Ai-based online learning management system

Abstract

The AI-based online learning management system according to the present invention measures and analyzes the concentration level of students in real time in an online learning environment to provide customized lecture design support and feedback to instructors, and comprises: a customized design support module that receives the instructor's lecture goals and plans and generates a customized design plan; an AI evaluation feedback module that analyzes learning, evaluation, and attitude data based on AI and generates a feedback report; and a feedback reflection design engine that automatically transmits improvement suggestions generated by the AI ​​evaluation feedback module to the customized design support module and reflects them in the lecture design scenario. The AI ​​evaluation feedback module comprises: a concentration analyzer that collects facial expression and gaze data of students through a webcam and voice response data through a microphone to generate a concentration curve by time period; a statistical analyzer that calculates the average value, standard deviation, median, IQR, and histogram for the concentration curve; a student concentration profile manager that manages concentration records and fluctuation patterns for each student; a customized review video generator that identifies sections of decreased concentration and automatically generates review videos for those sections; and a feedback report generator that generates a report of the analysis results along with visualized statistics.

Description

Technology Field

[0001] The present invention relates to an AI-based online learning management system, and more specifically, to an online learning management system that utilizes a webcam and a microphone to measure and analyze the concentration of students in real time and provides instructors with support for customized lecture design and feedback. Background Technology

[0002] As the importance of online education has increased rapidly since the COVID-19 pandemic, there is a growing demand for improving the quality of education in remote learning environments.

[0003] Existing online learning platforms were limited to simply streaming lecture videos or providing recorded content, making it difficult to assess students' actual learning status or concentration levels.

[0004] As a result, instructors were unable to check students' understanding or participation in real time, making it difficult to conduct effective training, and students also experienced a decline in learning efficiency due to uniform educational content that did not match their individual learning levels.

[0005] Recently, online education platforms utilizing AI technology have emerged, but most are limited to simple learning progress management or grade analysis.

[0006] Furthermore, existing systems lack the functionality to objectively measure and analyze students' learning concentration, making it difficult for instructors to receive specific feedback for effective course design and improvement.

[0007] Particularly in online environments, there is a need for a system that can assess students' concentration in real time through facial expressions, eye contact, and vocal responses, and adjust lecture content based on this; however, platforms that comprehensively provide these functions have not yet been sufficiently developed.

[0008] Furthermore, there is an increasing demand for a system that classifies an instructor's speech into categories such as learning explanations, jokes, and encouragement, analyzes the impact of each speech type on student concentration, and automatically generates improved lecture designs based on this analysis. The problem to be solved

[0009] The present invention aims to provide an AI-based online learning management system that measures and analyzes the concentration of students in real time in an online learning environment and provides instructors with support for customized lecture design and improvement feedback.

[0010] In addition, the present invention provides a system that generates a concentration curve by time period by comprehensively analyzing facial expressions, gaze, and voice response data of students collected through a webcam and microphone, and automatically generates and provides customized review videos to individual students based on this.

[0011] In addition, the present invention automatically classifies an instructor's voice data by subtopics such as learning explanations, jokes, and encouragement, analyzes the impact of each utterance type on student concentration, and provides the instructor with specific suggestions for improving teaching methods.

[0012] In addition, the present invention provides a feedback loop structure capable of continuous improvement of lecture quality by automatically transmitting the analysis results of the AI ​​evaluation feedback module to the lecture design support module. means of solving the problem

[0013] The AI-based online learning management system according to the present invention measures and analyzes the concentration level of students in real time in an online learning environment to provide customized lecture design support and feedback to instructors, and comprises: a customized design support module that receives the instructor's lecture goals and plans and generates a customized design plan; an AI evaluation feedback module that analyzes learning, evaluation, and attitude data based on AI and generates a feedback report; and a feedback reflection design engine that automatically transmits improvement suggestions generated by the AI ​​evaluation feedback module to the customized design support module and reflects them in the lecture design scenario. The AI ​​evaluation feedback module comprises: a concentration analyzer that collects facial expression and gaze data of students through a webcam and voice response data through a microphone to generate a concentration curve by time period; a statistical analyzer that calculates the average value, standard deviation, median, IQR, and histogram for the concentration curve; a student concentration profile manager that manages concentration records and fluctuation patterns for each student; a customized review video generator that identifies sections of decreased concentration and automatically generates review videos for those sections; and a feedback report generator that generates a report of the analysis results along with visualized statistics.

