An intelligent multimedia interactive teaching and assessment system and method

Abstract

The present invention discloses an intelligent multimedia interactive teaching and assessment system and method, which relates to the field of educational information technology, including obtaining user information and learning status, generating learning portraits; based on the learning portraits, using generative AI to dynamically generate personalized teaching content, and marking the user's abnormal status in real time; storing the user's abnormal status marking results in the dynamic feature field of the learning portrait, and dynamically optimizing the learning path; based on the optimized learning path, dynamically generating personalized assessment tasks, and recording the user's answer data and learning behavior data; using the answer data and learning behavior data, through semantic analysis methods, to perform multi-dimensional evaluation of the user's performance. The present invention realizes multi-dimensional evaluation of learners' performance through comprehensive analysis models and multimodal feature vectors, covering multiple aspects such as knowledge mastery, behavioral performance and emotional state, and provides learners with detailed feedback and improvement suggestions.

Description

Technical Field

[0001] The present invention relates to the field of educational information technology, and in particular to an intelligent multimedia interactive teaching and assessment system and method. Background Art

[0002] With the rapid development of artificial intelligence technology and educational informatization, the application of intelligent teaching systems in personalized learning has become a research hotspot; the traditional education model is usually based on fixed teaching content and unified teaching methods, which makes it difficult to provide personalized guidance and dynamic adjustments based on individual differences of learners; in recent years, the rise of generative artificial intelligence technology has made it possible to dynamically generate teaching content and assessment tasks; generative AI combines semantic analysis and data-driven models to provide more dimensional intelligent support for the education system; in addition, learner behavior analysis and emotional state assessment have also become an important foundation for personalized teaching. Through real-time data collection and analysis, the learner's learning status can be more accurately portrayed.

[0003] However, existing technologies still have many shortcomings in the field of personalized teaching; for example, although some intelligent teaching systems can provide personalized recommendations based on learners' knowledge mastery, they usually lack in-depth mining of learning behavior data and emotional feedback data, resulting in an incomplete understanding of users' learning status; in addition, most existing assessment systems use statically set test questions, which makes it difficult to dynamically adjust task difficulty and form according to learners' real-time status, and cannot meet the needs of personalized assessment; therefore, there is an urgent need for a full-process intelligent teaching method that can dynamically generate teaching content and assessment tasks, and evaluate learners' performance in multiple dimensions. Summary of the invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides an intelligent multimedia interactive teaching and assessment method to solve the problem that the prior art does not fully understand the user's learning status and cannot meet the needs of personalized assessment.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] In a first aspect, the present invention provides an intelligent multimedia interactive teaching and assessment method, which includes obtaining user information and learning status, and generating a learning portrait;

[0008] Based on learning portraits, generative AI is used to dynamically generate personalized teaching content and mark users’ abnormal status in real time;

[0009] Store the user's abnormal status marking results in the dynamic feature field of the learning profile to dynamically optimize the learning path;

[0010] Based on the optimized learning path, personalized assessment tasks are dynamically generated, and the user's answer data and learning behavior data are recorded;

[0011] Using answer data and learning behavior data, a multi-dimensional evaluation of user performance is performed through semantic analysis methods.

[0012] As a preferred solution of the intelligent multimedia interactive teaching and assessment method of the present invention, wherein: the user information includes user attribute data, learning preference data and learning scenario data;

[0013] The learning status includes knowledge mastery data and learning behavior data;

[0014] The specific steps of generating a learning portrait are as follows:

[0015] Collect static features from user attribute data, learning preference data, and learning scenario data;

[0016] Collect dynamic features from knowledge acquisition data and learning behavior data;

[0017] Combine static features and dynamic features through data structures to form user data sets;

[0018] Perform data preprocessing on user data sets and construct multi-dimensional feature vectors;

[0019] Based on multi-dimensional feature vectors, a clustering model is used to analyze users in groups, generate personalized user labels, and describe individual learning portraits by combining user information and learning status;

[0020] Store the learning profiles in structured data.

[0021] As a preferred solution of the intelligent multimedia interactive teaching and assessment method of the present invention, wherein: based on the learning portrait, the generative AI is used to dynamically generate personalized teaching content, and the specific steps are as follows:

[0022] Read the stored learning profile structured data and extract static and dynamic features;

[0023] Based on static and dynamic features, extract and generate requirements;

[0024] Transform generation requirements into natural language structured input to guide generative AI in real time;

[0025] Dynamically generate teaching content through generative AI;

[0026] The teaching content is quality-checked and formatted, converted into a display format suitable for user terminals, and the teaching content is displayed in real time.

[0027] As a preferred solution of the intelligent multimedia interactive teaching and assessment method of the present invention, the specific steps of real-time marking of abnormal status of users are as follows:

[0028] Collect the interactive behavior data between users and teaching content in real time and transmit it to the backend through the real-time transmission protocol;

[0029] The backend cleans and time-aligns the interactive behavior data, and extracts learning behavior data and emotional feedback data;

[0030] Use the in-memory database to cache learning behavior data and emotional feedback data;

[0031] Build a comprehensive analysis model based on the learning behavior data and emotional feedback data in the in-memory database;

[0032] The comprehensive analysis model is used to evaluate the user's current learning status, calculate the comprehensive score, and mark abnormal status.

[0033] As a preferred solution of the intelligent multimedia interactive teaching and assessment method of the present invention, wherein: the abnormal state marking result of the user is stored in the dynamic feature field of the learning portrait, and the learning path is dynamically optimized. The specific steps are as follows:

[0034] Read the learning profile, store the abnormal type in the abnormal status information into the dynamic feature field, and store the updated learning profile into the memory database;

[0035] Read the learning profile from the memory database and extract the user's dynamic learning status features;

[0036] Combine the user's dynamic learning status characteristics, abnormal status marks and learning behavior data, and use the rule engine and data-driven model to generate learning path optimization suggestions;

[0037] Adjust the learning path according to the optimization suggestions and store the optimized results in the in-memory database.

