An Artificial Intelligence-Based DISC Type Analysis and Customized Lecture Recommendation System

Abstract

The present invention relates to an artificial intelligence-based DISC type analysis and customized lecture recommendation system. Specifically, it is a system for analyzing a user's personality type through DISC testing, identifying areas requiring improvement using an artificial intelligence algorithm, and recommending lectures suitable for those areas. This invention aims to realize personalized learning by analyzing disk types and recommending the most suitable lectures to the user. By reflecting the user's personality type and learning style, this system enhances learning efficiency and helps the user improve in areas where necessary. In particular, the recommendation algorithm utilizing AI becomes more sophisticated over time, providing a learning environment that enables the user to grow continuously.

Description

Technology Field

[0001] The present invention relates to an artificial intelligence-based DISC type analysis and customized lecture recommendation system. Specifically, it is a system for analyzing a user's personality type through DISC testing, identifying areas requiring improvement using an artificial intelligence algorithm, and recommending lectures suitable for those areas. Background Technology

[0002] In modern society, the importance of self-development and education is growing day by day. However, providing personalized education tailored to each individual's inclinations and learning styles is not easy.

[0003] The DISC test can help identify a person's tendencies by analyzing their personality type. However, existing e-learning systems often fail to adequately reflect a user's DISC type and frequently provide standardized lectures.

[0004] Accordingly, the present invention proposes a system that utilizes artificial intelligence (AI) to analyze a user's disk type and recommends customized lectures tailored to it. By suggesting necessary content based on individual preferences, this system can enhance the efficiency and satisfaction of learning. Prior art literature

[0005] Korean Registered Patent Publication No. 10-2712039 Korean Registered Patent Publication No. 10-0972413 The problem to be solved

[0006] The objective of the present invention is to realize personalized learning by analyzing disk types based on artificial intelligence and recommending the most suitable lectures to the user. means of solving the problem

[0007] To solve the above-mentioned problem, the present invention includes a customized lecture recommendation server (100) for providing customized recommended lectures by receiving test performance information based on disk analysis test information from a tester terminal (200) and analyzing the provided test performance information based on artificial intelligence. The customized lecture recommendation server (100) includes a disk test database unit (110) for storing disk analysis test information including predetermined questions, a disk type analysis unit (120) for providing analysis result information including personality types analyzed through the provided test performance information received via the tester terminal (200) after performing disk analysis test information, an artificial intelligence lecture recommendation unit (130) for providing recommended lecture information by analyzing the analysis result information provided through the disk type analysis unit (120) through a customized recommendation AI model for recommending customized lectures, a user interface unit (140) for providing a learning dashboard for checking recommended lecture information and profiles, and recommended through the artificial intelligence lecture recommendation unit (130). An artificial intelligence-based disk type analysis and customized lecture recommendation system is provided, which includes a growth tracking and feedback unit (150) for tracking the examiner's learning progress when taking a lecture according to recommended lecture information and receiving feedback information about the lecture from the examiner terminal. Effects of the invention

[0008] This invention aims to realize personalized learning by analyzing disk types and recommending the most suitable lectures to the user. By reflecting the user's personality type and learning style, this system enhances learning efficiency and helps the user improve in areas where necessary. In particular, the recommendation algorithm utilizing AI becomes more sophisticated over time, providing a learning environment that enables the user to grow continuously. Brief explanation of the drawing

[0009] FIG. 1 is a diagram briefly illustrating the overall configuration of the present invention. FIG. 2 is a block diagram illustrating the configuration of the customized lecture recommendation server (100) of the present invention. FIG. 3 is an exemplary diagram of the overall configuration of the present invention. Specific details for implementing the invention

[0010] Terms and words used in this specification and claims should not be interpreted as being limited to their ordinary or dictionary meanings, but should be interpreted in a meaning and concept consistent with the technical spirit of the invention, based on the principle that the inventor can appropriately define the concept of the terms to best describe their invention.

[0011] Preferred embodiments of the present invention will be described in detail below with reference to the attached drawings.