[0014] It may also include a learning data management server that stores lecture videos, materials, user information, concentration profiles, and customized review content.

[0015] In addition, the above AI evaluation feedback module may include an instructor voice subtopic analyzer that classifies instructor voice by subtopic, such as learning explanations, jokes, and encouragement, and matches it with a concentration curve to evaluate the impact of each utterance type on concentration.

[0016] In addition, the instructor voice subtopic analyzer can transcribe the instructor's voice data and classify it into at least one subtopic among explanation of learning content, jokes, encouragement, questions, and feedback through natural language processing, and quantify the effect of improving concentration for each subtopic by matching the amount of change in the student's concentration with each subtopic segment.

[0017] In addition, the concentration analyzer can extract changes in facial expressions, gaze direction, and head movements from the student's webcam video, and analyze response frequency, voice tone, and engagement from the microphone voice to calculate a concentration score in the range of 0 to 100 in real time.

[0018] In addition, the customized review video generator can identify time intervals in an individual student's concentration curve that have decreased by more than 20% compared to a reference value, and automatically edit the lecture video of the corresponding interval to generate a personalized review video of 3 to 10 minutes in length.

[0019] In addition, the above-mentioned student concentration profile manager can accumulate and store concentration records by class, unit, and topic, and generate an individual learning characteristic profile by analyzing concentration patterns by time of day, changes in concentration by day of the week, and the degree of concentration maintenance by lecture length.

[0020] In addition, the above feedback report generator can automatically generate a comprehensive report including a concentration curve graph, statistical analysis results, concentration rankings by student, a concentration distribution histogram, an effect analysis by subtopic, and text suggestions for lecture improvement.

[0021] In addition, the above AI evaluation feedback module can detect the occurrence of a concentration gap when the concentration distribution exhibits a bimodal shape or the standard deviation exceeds a reference value, and automatically propose separate support measures for the group of students with reduced concentration.

[0022] In addition, the above system can sequentially perform the steps of inputting lecture objectives, requesting lecture design, providing lecture content, collecting concentration data, generating concentration curves by time period, statistical analysis, collecting instructor voice data and classifying subtopics, analyzing concentration and matching subtopics, updating student concentration profiles, generating AI evaluation feedback, generating customized review videos, generating feedback reports, designing based on feedback, proposing design changes, finalizing instructor designs, and distributing new lecture plans. Effects of the invention

[0023] According to the AI-based online learning management system of the present invention, by utilizing a webcam and a microphone to measure and analyze the concentration of students in real time, it supports instructors in conducting effective education even in an online environment.

[0024] In addition, the present invention can provide a personalized learning experience by objectively identifying a student's learning pattern through the generation of concentration curves by time period and statistical analysis, and by automatically identifying sections where concentration drops to generate individually customized review videos.

[0025] Furthermore, the present invention presents specific improvement measures to enable instructors to develop more effective teaching methods by analyzing instructors' voice data by subtopic and matching it with concentration curves to quantitatively evaluate the educational effectiveness of specific speech types.

[0026] In addition, the present invention can achieve continuous improvement in lecture quality and optimization of educational effects through a feedback loop structure that automatically reflects AI evaluation feedback in lecture design. Brief explanation of the drawing

[0027] FIG. 1 is a block diagram showing an AI-based online learning management system according to one embodiment of the present invention. FIG. 2 is a block diagram showing an online learning service providing server according to one embodiment. FIG. 3 is a block diagram showing a learning data management server according to one embodiment. Specific details for implementing the invention

[0028] Embodiments of the present invention will be described below with reference to the attached drawings. Unless otherwise specifically defined or mentioned, terms indicating direction used in this description are based on the state shown in the drawings. Additionally, throughout each embodiment, the same reference numeral indicates the same component. Meanwhile, the thickness or dimensions of each component shown in the drawings may be exaggerated for convenience of explanation and do not imply that it must actually be constructed according to the corresponding dimensions or ratios between components.

[0030] With reference to FIG. 1, the overall configuration of an AI-based online learning management system according to an embodiment of the present invention will be described. FIG. 1 is a block diagram showing an AI-based online learning management system according to an embodiment of the present invention.