[0038] As a preferred solution of the intelligent multimedia interactive teaching and assessment method of the present invention, wherein: based on the optimized learning path, the personalized assessment task is dynamically generated, and the user's answer data and learning behavior data are recorded. The specific steps are as follows:

[0039] Read the user's dynamic learning state features, abnormal state marks and learning behavior data from the optimized learning path, and construct a task to generate an input feature set;

[0040] Based on the task generation input feature set, the dynamic multi-objective Wasserstein distance is used to adjust the difficulty and diversity of the generated tasks in real time. The expression is:

[0041] ;

[0042] in, represents the dynamic multi-objective Wasserstein distance, Represents the total number of knowledge points, represents the normalization coefficient, Represents the index of the knowledge point, represents the mastery attenuation factor, Indicates the user's The quantitative value of the mastery level of each knowledge point, Indicates The difficulty coefficient of each knowledge point, represents the interest amplification factor, Indicates the user's The quantitative value of the interest preference intensity of each knowledge point, represents the abnormal state weight factor, Indicates the abnormal state strength value, represents the integral differential element, represents the periodic change of task difficulty, is a mathematical constant, is the time variable, is the upper limit of the integral, Represented by natural numbers The exponential function with base , Represented by natural numbers Logarithmic function with base ;

[0043] Setting basic task thresholds and medium task threshold ;

[0044] Input the task generation input feature set and dynamic Wasserstein distance into the generator;

[0045] When the generator outputs When it is a basic task, it is to review the content that has been mastered;

[0046] When the generator outputs When , it indicates a task of medium difficulty, which is targeted to strengthen the knowledge points that have not been mastered;

[0047] When the generator outputs When it is a high-difficulty task, it focuses on expandability and challenge;

[0048] Use the discriminator to evaluate the difficulty adaptability and diversity of the assessment tasks and optimize the generator output;

[0049] After the user completes the assessment task, the answer data is recorded in real time, combined with the input feature set generated by the task, and the user's dynamic learning status is updated.

[0050] As a preferred solution of the intelligent multimedia interactive teaching and assessment method of the present invention, wherein: the answer data and learning behavior data are used to perform a multi-dimensional evaluation of the user's performance through a semantic analysis method, and the specific steps are as follows:

[0051] Extract semantic vectors from answer data and generate multimodal feature representations in combination with learning behavior data;

[0052] The multimodal feature vector is combined with the domain knowledge graph to perform semantic reasoning for knowledge enhancement and calculate the matching score between the user's answer and the knowledge point. The expression is:

[0053] ;

[0054] in, Represented as user answers and knowledge points High dimensional representation of The matching score of The index variable representing the knowledge point, represents the vector concatenation operation, represents the bias vector, represents the multimodal feature vector, represents the weight matrix;

[0055] Based on user answers and knowledge points High dimensional representation of Matching scores and learning behavior data are used to calculate the user's multi-dimensional performance indicators;

[0056] Update the calculation results of the user's multi-dimensional performance indicators to the dynamic feature fields of the learning profile to optimize the dynamic learning profile;

[0057] Combine the optimized dynamic learning profile with the multi-dimensional evaluation results to dynamically generate assessment tasks.

[0058] In a second aspect, the present invention provides an intelligent multimedia interactive teaching and assessment system, including an information acquisition module, a content generation module, a path optimization module, a task generation module and a performance evaluation module;

[0059] The information acquisition module is used to obtain user information and learning status and generate a learning portrait;

[0060] The content generation module uses generative AI to dynamically generate personalized teaching content based on learning portraits and mark users' abnormal status in real time;

[0061] The path optimization module stores the abnormal status marking result of the user in the dynamic feature field of the learning profile and dynamically optimizes the learning path;

[0062] The task generation module dynamically generates personalized assessment tasks based on the optimized learning path, and records the user's answer data and learning behavior data;

[0063] The performance evaluation module uses the answer data and learning behavior data to perform a multi-dimensional evaluation of the user's performance through a semantic analysis method.

[0064] In a third aspect, the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, any step of the intelligent multimedia interactive teaching and assessment method as described in the first aspect of the present invention is implemented.

[0065] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, any step of the intelligent multimedia interactive teaching and assessment method as described in the first aspect of the present invention is implemented.

[0066] The beneficial effects of the present invention are as follows: the present invention significantly improves the personalization level of intelligent teaching by constructing a dynamic learning portrait and combining it with generative AI technology and semantic analysis methods; first, by deeply integrating the static and dynamic features of learners, a more comprehensive learning portrait is generated, which can accurately portray the learner's knowledge mastery, learning behavior characteristics and emotional feedback; second, by dynamically generating teaching content and assessment tasks based on the learning portrait, not only can the difficulty and form of the teaching content be adjusted in real time, but also the learning path can be dynamically optimized to make the learning process more adaptable; at the same time, the present invention realizes a multi-dimensional evaluation of learners' performance through a comprehensive analysis model and a multimodal feature vector, covering multiple aspects such as knowledge mastery, behavioral performance and emotional state, and provides learners with detailed feedback and improvement suggestions; in summary, the present invention significantly improves the adaptability of teaching content and assessment tasks, optimizes the learning experience, and provides a new technical implementation path for educational informatization. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use 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 ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0068] Figure 1 This is a flow chart of the intelligent multimedia interactive teaching and assessment method in Example 1.

[0069] Figure 2 This is a module diagram of the intelligent multimedia interactive teaching and assessment system in Example 1. DETAILED DESCRIPTION

[0070] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0071] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0072] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0073] Example 1, reference Figure 1 and Figure 2 , which is the first embodiment of the present invention, provides an intelligent multimedia interactive teaching and assessment method, comprising the following steps:

[0074] S1. Obtain user information and learning status, and generate a learning profile.

[0075] Furthermore, the user information includes user attribute data, learning preference data, and learning scenario data;

[0076] Specifically, user attribute data refers to the user's basic identity, background and other information (such as age, gender, learning stage, etc.); learning preference data refers to the user's learning style, interests and goals; learning scenario data refers to environmental information such as device type, time period, network status, etc. when the user is learning.

[0077] The learning status includes knowledge mastery data and learning behavior data;

[0078] Specifically, knowledge mastery data refers to the user's current mastery of knowledge points, weaknesses and strengths, etc.; learning behavior data refers to the user's learning frequency, learning time, task completion rate and other learning behavior data.