[0013] The present invention relates to an artificial intelligence-based DISC type analysis and customized lecture recommendation system. Specifically, it is a system for analyzing a user's personality type through DISC testing, identifying areas requiring improvement using an artificial intelligence algorithm, and recommending lectures suitable for those areas.

[0015] Referring to FIG. 1, the present invention has a customized lecture recommendation server (100) and an inspector terminal (200) communicating via a network.

[0016] A network is a network that serves as a transmission path for web pages; it may be a closed network such as a LAN (Local Area Network) or WAN (Wide Area Network), but it is desirable for it to be an open network such as the Internet. The Internet refers to a global open computer network structure that provides the TCP / IP protocol and various services existing at its upper layers, namely HTTP (HyperText Transfer Protocol), Telnet, FTP (File Transfer Protocol), DNS (Domain Name System), SMTP (Simple Mail Transfer Protocol), SNMP (Simple Network Management Protocol), NFS (Network File Service), and NIS (Network Information Service).

[0017] The above-mentioned tester terminal can be implemented in various forms. For example, the terminal described in this specification may be a mobile terminal such as a smartphone, tablet PC, PDA (Personal Digital Assistant), PMP (Portable Multimedia Player), or MP3 player, as well as a fixed terminal such as a smart TV or desktop computer.

[0019] The customized lecture recommendation server (100) receives inspection performance information based on disk analysis inspection information from the inspector terminal (200), analyzes the provided inspection performance information based on artificial intelligence, and provides customized recommended lectures.

[0021] Referring to FIG. 2, the customized lecture recommendation server (100) is configured to include a disk inspection database unit (110), a disk type analysis unit (120), an artificial intelligence lecture recommendation unit (130), a user interface unit (140), and a growth tracking and feedback unit (150).

[0023] Each component is explained in detail as follows.

[0025] The disk test database section (110) stores disk analysis test information including predetermined items. The disk analysis test information is a personality type test for determining personality type and includes predetermined item information.

[0026] The above disk analysis test information can be classified into four types: dominant (D), sociable (I), stable (S), and cautious (C), and can provide result information for the test.

[0028] The disk type analysis unit (120) is intended to perform disk analysis test information, receive the provided test performance information through the tester terminal (200), and provide analysis result information including the personality type analyzed through the provided test performance information.

[0029] The above analysis results are stored in the user profile of the user interface unit (140) and are intended to provide areas requiring improvement by analyzing the test performance information based on artificial intelligence.

[0031] The artificial intelligence lecture recommendation unit (130) is intended to provide recommended lecture information by analyzing the analysis result information provided through the disk type analysis unit (120) using a customized recommendation AI model for recommending customized lectures.

[0033] The above artificial intelligence lecture recommendation unit (130) is composed of a lecture database unit (131), a customized recommendation AI algorithm provision unit (132), and a recommendation result update unit (133).

[0035] The lecture database section (131) stores multiple lecture information in a database. The lecture information can be stored by dividing the courses according to personality types, and for example, the lecture information included in the DS type as a result of disk inspection can be organizational communication conversation methods, customer service methods, conflict case analysis, etc.

[0037] The customized recommendation AI algorithm providing unit (132) is intended to generate a customized recommendation AI model for recommending customized lectures to the examiner by learning the analysis result information of multiple lecture information and disk analysis inspection information stored in the lecture database unit (131) and the inspection performance information of disk analysis inspection information using artificial intelligence.

[0039] Through the above-mentioned custom recommendation AI algorithm provider (132), a lecture suitable for the area that the user needs to improve can be recommended. For example, a lecture to improve communication skills can be recommended to a user of the cautious type (C).

[0041] The recommendation result update unit (133) is intended to transmit the examiner's feedback information and learning performance information regarding the recommended lecture information to the customized recommendation AI algorithm provision unit (132).

[0042] By transmitting learning performance information to the customized recommendation AI algorithm provision unit (132) through the above recommendation result update unit (133), the algorithm of the customized recommendation AI model can be continuously updated through artificial intelligence learning, thereby increasing the accuracy of lecture recommendations.