[0031] An AI-based online learning management system includes an online learning service provision server (100), a learning data management server (200), and a client terminal (300). The online learning service provision server (100) is a central server that manages and provides online learning services between students and instructors, and performs functions such as supporting lecture design, generating AI evaluations and feedback, and managing data storage. The learning data management server (200) acts as a database server that stores and manages user information, content, and analysis results in conjunction with the online learning service provision server (100). The client terminal (300) provides a user interface that allows students and instructors to access the system and use the services via the web or an application.

[0032] The online learning service providing server (100) and the learning data management server (200) are connected via a wired or wireless network and perform real-time data communication using REST API or GraphQL protocols. The client terminal (300) performs secure communication via the HTTPS protocol and can collect real-time webcam and microphone data using WebRTC technology.

[0034] Referring to FIG. 2, the configuration of an online learning service providing server (100) will be described. FIG. 2 is a block diagram showing an online learning service providing server according to one embodiment.

[0035] The online learning service providing server (100) includes a custom design support module (101), an AI evaluation feedback module (102), and a feedback reflection design engine (103).

[0036] The custom design support module (101) generates and recommends a lecture scenario based on information such as lecture objectives, topics, and target student demographics entered by the instructor, and reflects the analysis results of the AI ​​evaluation feedback module (102). The instructor can modify and finalize the proposed design, and the system saves the finalized design as the next lecture plan.

[0037] The customized design support module (101) includes a lecture goal analyzer, a target student group profiling engine, a content recommendation engine, and a scenario generator. The lecture goal analyzer analyzes the educational goals entered by the instructor using natural language processing technology and classifies lecture types such as knowledge transfer type, skill acquisition type, discussion type, and project type. The target student group profiling engine comprehensively analyzes the learner's age group, prior knowledge level, learning preference, concentration pattern, etc., to set a customized educational direction.

[0038] The content recommendation engine utilizes a deep learning-based collaborative filtering algorithm to analyze successful cases of lectures with similar goals, and extracts and recommends content elements that showed high levels of concentration within those lectures. The scenario generator uses a GPT-based conversational AI model to automatically generate optimized scenarios for the lecture introduction, body, and conclusion stages.

[0039] The AI ​​evaluation feedback module (102) analyzes learning, evaluation, and attitude data to provide feedback to the instructor and support the improvement of the lecture. The AI ​​evaluation feedback module (102) includes an concentration analyzer (1021), a statistical analyzer (1022), a student concentration profile manager (1023), a customized review video generator (1024), a feedback report generator (1025), and an instructor voice subtopic analyzer (1026).

[0040] The concentration analyzer (1021) collects webcam and microphone data in real time to generate a concentration curve over time. The concentration analyzer (1021) includes a face recognition module, an eye tracking module, an expression analysis module, a voice analysis module, and a concentration score calculation module.

[0041] The face recognition module utilizes the OpenCV library and the Dlib face landmark detection algorithm to detect and track the student's face region in webcam footage in real time. The eye tracking module analyzes pupil position and head direction to determine whether the student is looking at the screen. The facial expression analysis module uses a CNN-based emotion recognition model to classify emotional states such as concentration, confusion, boredom, and interest from the student's facial expressions.

[0042] The voice analysis module performs voice activity detection (VAD), voice energy analysis, and utterance frequency measurement from the student's microphone input. The voice analysis module utilizes the LibROSA library to extract Mel-Frequency Cepstrum Coefficient (MFCC) features and analyzes the emotional tone and engagement of the voice.

[0043] The concentration score calculation module calculates a concentration score between 0 and 100 in real time by combining the results of face recognition, eye tracking, facial expression analysis, and voice analysis. The concentration score is calculated by applying weights of 40% for gaze rate, 30% for facial expression analysis results, 20% for voice participation, and 10% for face detection stability.

[0044] The statistical analyzer (1022) calculates the mean, standard deviation, median, IQR, and histogram for the concentration curve to interpret the distribution and characteristics of the concentration. The statistical analyzer (1022) includes a descriptive statistics analysis module, a distribution shape analysis module, and an outlier detection module.

[0045] The descriptive statistics analysis module utilizes the NumPy and SciPy libraries to calculate the mean, standard deviation, variance, minimum, maximum, and quartiles of the concentration data. The distribution shape analysis module tests the normality of the concentration distribution and identifies unimodal, bimodal, and multimodal distributions. Distribution shape analysis is performed using the Shapiro-Wilk normality test and the Kolmogorov-Smirnov test.