[0079] Collect static features from user attribute data, learning preference data, and learning scenario data;

[0080] Collect dynamic features from knowledge acquisition data and learning behavior data;

[0081] Combine static features and dynamic features through data structures to form user data sets;

[0082] Perform data preprocessing on user data sets and construct multi-dimensional feature vectors;

[0083] Specifically, data preprocessing includes cleaning, standardization and feature extraction. Data cleaning can be performed by removing outliers (using IQR or Z-score methods) and filling missing values ​​(mean value filling or interpolation methods); data standardization can use the Z-score method to convert data into zero mean and unit variance, or use Min-Max normalization to compress values ​​to a specific range; feature extraction can reduce dimensionality based on principal component analysis (PCA), or screen key features through feature selection algorithms (such as based on mutual information or feature importance).

[0084] Based on multi-dimensional feature vectors, a clustering model is used to analyze users in groups, generate personalized user labels, and describe individual learning portraits by combining user information and learning status;

[0085] Preferably, clustering analysis is performed on users by using a clustering model based on multidimensional feature vectors. First, static features (such as age, gender, learning preference, device type, etc.) and dynamic features (such as knowledge mastery level, learning behavior data, etc.) of users are collected, and the data is cleaned to remove missing values ​​and outliers. Then, standardization is performed to ensure that the features have consistent metrics, and key features are further extracted to construct multidimensional feature vectors. On this basis, a suitable clustering algorithm (such as K-Means, DBSCAN or hierarchical clustering) is selected to perform cluster analysis on the multidimensional feature vectors to form several cluster categories representing different user groups. According to the clustering results, personalized labels are generated for each type of user. These labels are refined by analyzing the statistical results of the main feature dimensions of each type of user, such as "visual learner", "logical learner" or "video preference type". Combining the personalized labels, static features and dynamic features of users, a learning portrait describing personal learning characteristics is further generated, including learning goals, knowledge mastery level, learning behavior patterns (such as learning frequency and task completion rate), and points of interest. Ultimately, the learning portrait is stored in the form of structured data, which is used for the subsequent dynamic generation of personalized teaching content and optimization of learning paths, thereby achieving more targeted and adaptive teaching design, significantly improving user learning efficiency and experience.

[0086] Store the learning profiles in structured data.

[0087] S2. Based on learning portraits, use generative AI to dynamically generate personalized teaching content and mark users' abnormal status in real time.

[0088] Furthermore, the stored learning profile structured data is read to extract static and dynamic features;

[0089] Based on static and dynamic features, extract and generate requirements;

[0090] Specifically, the user's learning background is determined based on static features, including their academic stage, interest areas, and content preferences obtained through analysis of their learning scenarios (such as commonly used equipment, learning time periods) and learning preference data (such as visual learning, auditory learning), for example, whether the user prefers graphic explanations, video teaching, or text reading; secondly, the user's current learning goals and knowledge points that need to be improved are clarified based on dynamic features, for example, weak links are identified based on the mastery of knowledge points, and the user's learning habits (such as learning time, completion rate, etc.) are analyzed based on learning behaviors; finally, the user's learning background, learning goals, and content preferences obtained through analysis of learning preference data are integrated to form a clear demand description, that is, the refined generated needs, which are used to guide the subsequent generation of teaching content.

[0091] Transform generation requirements into natural language structured input to guide generative AI in real time;

[0092] Dynamically generate teaching content through generative AI;

[0093] Specifically, according to the refined user learning background, learning goals and content preferences, they are organized into clear text descriptions to ensure that each demand point has a clear statement, such as the user's attribute information (such as "high school students"), learning goals (such as "mastering the application of quadratic functions"), and content form preferences (such as "explaining through charts and examples"); then, these needs are organized into structured natural language input in a logical order to ensure that the generative AI can accurately understand them, such as clearly expressing them in three parts: "user background", "learning goals" and "generation requirements"; finally, the text input is passed to the generative AI in real time to dynamically generate personalized teaching content that meets user needs, ensuring that the content is accurate and highly matched with the user portrait.

[0094] The teaching content is quality-checked and formatted, converted into a display format suitable for user terminals, and the teaching content is displayed in real time.

[0095] Specifically, the generated teaching content is directly loaded into the quality inspection process, and grammatical errors, incoherent expressions and possible knowledge errors are identified and corrected through grammar checking and semantic analysis, while language fluency and knowledge completeness are evaluated; after the inspection is completed, the presentation of the content is adjusted according to the device characteristics of the user terminal (such as screen size, resolution, and supported media formats), the text is re-segmented into a length suitable for screen display, and pictures and video files are compressed to adapt to device performance. At the same time, annotations, icons or interactive elements are added to key parts according to the characteristics of the content; then, the processed teaching content is converted into a display format that is suitable for the target terminal, such as the HTML structure displayed on the web page, the adaptive interface layout suitable for mobile devices, or the three-dimensional media format supported by virtual reality devices; during the display process, the user's interactive behavior with the teaching content (such as reading time, operation trajectory) is collected in real time, and the layout or display method of the content is dynamically adjusted according to these data to further enhance the user experience; ultimately, it is ensured that the output of each step can provide high-quality input for the next step, thereby realizing real-time and efficient display and dynamic optimization of the teaching content.

[0096] Furthermore, real-time monitoring of users’ learning behavior data and emotional feedback data is carried out as follows:

[0097] Collect the interactive behavior data between users and teaching content in real time and transmit it to the backend through the real-time transmission protocol;

[0098] Specifically, the front-end embedded tools are used to collect real-time data on the interaction between users and teaching content, including clicks, dwell time, answer records, page switching and other operations. At the same time, facial expression data is captured through the camera and voice signals are collected using the microphone. All data are timestamped to ensure synchronization and transmitted to the back-end for processing in a low-latency manner through the WebSocket protocol.

[0099] The backend cleans and time-aligns the interactive behavior data, and extracts learning behavior data and emotional feedback data;

[0100] Specifically, after the backend receives the interactive behavior data, it removes missing values, duplicate values, and outliers through a data cleaning step. At the same time, it performs time alignment on the multimodal data based on timestamps to ensure the synchronization of learning behavior data and emotional data. Then, it extracts learning behavior data (such as click frequency, dwell time, and answer accuracy) and emotional feedback data (such as facial expression emotion categories and voice emotional characteristics) from the cleaned data for subsequent analysis.

[0101] Use the in-memory database to cache learning behavior data and emotional feedback data;

[0102] Preferably, an in-memory database (such as Redis) can support high-concurrency access and fast read and write operations by temporarily storing learning behavior data and emotional feedback data in memory, providing timely data input for subsequent comprehensive analysis models, and synchronizing the processed results to persistent storage for subsequent use when necessary.

