system, and calculated the typical teaching preferences of each group of managers in each teaching segment.

6. The educational course script generation system based on cosmic package large model fusion data processing according to claim 1, characterized in that, The method for dynamically balancing the degree of personalization is as follows: The statistically obtained teaching preference data of user groups and teaching preference data of administrator groups are input into the Zhoubao large model for meta-learning training to construct a pre-mapping table of teacher and student dual subject characteristics and teaching preferences; Based on the user's basic information input by the administrator and the administrator's own teaching style, the corresponding optimal teaching preference data is retrieved from the pre-mapping table; if a small amount of personal usage data of a single user or administrator is collected, the corresponding personal characteristics are compared with the group characteristics in the pre-mapping table, and the degree of overlap between the two in various dimensions is statistically analyzed. Adjust the degree of personalization and the proportion of extended content based on the degree of overlap. If the degree of overlap is high, use a higher degree of personalization; if the degree of overlap is low, reduce the degree of personalization and increase the extended content.

7. The educational course script generation system based on the fusion data processing of the large cosmic package model according to claim 1, characterized in that, The method for obtaining the optimal teaching path is as follows: The administrator inputs the teaching objectives, and determines the starting and ending knowledge points of the teaching from the multi-dimensional teaching knowledge graph based on the teaching objectives; searches all teaching paths from the starting knowledge point to the ending knowledge point, and each path contains a series of continuous knowledge point nodes and directed edges; Based on the ranking of the relative importance of the teaching nodes, the path containing the most important knowledge points is selected as the optimal teaching path. If the administrator selects interdisciplinary integration, interdisciplinary integration points related to the teaching content will be automatically inserted into the optimal teaching path; if the administrator selects special education adaptation, a corresponding special education adaptation strategy will be automatically added to each knowledge point in the optimal teaching path.

8. The educational course script generation system based on the fusion data processing of the large cosmic package model according to claim 1, characterized in that, The method for generating unique identifiers is as follows: Based on the optimal teaching path and the teaching preferences of teachers and students, the teaching process is divided into a predetermined number of teaching segments, and a reasonable teaching time is allocated to each teaching segment. Based on the needs of the teaching content, the system automatically generates corresponding script game videos, ensuring complete semantic alignment between all multimodal resources and text content; it generates a unique content identifier for each generated teaching content segment; and it uploads original teaching content created by administrators to the platform, generating a unique copyright certificate for the original content.

9. The educational course script generation system based on cosmic package large model fusion data processing according to claim 1, characterized in that, The method for identifying dynamic events in the classroom is as follows: Analyze the dialogue between users and administrators in the classroom to extract the number of times users spoke, the content of their questions, and the accuracy of their answers; use computer vision technology to analyze users' facial expressions and body movements to extract information on changes in users' attention and emotions. By analyzing the continuous changing trends of classroom status data, dynamic events in the classroom can be identified. If the number of times a user speaks continues to decrease in three consecutive teaching sessions, or if the user's level of concentration shows a significant downward trend, it can be determined that the user is experiencing learning fatigue. If, after any knowledge point has been explained, the user raises multiple related questions, or the user's correct answer rate is significantly lower than the average level, it is determined that the user has not firmly grasped the corresponding knowledge point.

10. The educational course script generation system based on cosmic package large model fusion data processing according to claim 1, characterized in that, The method for dynamically adjusting the model iteration frequency is as follows: The cloud deploys a complete cosmic package model and a multi-dimensional teaching knowledge graph, responsible for handling complex script generation tasks and model iteration optimization; the edge deploys a lightweight cosmic package model and cached commonly used teaching content, responsible for handling real-time classroom perception and dynamic adjustment tasks. When a network connection is available, the cloud synchronizes the teaching scripts and resources to the edge device in advance according to the administrator's teaching plan; when there is no network or the network bandwidth is low, the edge device runs independently, providing core script generation and dynamic adjustment functions; after the network is restored, the edge device synchronizes classroom feedback data and administrator's modification opinions to the cloud for iterative optimization.

Images