[0046] The outlier detection module utilizes the IQR and Z-score methods to identify students with abnormally high or low concentration values. For students classified as outliers, it proposes separate learning support measures.

[0047] The student-specific concentration profile manager (1023) manages the cumulative concentration records and fluctuation patterns for each student. The student-specific concentration profile manager (1023) includes a personal profile creation module, a pattern analysis module, and a predictive modeling module.

[0048] The individual profile creation module accumulates and stores records of each student's concentration levels by class, unit, and topic in a database. The pattern analysis module analyzes changes in concentration by time of day, differences in concentration by day of the week, and the degree of concentration maintenance based on lecture length. The pattern analysis module utilizes the ARIMA model, a time-series analysis technique, to model the trends in individual students' concentration changes.

[0049] The predictive modeling module predicts concentration levels in future lectures based on past concentration patterns and provides advance notifications for when a decline in concentration is expected.

[0050] The customized review video generator (1024) identifies sections of low concentration, edits and extracts the corresponding parts, and automatically generates a customized review video for the student. The customized review video generator (1024) includes a low concentration section detection module, a video editing module, and a personalized recommendation module.

[0051] The degradation detection module automatically identifies time intervals in an individual student's concentration curve where the level drops by more than 20% compared to the individual average. The video editing module utilizes the FFmpeg library to automatically edit the lecture video of the identified degradation intervals and generates a review video ranging from 3 to 10 minutes in length. During the video editing process, an additional 30 seconds are included before and after the relevant section to provide context, and accessibility is enhanced through an automatic subtitle generation function.

[0052] The personalized recommendation module customizes the playback speed, repetition sections, and additional explanatory materials of review videos for each individual, taking into account the student's learning preferences and level of understanding.

[0053] The feedback report generator (1025) generates a report with the analysis results and visualized statistics. The feedback report generator (1025) includes a data visualization module, a report template engine, and an improvement suggestion generation module.

[0054] The data visualization module utilizes Matplotlib, Seaborn, and Plotly libraries to generate concentration curve graphs, statistical analysis results, concentration rankings by student, and concentration distribution histograms. The report template engine provides various report templates based on instructor requirements and can output in PDF or HTML format.

[0055] The improvement suggestion generation module generates specific lecture improvement plans in natural language based on the results of concentration analysis. The suggestions include analysis of when concentration drops, recommendations for effective utterance types, suggestions for adjusting lecture pace, and methods to enhance interaction.

[0056] The instructor voice subtopic analyzer (1026) classifies and tags the instructor voice by subtopic, such as learning explanation, jokes, and encouragement, and evaluates the concentration impact of specific utterance types by matching analysis with concentration curves. The instructor voice subtopic analyzer (1026) includes a voice transcription module, a subtopic classification module, a time synchronization module, and an effect analysis module.

[0057] The speech transcription module converts the instructor's speech into text in real time using the Google Speech-to-Text API or the OpenAI Whisper model. The subtopic classification module utilizes a BERT-based natural language processing model to automatically classify the transcribed text into subtopics such as explanations of learning content, jokes, encouragement, questions, feedback, and examples.

[0058] The time synchronization module accurately matches the utterance time of each subtopic with the students' concentration curve. The effect analysis module measures the change in concentration before and after utterance for each subtopic and quantifies the concentration improvement effect for each subtopic by testing statistical significance. Paired t-tests and ANOVA analysis are utilized in the effect analysis to verify the statistical significance of the effect for each subtopic.

[0060] The feedback-reflecting design engine (103) automatically transmits improvement suggestions generated by the AI ​​evaluation feedback module (102) to the custom design support module (101) and enables them to be reflected in the lecture design scenario. The feedback-reflecting design engine (103) includes a feedback collection module, an improvement proposal generation module, and a design update module.

[0061] The feedback collection module collects and integrates analysis results and improvement suggestions from each component of the AI ​​evaluation feedback module (102). The improvement proposal generation module generates specific lecture design changes based on the collected feedback. The design update module transmits the generated improvement proposal to the scenario generator of the custom design support module (101) to automatically reflect it in the next lecture design.