[0103] Build a comprehensive analysis model based on the learning behavior data and emotional feedback data in the in-memory database;

[0104] Specifically, the user learning behavior data and emotional feedback data are read from the memory database (such as Redis), merged according to the user ID, and a complete data structure containing behavioral characteristics and emotional characteristics is generated;

[0105] The Z-score method is used to standardize numerical features (such as click-through rate, video viewing time, correct answer rate, fatigue level, and confusion level) and convert them into zero mean and unit variance forms; One-Hot encoding is used for discrete features (such as emotional state) and converted into vector representation, and all features are spliced ​​into a unified input feature vector to form the input data of the comprehensive analysis model;

[0106] A comprehensive analysis model is constructed using a multi-layer perceptron (MLP). The relationship between behavior and emotional characteristics is modeled through the weight calculation and activation functions of the neural network (such as ReLU and Sigmoid). The user's comprehensive learning status score is output to evaluate the performance of the learning status.

[0107] The comprehensive analysis model is used to evaluate the user's current learning status, calculate the comprehensive score, and mark abnormal status.

[0108] Specifically, by collecting user behavior data and emotional feedback data in real time, using data cleaning and time alignment methods to remove outliers and ensure data synchronization, the cleaned features are organized into multimodal input vectors and input into the comprehensive analysis model. The comprehensive analysis model captures the nonlinear correlation between features through a deep learning algorithm, and uses the learned weights and bias values ​​to calculate the user's current learning status score, including the mastery of knowledge points, behavioral focus, and emotional engagement, etc. The score is then interpreted in multiple dimensions according to the preset threshold to evaluate the learning status. When a score indicator is found to be lower than the critical value, it is marked as an abnormality and annotated in detail based on the abnormality type. Finally, the abnormal status and evaluation results are integrated into the user's dynamic learning features and stored in the dynamic field of the learning portrait, providing real-time available update data for subsequent learning path optimization or assessment task generation.

[0109] S3. Store the user's abnormal status marking results in the dynamic feature field of the learning profile and dynamically optimize the learning path.

[0110] Furthermore, the learning profile is read, the abnormality type in the abnormal status information is stored in the dynamic feature field, and the updated learning profile is stored in the memory database;

[0111] Specifically, obtain the latest user learning portrait from the in-memory database or cache, extract dynamic feature fields and related basic information; identify new abnormal states (such as stagnant learning progress, high error rate, etc.) based on user learning behavior data or real-time monitoring results, and extract the abnormal type and trigger conditions; write the parsed abnormal state information (including abnormal type and related data) into the dynamic feature field of the learning portrait, overwriting or appending the latest state; synchronously store the updated learning portrait in the in-memory database to ensure that subsequent steps can quickly access the latest data.

[0112] Read the learning profile from the memory database and extract the user's dynamic learning status features;

[0113] Combine the user's dynamic learning status characteristics, abnormal status marks and learning behavior data, and use the rule engine and data-driven model to generate learning path optimization suggestions;

[0114] Specifically, first, extract the user's dynamic learning status characteristics, including learning progress, knowledge point mastery, task completion efficiency, etc., and combine abnormal status marks (such as high error rate, unfinished tasks, decreased attention, etc.) to form a full picture of the user's current learning status; at the same time, integrate the user's learning behavior data (such as learning time, interaction frequency, answer records, etc.) to analyze their learning behavior patterns and trends. Subsequently, the rule engine is used to process deterministic problems (such as exception handling or simple logical judgment under fixed rules), and the data-driven model is used to model and analyze complex problems (such as the potential effect or trend prediction of learning path optimization). Finally, the processing results of the rule engine and the prediction results of the data-driven model are combined to generate personalized learning path optimization suggestions for users, such as giving priority to review tasks of weak knowledge points, adjusting the difficulty level of learning tasks, or recommending task forms that are more suitable for users' learning habits, so as to improve learning efficiency and knowledge mastery, and provide a basis for the subsequent sorting, difficulty adjustment and form optimization of learning tasks.

[0115] Furthermore, the learning behavior data of users (such as learning time, interaction frequency, answering records, etc.) are integrated to analyze their learning behavior patterns and trends. The specific steps are as follows: In the data collection and cleaning stage, the user's learning time, interaction frequency, answering records and other behavioral data are first collected, and these data are cleaned to remove outliers and duplicate values, and the missing data are supplemented or eliminated to ensure the integrity and accuracy of the data. Subsequently, through feature extraction and processing, the sliding window analysis method is applied to time series data such as learning time and interaction frequency to extract the behavior change trend, and the user's participation and learning efficiency indicators are constructed by combining the interaction frequency and the correct answer rate. In the process of behavioral pattern and trend analysis, the time series analysis method (such as ARIMA model) is used to predict the future trend of users' learning behavior changes, and the user behavior is divided into different modes (such as high-frequency short-term learning, low-frequency long-term learning) through cluster analysis (such as K-Means or DBSCAN), and the typical behavioral characteristics are summarized. In addition, the rule engine is used to determine whether there is obvious abnormal behavior (such as a sudden decrease in learning time or a decrease in interaction frequency), and deep learning models (such as LSTM) are used to detect potential non-obvious abnormal behavior trends, providing data support for subsequent learning path optimization.

[0116] Adjust the learning path according to the optimization suggestions and store the optimized results in the in-memory database.

[0117] Specifically, first, the user's dynamic learning state characteristics, abnormal state markers and behavior data are extracted from the generated optimization suggestions, and the various task nodes in the learning path are reordered in combination with the processing results of the rule engine and the prediction results of the data-driven model. For example, the tasks of knowledge points that the user has not mastered are adjusted to the priority position of the path, while reducing the initial difficulty of the task; then, the task duration allocation is optimized according to the user's learning behavior pattern (such as high-frequency short-term learning or low-frequency long-term learning), more complex tasks are broken down into multiple small steps to reduce the user's cognitive burden, and the recommended form of the task (such as video, chart or text) is adjusted to a presentation method that is more in line with the user's preference; then, the deep reinforcement learning algorithm is used to further adjust the transition logic between task nodes, and by modeling the connectivity between different tasks, the learning path is ensured to form coherence and progression between knowledge points; finally, the optimized learning path is reorganized into structured data form and stored in the memory database, while synchronizing the user's dynamic learning state characteristics and optimization results, so that the subsequent task generation module can call in real time and dynamically adjust the subsequent learning plan.