[0062] In the feedback incorporation process, instructor approval is obtained to ensure that automated improvement suggestions align with the instructor's educational philosophy.

[0064] The configuration of the learning data management server (200) is described with reference to FIG. 3. FIG. 3 is a block diagram showing a learning data management server according to one embodiment.

[0065] The learning data management server (200) includes a content repository (201), a user profile / history DB (202), an evaluation / test DB (203), an intensity profile DB (204), and a customized review content repository (205).

[0066] The content repository (201) stores and manages educational content such as lecture videos, educational materials, assignments, and reference materials. The content repository (201) includes a video storage module, a document management module, and a metadata management module.

[0067] The video storage module efficiently stores lecture videos of various resolutions and formats and performs adaptive bitrate encoding for streaming services. The document management module stores educational materials such as PDFs, PPTs, and DOCs with version control. The metadata management module manages information such as creation date, modification date, tags, categories, and access rights for each piece of content.

[0068] The content repository utilizes Content Delivery Network (CDN) technology to provide fast content delivery to users worldwide and ensures data stability by establishing an automated backup and disaster recovery system.

[0069] The user profile / history DB (202) manages account information, learning history, and performance records of students and instructors. The user profile / history DB (202) includes a user account management module, a learning history tracking module, and a performance analysis module.

[0070] The user account management module is responsible for user authentication, permission management, and privacy protection. The learning history tracking module tracks and stores course enrollment, progress rates, assignment submission status, and exam records in real time. The performance analysis module analyzes trends in learning outcomes and measures individual learning effectiveness.

[0071] The evaluation / test DB (203) stores and manages a question bank, test results, and evaluation criteria. The evaluation / test DB (203) includes a question bank management module, a test result storage module, and an evaluation analysis module.

[0072] The question bank management module classifies and manages various types of questions, such as multiple-choice, short-answer, and descriptive questions. The test result storage module stores detailed information such as answers, scores, and time taken for each student. The evaluation analysis module evaluates the quality of questions by analyzing the correct answer rate, discrimination, and difficulty level for each item.

[0073] The concentration profile DB (204) stores concentration curves, statistical analysis results, and cumulative records for each student. The concentration profile DB (204) includes a real-time data storage module, a statistical data management module, and a pattern analysis result storage module.

[0074] The real-time data storage module efficiently stores second-by-second concentration scores generated by the concentration analyzer. The statistical data management module enables rapid retrieval by pre-calculating and storing daily, weekly, and monthly concentration statistics. The pattern analysis result storage module stores individual concentration patterns, prediction model parameters, and outlier detection results.

[0075] The customized review content repository (205) stores automatically generated customized review videos and related metadata. The customized review content repository (205) includes a personalized review video management module, a recommendation algorithm data module, and a usage statistics tracking module.

[0076] The individual review video management module systematically classifies and stores review videos generated for each student. The recommendation algorithm data module collects effectiveness data for review videos and utilizes it as training data for generating improved review videos in the future. The usage statistics tracking module measures the viewership, re-viewing rates, and learning effectiveness of review videos to continuously improve the personalized recommendation system.

[0077] The client terminal (300) includes a custom-designed UI for instructors (301), an AI feedback report UI for instructors (302), a custom review video provision UI for students (303), and a notification / communication module (304).

[0078] The instructor-customized design UI (301) provides a user interface for creating and finalizing AI-recommended lecture designs. The instructor-customized design UI (301) includes a lecture information input interface, a design recommendation screen, a scenario editing tool, and a design saving function.

[0079] The lecture information input interface provides an intuitive form for entering lecture objectives, target student demographics, estimated duration, and key learning content. The design recommendation screen allows for the visual comparison of various AI-generated lecture scenario options. The scenario editing tool offers the ability to rearrange and modify lecture components using a drag-and-drop method.

[0080] The AI ​​feedback report UI (302) for instructors provides statistical analysis, concentration curves, and results of effect analysis by subtopic. The AI ​​feedback report UI (302) for instructors includes a dashboard screen, a detailed analysis screen, an improvement suggestion screen, and a comparative analysis screen.

[0081] The dashboard screen provides summary information that allows you to grasp key metrics at a glance. The detailed analysis screen enables interactive exploration of concentration curves, statistical charts, and detailed analysis results by student. The improvement suggestion screen presents specific lecture improvement plans generated by AI, organized by priority.

[0082] The UI (303) for providing personalized review videos for students provides an interface that allows viewing review videos during individual periods of low concentration. The UI (303) for providing personalized review videos for students includes a review video list screen, a video player, a study note function, and a progress tracking function.

[0083] The review video list screen displays individually created review videos organized by subject and date. The video player provides learning convenience features such as playback speed control, loop playback, and subtitle display. The study notes function allows users to take notes on and save important content during review.

[0084] The notification / communication module (304) supports communication between users, such as announcements, notifications, messengers, and forums. The notification / communication module (304) includes a real-time notification system, a messaging system, a forum management system, and a video conferencing integration function.

[0085] The real-time notification system sends alerts for missed progress, course closure notices, real-time warnings for declining concentration, and assignment submission deadlines via push notifications and email. The messaging system provides 1:1 messaging, group chat, and file sharing features between instructors and students. The forum management system operates a Q&A board, discussion forums, and a learning material sharing space.

[0086] The video conferencing integration feature supports real-time online classes by integrating with external video conferencing platforms such as Zoom, Google Meet, and Microsoft Teams, and imports participation data collected from these platforms into the system to utilize for concentration analysis.

[0087] The key operational processes of an AI-based online learning management system include the lecture design phase, the lecture execution phase, the analysis and feedback phase, and the improvement reflection phase.

[0088] The lecture design phase includes steps S100 to S120. In step S100, the instructor inputs information such as lecture objectives, target student demographics, and topics through the instructor-customized design UI (301) of the client terminal (300). Lecture objectives are classified into knowledge transfer, skill acquisition, and improvement of problem-solving abilities, and the target student demographics are subdivided into age groups, prior knowledge levels, learning preferences, etc.

[0089] In step S110, the custom design support module (101) analyzes input information to generate and recommend an AI-based lecture design plan. During the design plan generation process, components that showed high levels of concentration in past lectures under similar conditions are prioritized. The generated design plan includes scenarios for the lecture introduction, main body, and conclusion stages, as well as estimated time required and interaction methods.

[0090] In step S120, the instructor reviews the recommended design plan, modifies it as necessary, and finalizes it. The finalized lecture design plan is stored in the content repository (201) of the learning data management server (200), and content for conducting the lecture is prepared.

[0091] The lecture progress and data collection stage includes steps S200 to S229. In step S200, the concentration analyzer (1021) collects webcam and microphone data of students in real time during the lecture. Webcam data collection is performed at a resolution of 30fps, and preprocessing for face recognition and eye tracking is performed in real time. Microphone data is collected at a sampling rate of 44.1kHz, and background noise removal and voice activity detection are performed.

[0092] In step S210, the concentration analyzer (1021) comprehensively analyzes the collected data to generate a concentration curve for each individual student over time. Concentration scores are calculated at 10-second intervals and stored in the concentration profile DB (204) in real time. During the process of generating the concentration curve, a smoothing algorithm is applied to correct for temporary data loss or noise.

[0093] In step S220, the statistical analyzer (1022) calculates descriptive statistics for the generated concentration curve. The mean, standard deviation, median, interquartile range, minimum, and maximum values ​​are calculated, and the shape of the concentration distribution is analyzed. If the concentration distribution shows a bimodal shape, it suggests the possibility that there is a gap in understanding between learner groups, and further analysis is performed.

[0094] The voice analysis and subtopic matching step includes steps S225 through S229. In step S225, the instructor voice subtopic analyzer (1026) collects the instructor's voice data in real time and automatically records it during the lecture. The voice data is stored in a lossless compression format and encryption is applied for privacy protection.

[0095] In step S226, the speech transcription module converts the collected speech data into text, and the subtopic classification module automatically classifies the transcribed text into subtopics such as training explanations, jokes, encouragement, questions, feedback, and example explanations. A BERT model specialized for the training domain is used to improve classification accuracy, and sections with low classification confidence are marked for manual review.

[0096] In step S227, the time synchronization module accurately matches the utterance time of each subtopic with the student's concentration curve. Time synchronization between voice data and concentration data is performed based on timestamps, and time errors caused by network delays are automatically corrected.