[0118] S4. Based on the optimized learning path, dynamically generate personalized assessment tasks and record the user's answer data and learning behavior data.

[0119] Furthermore, the user's dynamic learning state features, abnormal state marks and learning behavior data are read from the optimized learning path to construct the task generation input feature set;

[0120] Among them, the task generation input feature set includes knowledge point mastery , Difficulty Coefficient , interest preferences and abnormal status intensity .

[0121] Based on the task generation input feature set, the dynamic multi-objective Wasserstein distance is used to adjust the difficulty and diversity of the generated tasks in real time. The expression is:

[0122] ;

[0123] in, Represents the dynamic multi-objective Wasserstein distance, which is used to adjust the overall difficulty of the generated task. The higher the value, the more complex the generated task. The value range is , Indicates the total number of knowledge points, that is, the total number of knowledge points involved in the user's current learning path. Represents the normalization coefficient, which is used to average the difficulty contribution of all knowledge points. Represents the index of the knowledge point, that is, the index of the currently calculated Knowledge points, represents the mastery attenuation factor, Indicates the user's The quantitative value of the mastery level of each knowledge point, the value range , Indicates The difficulty coefficient of each knowledge point, the value range , Represents the interest amplification factor, which is used to control the influence weight of the quantitative value of the user's interest preference strength on the difficulty of the generated task. The value is a positive real number. Indicates the user's The quantitative value of the interest preference intensity of each knowledge point, the value range , Represents the abnormal state weight factor, which is used to control the degree of adjustment of the abnormal state intensity value on the task difficulty. The value is a positive real number. Indicates the abnormal state strength value, represents the integral differential element, represents the time point in the task generation process, Indicates the periodic change of task difficulty, which is used to dynamically adjust the time-varying characteristics of task generation. is a mathematical constant that defines the period of the sine function. is a time variable, indicating the time point in the task generation process, is the upper limit of the integral, indicating the end point of the time range (such as the total duration of a learning cycle), Represented by natural numbers An exponential function with a base of , used to calculate the exponential decay effect of mastery. Represented by natural numbers A logarithmic function with base is used to calculate the amplification effect of the quantitative value of interest preference intensity on task generation;

[0124] The introduction of multi-objective Wasserstein distance is more dynamic and personalized than traditional task generation technology based on static rules or single-objective optimization. It realizes real-time adjustment of task difficulty and diversity by comprehensively considering user mastery, interest preferences, knowledge point difficulty and abnormal status. Its dynamic factors can simulate learning state fluctuations, generate richer task content, avoid monotony, and flexibly respond to abnormal status to provide targeted guidance tasks. Compared with the traditional method of task allocation based on fixed rules or single dimension, this technology significantly improves the accuracy of task generation and user learning experience.

[0125] Specifically, The calculation of relies on the commonly used mastery assessment models in the existing technology, such as Bayesian Knowledge Tracing (BKT) or Deep Knowledge Tracing (DKT), which is obtained by counting the user's mastery level on each knowledge point; Calculated based on the difficulty parameters of knowledge points or test questions in the existing technology, usually derived from the IRT (Item Response Theory) model or the difficulty coefficient predefined by domain experts It is based on the recommendation algorithms in the existing technology, such as collaborative filtering, matrix decomposition or implicit feedback modeling technology, and is obtained by summarizing the user's interest in each knowledge point.

[0126] Specifically, The specific value of is used in the dynamic Wasserstein distance formula to reflect the user's The familiarity of a knowledge point; when it is at a high value (close to 1), it means that the user is relatively familiar with the knowledge point, and the formula reduces its impact on the task difficulty through exponential decay; when it is at a low value (close to 0), it means that the user is not familiar with the knowledge point, and the formula amplifies it exponentially to make it the focus of task generation.

[0127] Specifically, The value of reflects the user's interest in the knowledge point; when it is at a high value (close to 1), it means that the user is very interested in the knowledge point. Amplify its importance and generate more tasks related to this knowledge point; when it is at a low value (close to 0): it means that the user has low interest in this knowledge point, the result of the logarithmic function in the formula tends to 0, and the priority of this knowledge point in task generation is reduced.

[0128] Setting basic task thresholds and medium task threshold ;

[0129] It should be noted that the basic task threshold and medium task threshold It is an important indicator for dividing the difficulty level of tasks, aiming to output the value through the generator Accurately classify task types. Among them, the basic task threshold Defined as the upper limit of low-difficulty tasks, it means that the generator output Value less than When the task threshold is 1, the task is suitable for consolidating the content that has been mastered and focusing on reviewing basic knowledge; Defined as the upper limit of medium-difficulty tasks, Tasks within the scope are suitable for strengthening knowledge points that have not yet been mastered, focusing on targeted skill improvement. and medium task threshold The numerical value can realize the accurate division of task difficulty and meet the needs of different learning objectives.

[0130] Specifically, the basic task threshold and medium task threshold The setting is based on the specific requirements of the generator output data distribution and task classification. The specific implementation steps are as follows: First, collect the output data of the generator on different task samples , and statistically analyze these data to obtain their value range and distribution characteristics. Then, by analyzing the distribution curve of the data, key statistical indicators such as the mean, median and standard deviation of the output value are calculated to provide a scientific basis for the definition of the threshold. On this basis, the threshold is initially set according to the data distribution. is the 30th percentile of the distribution, representing the upper limit of the basic task; at the same time, Set it as the 70th percentile of the distribution as the upper limit of the medium task. Then, set the initial basic task threshold and medium task threshold Apply it to the task division model to observe its actual impact on the task classification effect; if the classification is not accurate enough or does not meet expectations, adjust the quantile based on feedback and reset the threshold. Finally, confirm the final optimized basic task threshold and medium task threshold Finally, it is solidified into the task generation rules to guide the difficulty division and allocation of subsequent tasks, ensuring the scientificity and practicality of task division.

[0131] Input the task generation input feature set and dynamic Wasserstein distance into the generator;

[0132] When the generator outputs When it is a basic task, it is to review the content that has been mastered;

[0133] When the generator outputs When , it indicates a task of medium difficulty, which is targeted to strengthen the knowledge points that have not yet been mastered;

[0134] When the generator outputs When it is a high-difficulty task, it focuses on expandability and challenge;

[0135] Use the discriminator to evaluate the difficulty adaptability and diversity of the assessment tasks and optimize the generator output;

[0136] For example, for difficulty adaptability: according to the user's knowledge mastery and task difficulty The error is calculated based on the matching degree; for diversity: it is evaluated based on the scope of knowledge points covered by the task or the difference in form.

[0137] After the user completes the assessment task, the answer data is recorded in real time, combined with the input feature set generated by the task, and the user's dynamic learning status is updated.

[0138] Specifically, after the user completes the assessment task, the answers submitted by the user are recorded in real time, including text, voice, multiple-choice options or other forms of answers, and the answer time and operation behaviors such as clicks and slides are recorded. The collected answer data is then cleaned to remove missing values ​​and outliers; for voice data, speech-to-text technology is used to transcribe it into text; the behavioral data is unified in timestamp format to ensure synchronization, and then the text answer data is converted into semantic vectors, and combined with the behavioral data to generate multimodal feature representations, including semantic features and behavioral features. At the same time, the knowledge points involved in the answer are marked and the mastery is quantified, and the extracted answer features are combined with the task generation input features. The collected knowledge point difficulty coefficient, mastery attenuation factor, interest preference intensity and other features are fused to form a complete learning status update input vector. The fused features are input into dynamic update models such as deep reinforcement learning or Bayesian knowledge tracking models to update the user's knowledge mastery status, abnormal status markers and learning behavior characteristics. Based on the updated dynamic learning status, new learning path optimization suggestions are generated through the rule engine and data-driven model, including task sorting adjustment and dynamic difficulty allocation. Finally, the updated learning status is synchronously stored in the dynamic feature field of the user portrait, and visual feedback content is generated to show the user's current learning status and progress.

[0139] S5. Using answer data and learning behavior data, a multi-dimensional evaluation of user performance is performed through semantic analysis methods.

[0140] Furthermore, semantic vectors are extracted from the answer data (text / speech) and combined with learning behavior data (such as answer time, mouse trajectory) to generate multimodal feature representations;

[0141] Specifically, text answers are used to extract semantic vectors using BERT; voice answers are used to convert to text through Wav2Vec2.0 and obtain voice emotion features; learning behavior data are used to extract operational features through time series models (such as TCN). All extracted features are then normalized to ensure that data of different modalities have consistent scales. Next, the text semantic vectors, voice emotion features, and behavior features are concatenated to generate a preliminary multimodal feature representation, and the fusion process is further optimized through the attention mechanism, assigning dynamic weights to each modality to highlight the more important sources of information in the current learning scenario; finally, a fused multimodal feature vector is generated through a weighted operation, which not only retains the independent feature information of each modality, but also realizes the effective integration of cross-modal information through dynamic weight adjustment, providing support for subsequent semantic reasoning and user performance evaluation.

[0142] Input multimodal feature vectors and combine them with domain knowledge graphs to perform knowledge-enhanced semantic reasoning;

[0143] Specifically, K-BERT is used to dynamically match user answers with related nodes in the knowledge graph to expand the implicit semantics of the answers; the relationship between knowledge points is modeled through graph neural networks (GNNs) to generate high-dimensional representations of each knowledge point. ;

[0144] Calculate the matching score between the user's answer and the knowledge point. The expression is:

[0145] ;

[0146] in, Represented as user answers and knowledge points High dimensional representation of The degree of matching, The index variable representing the knowledge point, represents the vector concatenation operation, represents the bias vector, which is used to adjust the output of the linear transformation. represents the multimodal feature vector, Represents the weight matrix, which is used to perform linear transformation on the input features;

[0147] Based on user answers and knowledge points High dimensional representation of Matching scores and learning behavior data are used to calculate the user's multi-dimensional performance indicators;

[0148] Among them, the multi-dimensional performance indicators include knowledge mastery , depth of thinking , problem solving skills and emotional engagement ;

[0149] Specifically, knowledge mastery , the calculation steps are as follows: use a semantic analysis model (such as BERT or K-BERT) to process the user's answer and find the semantic relevance between the answer and each knowledge point in the knowledge graph; assign a matching score to each knowledge point, the higher the score, the better the knowledge point is mastered; collect the matching scores of all knowledge points and calculate the average of these scores; the result is the user's knowledge mastery, which indicates the overall mastery level of the relevant knowledge points in the current learning task.

[0150] Specific, deep thinking , the calculation steps are as follows: according to the knowledge points involved in the user's answer, find the corresponding positions of these knowledge points in the knowledge graph; the knowledge graph is usually a hierarchical structure, the deeper the level of the knowledge point, the more complex the knowledge point; find the hierarchical depth of all knowledge points matched in the user's answer, and record these depth values; calculate the average of these depth values, and the result is the user's thinking depth, which indicates the complexity of the knowledge points involved in his answer.

[0151] Specific, problem-solving skills ,The calculation steps are as follows: record the total time from the start of the assessment task to the submission of the answer by the user, which is called "answer time"; according to the conventional task standards, the shorter the answer time, usually means the higher the problem-solving ability; normalize the user's answer time and convert it into a score to ensure that the higher the score, the stronger the problem-solving ability; the final score is the evaluation result of the problem-solving ability.

[0152] Specific, emotional engagement , the calculation steps are as follows: perform sentiment analysis on the user's voice answer, extract voice features (such as speaking speed, intonation changes, pause time, etc.), and evaluate its emotional expression in combination with the voice emotion model; use the emotion model to generate an emotion score for each answer, the higher the score, the higher the emotional involvement; if the answer contains multiple voice segments, calculate the average of the emotion scores of all voice segments; the final average score is the emotional involvement, which indicates the user's emotional involvement level when answering.

[0153] The calculation results of the user's multi-dimensional performance indicators Updated the dynamic feature fields of learning portraits to optimize dynamic learning portraits;

[0154] Specifically, if KM (knowledge mastery) does not reach 70%, it means that the user's mastery of the current knowledge point is insufficient (i.e., the average matching score between the answer and the knowledge point is less than 70%), and the learning path should be adjusted to prioritize the basic review tasks of the relevant knowledge points to help users consolidate the core concepts;

[0155] If RD (depth of thinking) is less than 3 layers: it means that the knowledge points associated with the user's answer are not deep enough (that is, the average depth of matching in the knowledge graph is less than 3 layers). The learning path should be adjusted to add extension tasks that require logical reasoning or deep exploration to help users understand more complex knowledge points.