[0097] In step S228, the effect analysis module measures the change in concentration before and after utterance for each subtopic. The impact on concentration for each subtopic is calculated by comparing the change in concentration in the 30-second interval before the start of the utterance and the 30-second interval after the end of the utterance. In step S229, the effect on concentration for each subtopic is quantitatively evaluated, and statistical significance is tested to identify specific results to be included in the instructor feedback.

[0098] The profile management and feedback generation steps include steps S230 through S330. In step S230, the student-specific concentration profile manager (1023) updates the concentration data of individual students as a cumulative record. The individual profile stores daily, weekly, and monthly concentration statistics along with trends in changes to learning patterns.

[0099] In step S300, the feedback report generator (1025) synthesizes all collected analysis results to generate AI evaluation feedback. The generated feedback includes an overview of the overall class concentration, detailed analysis by student, analysis of effectiveness by subtopic, and suggestions for improvement.

[0100] In step S310, the customized review video generator (1024) identifies the section where an individual student’s concentration drops and automatically generates a customized review video. During the review video generation process, related supplementary materials are added along with the lecture content of the section where concentration drops, and the explanation speed and difficulty are adjusted to suit the individual’s learning level.

[0101] In step S320, the feedback report generator (1025) generates a comprehensive report for the instructor. The report includes data visualization charts, statistical analysis results, and improvement suggestions, and is organized in an intuitive format so that the instructor can easily understand it. In step S330, the generated report is provided to the instructor through the instructor AI feedback report UI (302).

[0102] The feedback incorporation and design improvement steps include steps S400 to S430. In step S400, the feedback incorporation design engine (103) analyzes the generated AI feedback to derive improvements to be incorporated into the next lecture design. The improvements include methods to improve the point at which concentration drops, suggestions for utilizing effective speech types, and methods to adjust the lecture structure.

[0103] In step S410, the feedback-reflecting design engine (103) generates a modified lecture design based on the derived improvements. The new design is generated in a way that improves upon the problems of the existing design and strengthens the components that showed high concentration.

[0104] In the S420 stage, the instructor reviews the design changes proposed by the AI ​​and makes the final decision. Based on their educational philosophy and experience, the instructor may accept or modify the AI ​​suggestions, but the final decision-making authority rests with the instructor. In the S430 stage, the newly confirmed lesson plan is saved as the next lesson plan, completing the cycle of continuous educational quality improvement.

[0105] The AI-based online learning management system implements a streaming architecture utilizing Apache Kafka for processing large volumes of real-time data. Webcam and microphone data are collected in real-time via the WebRTC protocol, and the collected data is distributed to each analysis module for processing through Kafka message queues.

[0106] To optimize data processing performance, real-time intensity score calculation and temporary storage are performed using Apache Spark Streaming and Redis in-memory database. For long-term storage, MongoDB and InfluxDB are utilized to support the efficient storage and retrieval of time-series data.

[0107] The face recognition model for concentration analysis utilizes a lightweight CNN model based on MobileNetV3 to ensure real-time processing performance. The emotion recognition model uses a ResNet-50 model pre-trained on the FER-2013 dataset, fine-tuned using the training domain data.

[0108] The subtopic classification model for natural language processing utilizes a KoBERT model additionally trained with training-related text. To continuously improve the model's performance, an MLOps pipeline is established to perform automatic retraining on new data and model version management.

[0109] The system implements a multi-layered security scheme to comply with the GDPR and the Personal Information Protection Act. All biometric data is stored using AES-256 encryption, and the TLS 1.3 protocol is used during transmission. Face recognition data minimizes the risk of personal information exposure by storing only extracted feature vectors instead of the original images.

[0110] Only concentration pattern data with personally identifiable information removed through the application of data anonymization technology is used for research and system improvement purposes, and users can request the deletion of their personal data at any time.