[0156] If PSA (problem solving ability) is lower than 0.6: it means that the user is inefficient in answering questions (i.e. the result of the problem solving ability scoring formula is lower than 0.6). The task design logic should be optimized, such as reducing the complexity of the question or breaking the task into smaller subtasks to improve the user's problem-solving efficiency.

[0157] If the EE (emotional engagement) score is lower than 50 points: it means that the user is not emotionally engaged enough or has low interest in the learning process (that is, the voice emotion analysis score is lower than 50 points). The task format should be adjusted, such as adding interactive content, introducing situational tasks, or adopting gamification design to stimulate user participation.

[0158] Combine the optimized dynamic learning profile with the multi-dimensional evaluation results to dynamically generate assessment tasks.

[0159] Specifically, based on the optimized dynamic learning portrait, the user's knowledge mastery, multi-dimensional performance indicators (such as thinking depth, emotional engagement, problem-solving ability) and abnormal state marking information are extracted, and these data constitute the input basis for task generation; then, based on the extracted input basis, the task difficulty is calculated through a formula, and the priority and adaptability of the assessment task are determined by comprehensively considering the difficulty of knowledge points, the intensity of user interest preferences and the impact of abnormal states; then, the dynamic generation mechanism is used to generate different types of tasks, including review tasks (consolidating mastered knowledge), targeted reinforcement tasks (improving weak links) and expansion tasks (increasing challenges); the generated tasks are checked for adaptability through difficulty and diversity analysis steps to ensure that the task content meets the user's learning status and needs; if the task adaptability is insufficient, the input basis is adjusted and the task is regenerated; finally, the optimized and generated task content is directly converted into a format that adapts to the user terminal and presented to the user in real time, while recording the user's behavior data and answer data in the process of completing the task, providing a basis for subsequent task generation, and ensuring a high degree of match between dynamic task adjustment and user learning needs.

[0160] Based on the input feature set, the assessment priority and task difficulty of each knowledge point are calculated. The dynamic Wasserstein distance formula is used to adjust the diversity and adaptability of tasks: if the user's mastery of a certain knowledge point (KM) is low, basic review tasks are generated first to help consolidate related concepts; if the user's thinking depth (RD) is insufficient, extended tasks are generated, and open questions are designed to guide users to think deeply; if the user's problem-solving ability (PSA) is low, the task is broken down into multiple small steps to reduce complexity; if the user's emotional engagement (EE) is low, more interactive or situational tasks are generated to guide users to actively participate.

[0161] The task generation input feature set and the dynamic Wasserstein distance formula are input into the task generator, and the assessment tasks are dynamically generated through the generative algorithm (such as the task generation model based on Transformer). The assessment task types include: basic tasks: review the content that has been mastered and consolidate memory; reinforcement tasks: design medium-difficulty tasks for weak links to strengthen weaknesses; expansion tasks: design high-difficulty or open tasks to challenge the user's thinking ability.

[0162] Use the discriminator to evaluate the difficulty adaptability and diversity of the generated tasks to ensure that the tasks are consistent with the user's current learning status. If not, adjust the input features of the generator and regenerate the tasks.

[0163] This embodiment also provides an intelligent multimedia interactive teaching and assessment system, including: an information acquisition module, a content generation module, a path optimization module, a task generation module and a performance evaluation module; the information acquisition module is used to obtain user information and learning status and generate a learning portrait; the content generation module, based on the learning portrait, uses generative AI to dynamically generate personalized teaching content, and marks the user's abnormal status in real time; the path optimization module stores the user's abnormal status marking results in the dynamic feature field of the learning portrait, and dynamically optimizes the learning path; the task generation module, based on the optimized learning path, dynamically generates personalized assessment tasks, and records the user's answer data and learning behavior data; the performance evaluation module uses the answer data and learning behavior data to perform multi-dimensional evaluation of the user's performance through a semantic analysis method.

[0164] This embodiment also provides a computer device suitable for the intelligent multimedia interactive teaching and assessment method, including: a memory and a processor; the memory is used to store computer executable instructions, and the processor is used to execute computer executable instructions to implement the intelligent multimedia interactive teaching and assessment method proposed in the above embodiment.

[0165] The computer device may be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covered on the display screen, or a key, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse, etc.

[0166] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, the intelligent multimedia interactive teaching and assessment method proposed in the above embodiment is implemented; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (Static Random Access Memory, referred to as SRAM), electrically erasable programmable read-only memory (Electrically Erasable Programmable Read-Only Memory, referred to as EEPROM), erasable programmable read-only memory (Erasable Programmable Read Only Memory, referred to as EPROM), programmable read-only memory (Programmable Red-Only Memory, referred to as PROM), read-only memory (Read-Only Memory, referred to as ROM), magnetic storage, flash memory, magnetic disk or optical disk.

[0167] In summary, the present invention significantly improves the personalization level of intelligent teaching by constructing a dynamic learning portrait and combining it with generative AI technology and semantic analysis methods; first, through the deep integration of learners' static and dynamic features, a more comprehensive learning portrait is generated, which can accurately portray the learners' knowledge mastery, learning behavior characteristics and emotional feedback; secondly, based on the dynamic generation of teaching content and assessment tasks based on the learning portrait, it can not only adjust the difficulty and form of the teaching content in real time, but also dynamically optimize the learning path to make the learning process more adaptable; at the same time, the present invention realizes a multi-dimensional evaluation of learners' performance through a comprehensive analysis model and a multimodal feature vector, covering multiple aspects such as knowledge mastery, behavioral performance and emotional state, and provides learners with detailed feedback and improvement suggestions; in summary, the present invention significantly improves the adaptability of teaching content and assessment tasks, optimizes the learning experience, and provides a new technical implementation path for educational informatization.

[0168] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. An intelligent multimedia interactive teaching and assessment method, characterized by: include, Obtain user information and learning status, and generate learning portraits; Based on learning portraits, generative AI is used to dynamically generate personalized teaching content and mark users’ abnormal status in real time; Store the user's abnormal status marking results in the dynamic feature field of the learning profile to dynamically optimize the learning path; Based on the optimized learning path, personalized assessment tasks are dynamically generated, and the user's answer data and learning behavior data are recorded; Using answer data and learning behavior data, a multi-dimensional evaluation of user performance is performed through semantic analysis methods; Based on the optimized learning path, personalized assessment tasks are dynamically generated, and the user's answer data and learning behavior data are recorded. The specific steps are as follows: Read the user's dynamic learning state features, abnormal state marks and learning behavior data from the optimized learning path, and construct a task to generate an input feature set; Based on the task generation input feature set, the dynamic multi-objective Wasserstein distance is used to adjust the difficulty and diversity of the generated tasks in real time. The expression is: ； in, represents the dynamic multi-objective Wasserstein distance, Represents the total number of knowledge points, represents the normalization coefficient, Represents the index of the knowledge point, represents the mastery attenuation factor, Indicates the user's The quantitative value of the mastery level of each knowledge point, Indicates The difficulty coefficient of each knowledge point, represents the interest amplification factor, Indicates the user's The quantitative value of the interest preference intensity of each knowledge point, represents the abnormal state weight factor, Indicates the abnormal state strength value, represents the integral differential element, represents the periodic change of task difficulty, is a mathematical constant, is the time variable, is the upper limit of the integral, Represented by natural numbers The exponential function with base , Represented by natural numbers Logarithmic function with base ; Setting basic task thresholds and medium task threshold ; Input the task generation input feature set and dynamic Wasserstein distance into the generator; When the generator outputs When it is a basic task, it is to review the content that has been mastered; When the generator outputs When , it indicates a task of medium difficulty, which is targeted to strengthen the knowledge points that have not been mastered; When the generator outputs When it is a high-difficulty task, it focuses on expandability and challenge; Use the discriminator to evaluate the difficulty adaptability and diversity of the assessment tasks and optimize the generator output; After the user completes the assessment task, the answer data is recorded in real time, combined with the input feature set generated by the task, and the user's dynamic learning status is updated.

2. The intelligent multimedia interactive teaching and assessment method according to claim 1, characterized in that: The user information includes user attribute data, learning preference data and learning scenario data; The learning status includes knowledge mastery data and learning behavior data; The specific steps of generating a learning portrait are as follows: Collect static features from user attribute data, learning preference data, and learning scenario data; Collect dynamic features from knowledge acquisition data and learning behavior data; Combine static features and dynamic features through data structures to form user data sets; Perform data preprocessing on user data sets and construct multi-dimensional feature vectors; Based on multi-dimensional feature vectors, a clustering model is used to analyze users in groups, generate personalized user labels, and describe individual learning portraits by combining user information and learning status; Store the learning profiles in structured data.

3. The intelligent multimedia interactive teaching and assessment method according to claim 2, characterized in that: Based on the learning portrait, the generative AI is used to dynamically generate personalized teaching content. The specific steps are as follows: Read the stored learning profile structured data and extract static and dynamic features; Based on static and dynamic features, extract and generate requirements; Transform generation requirements into natural language structured input to guide generative AI in real time; Dynamically generate teaching content through generative AI; The teaching content is quality-checked and formatted, converted into a display format suitable for user terminals, and the teaching content is displayed in real time.

4. The intelligent multimedia interactive teaching and assessment method according to claim 3, characterized in that: The specific steps of marking the abnormal status of the user in real time are as follows: Collect the interactive behavior data between users and teaching content in real time and transmit it to the backend through the real-time transmission protocol; The backend cleans and time-aligns the interactive behavior data, and extracts learning behavior data and emotional feedback data; Use the in-memory database to cache learning behavior data and emotional feedback data; Build a comprehensive analysis model based on the learning behavior data and emotional feedback data in the in-memory database; The user's current learning status is evaluated through a comprehensive analysis model, the comprehensive score is calculated, and abnormal status is marked.

5. The intelligent multimedia interactive teaching and assessment method according to claim 4, characterized in that: The abnormal state marking result of the user is stored in the dynamic feature field of the learning portrait, and the learning path is dynamically optimized. The specific steps are as follows: Read the learning profile, store the abnormal type in the abnormal status information into the dynamic feature field, and store the updated learning profile into the memory database; Read the learning profile from the memory database and extract the user's dynamic learning status features; Combine the user's dynamic learning status characteristics, abnormal status marks and learning behavior data, and use the rule engine and data-driven model to generate learning path optimization suggestions; Adjust the learning path according to the optimization suggestions and store the optimized results in the in-memory database.

6. The intelligent multimedia interactive teaching and assessment method according to claim 5, characterized in that: The answer data and learning behavior data are used to conduct a multi-dimensional evaluation of the user's performance through a semantic analysis method. The specific steps are as follows: Extract semantic vectors from answer data and generate multimodal feature representations in combination with learning behavior data; The multimodal feature vector is combined with the domain knowledge graph to perform semantic reasoning for knowledge enhancement and calculate the matching score between the user's answer and the knowledge point. The expression is: ； in, Represented as user answers and knowledge points High dimensional representation of The matching score of The index variable representing the knowledge point, represents the vector concatenation operation, represents the bias vector, represents the multimodal feature vector, represents the weight matrix; Based on user answers and knowledge points High dimensional representation of Matching scores and learning behavior data are used to calculate the user's multi-dimensional performance indicators; Update the calculation results of the user's multi-dimensional performance indicators to the dynamic feature fields of the learning profile to optimize the dynamic learning profile; Combine the optimized dynamic learning profile with the multi-dimensional evaluation results to dynamically generate assessment tasks.

7. An intelligent multimedia interactive teaching and assessment system, based on the intelligent multimedia interactive teaching and assessment method according to any one of claims 1 to 6, characterized in that: Including, information acquisition module, content generation module, path optimization module, task generation module and performance evaluation module; The information acquisition module is used to obtain user information and learning status and generate a learning portrait; The content generation module uses generative AI to dynamically generate personalized teaching content based on learning portraits and mark users' abnormal status in real time; The path optimization module stores the abnormal status marking result of the user in the dynamic feature field of the learning profile and dynamically optimizes the learning path; The task generation module dynamically generates personalized assessment tasks based on the optimized learning path, and records the user's answer data and learning behavior data; The performance evaluation module uses the answer data and learning behavior data to perform a multi-dimensional evaluation of the user's performance through a semantic analysis method.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the intelligent multimedia interactive teaching and assessment method described in any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the intelligent multimedia interactive teaching and assessment method described in any one of claims 1 to 6 are implemented.

Images