[0112] Although preferred embodiments of the present invention have been described above, the technical concept of the present invention is not limited to the preferred embodiments described above and can be implemented in various ways within the scope that does not depart from the technical concept of the present invention as embodied in the claims. Explanation of the symbols

[0113] 100: Online learning service provider server 101: Custom Design Support Module 102: AI Evaluation Feedback Module 1021: Concentration Analyzer 1022: Statistical Analyzer 1023: Student Concentration Profile Manager 1024: Custom Review Video Generator 1025: Feedback Report Generator 1026: Instructor Voice Subtopic Analyzer 103: Feedback-Incorporated Design Engine 200: Training Data Management Server 201: Content Repository 202: User Profile / History DB 203: Evaluation / Test DB 204: Concentration Profile DB 205: Custom Review Content Repository 300: Client terminal 301: Custom UI Designed for Instructors 302: Instructor AI Feedback Report UI 303: UI providing customized review videos for students 304: Notification / Communication Module

Claims

Claim 1 In an AI-based online learning management system that measures and analyzes students' concentration levels in real time in an online learning environment to provide instructors with support for customized lecture design and feedback, the system comprises: a customized design support module that receives the instructor's lecture goals and plans as input and generates a customized design plan; and an AI evaluation feedback module that analyzes learning, evaluation, and attitude data based on AI to generate a feedback report. The system includes a feedback reflection design engine that collects and integrates analysis results and improvement suggestions from each component of the AI ​​evaluation feedback module to generate specific lecture design changes, and automatically transmits them to the custom design support module to reflect them in the next lecture design scenario; wherein the AI ​​evaluation feedback module comprises: an concentration analyzer that generates a concentration curve by time period by collecting facial expression and gaze data of students through a webcam and voice response data through a microphone; a statistical analyzer that calculates the average value, standard deviation, median, IQR, and histogram for the concentration curve; a student-specific concentration profile manager that manages concentration records and fluctuation patterns for each student; a custom review video generator that identifies sections of decreased concentration and automatically generates review videos for those sections; and a feedback report generator that generates a report along with visualized statistics from the analysis results of the concentration analyzer, statistical analyzer, student-specific concentration profile manager, and custom review video generator, and generates lecture improvement suggestions. An instructor voice subtopic analyzer that classifies instructor voice into subtopics of learning explanation, jokes, encouragement, questions, and feedback, and evaluates the influence of concentration by utterance type by matching with the aforementioned concentration curve;An AI-based online learning management system comprising: an instructor voice subtopic analyzer that transcribes the instructor's voice data and classifies it into at least one subtopic among explanation of learning content, jokes, encouragement, questions, and feedback through natural language processing, and quantifies the concentration improvement effect for each subtopic by matching each subtopic segment with the change in student concentration; and an AI evaluation feedback module that detects the occurrence of a concentration gap when the concentration distribution shows a bimodal shape or the standard deviation exceeds a reference value based on the analysis results of the statistical analyzer, and automatically proposes separate support measures for a group of students with reduced concentration. Claim 2 An AI-based online learning management system comprising, in claim 1, a learning data management server that stores lecture videos, materials, user information, concentration profiles, and customized review content. Claim 3 delete Claim 4 delete Claim 5 In claim 1, the concentration analyzer extracts changes in facial expressions, gaze direction, and head movements from a student's webcam video, and analyzes response frequency, voice tone, and engagement from microphone audio to calculate a concentration score in the range of 0 to 100 in real time, thereby creating an AI-based online learning management system. Claim 6 In claim 1, the customized review video generator is an AI-based online learning management system that identifies a time interval in an individual student's concentration curve that has decreased by 20% or more compared to a reference value, and automatically edits the lecture video of the corresponding interval to generate a personalized review video of 3 to 10 minutes in length. Claim 7 In claim 1, the above-mentioned student concentration profile manager is an AI-based online learning management system that accumulates and stores concentration records by class, unit, and topic, and generates individual learning characteristic profiles by analyzing concentration patterns by time of day, changes in concentration by day of the week, and the degree of concentration maintenance by lecture length. Claim 8 In claim 1, the feedback report generator is an AI-based online learning management system that automatically generates a comprehensive report including a concentration curve graph, statistical analysis results, concentration rankings by student, a concentration distribution histogram, an effect analysis by subtopic, and text for lecture improvement suggestions. Claim 9 delete Claim 10 In claim 1, the system is an AI-based online learning management system that sequentially performs the steps of inputting lecture objectives, requesting lecture design, providing lecture content, collecting concentration data, generating concentration curves by time period, statistical analysis, collecting instructor voice data and classifying subtopics, analyzing concentration and matching subtopics, updating concentration profiles for each student, generating AI evaluation feedback, generating customized review videos, generating feedback reports, designing with feedback reflected, proposing design changes, finalizing instructor designs, and distributing new lecture plans.

Images