[0044] The artificial intelligence lecture recommendation unit (130) further includes a lecture content style analysis unit (134), and the lecture content style analysis unit (134) analyzes lecture video and audio data to quantify the delivery speed, visual change rate, logical density, etc. of the lecture, and calculates an indicator to match this with the user's DISC personality type.

[0045] The lecture content style analysis unit (134) extracts metadata from video files stored in the lecture database unit (131) and each process is performed through the following detailed analysis modules.

[0047] The speech rate and tone analysis module is designed to quantify the instructor's speaking style by separating the audio track from the lecture video and applying noise removal preprocessing.

[0048] The specific processing steps are as follows.

[0049] First, to calculate the Net Speech Rate (Net SPM), instead of simply dividing the total time by the number of syllables, Voice Activity Detection (VAD) technology is applied to identify silent periods where the instructor does not speak and exclude them from the total playback time.

[0050] The 'Net SPM' is derived by calculating the total number of syllables of text converted via STT (Speech-to-Text) relative to the pure speaking time. This serves as a key indicator for measuring the 'sense of speed' preferred by Type D (dominant) learners, and a higher SPM classifies a lecture as having a higher density of information delivery.

[0051] Voice Activity Detection (VAD) is a technology that distinguishes human voices from background noise in audio signals to determine which segments contain voice and which do not. This technology is utilized in various fields, such as speech recognition, video conferencing, and voice encoding, and improves efficiency by initiating processing only when voice input is present or by eliminating unnecessary noise segments.

[0052] Second, as a frequency variability analysis, the fundamental frequency (F0), which determines the pitch of the human voice, is extracted from the audio waveform on a frame-by-frame basis. The standard deviation of the extracted F0 values ​​is calculated; if the deviation exceeds a preset threshold, it is classified as a dynamic tone, and if it falls below, it is classified as a calm tone. This serves as an indicator that lectures with large frequency fluctuations and distinct intonation are suitable for stimulating the interest of Type I (Social) learners, while tones with small fluctuations and stability provide comfort to Type S (Stable) learners.

[0054] The visual stimulus density analysis module is designed to analyze, on a frame-by-frame basis, the components of a video that influence a learner's visual attention, in addition to auditory elements.

[0055] To this end, the video is sampled at a rate of 1 to 5 frames per second, and the color histogram difference values ​​between adjacent frames are calculated. Points where the difference value exceeds a threshold are recognized as scene transitions, and the number of scene transitions per minute is calculated. Through this, it is determined that videos with frequent screen transitions are suitable for Type I (Social) individuals who feel less boredom, whereas videos with slow transitions and long durations of a single frame are suitable for Type C (Cautious) individuals who require deep thinking.

[0056] In addition, the pixel area of ​​the text within the frame is calculated using an OCR (Optical Character Recognition) engine, while the areas of people, graphs, and photos are calculated simultaneously using object recognition algorithms. If the proportion of text relative to the total valid area is high, it is classified as 'Information-Conveying Type,' and if the proportion of images is high, it is classified as 'Emotion-Conveying Type.' This serves as the basis for assigning a higher weight to lectures with high text density for Type C learners who prioritize evidence and data.

[0058] The linguistic structure analysis module is designed to identify the logical and emotional tendencies of a lecture by analyzing the semantic characteristics of the vocabulary used by the instructor, and it analyzes the frequency ratios of logical conjunctions (therefore, because, consequently, etc.) and emotional adjectives (great, joyful, worried, etc.) in the lecture script.

[0060] Specifically, natural language processing (NLP)-based morphological analysis is performed on lecture scripts converted into STT to tag nouns, verbs, adjectives, and adverbs. Additionally, the frequency of occurrence of words matching the 'logical vocabulary dictionary' (e.g., therefore, because, that is, consequently, proportionally) and the 'emotional vocabulary dictionary' (e.g., amazing, regrettable, passionate, happy) pre-built in the database is counted. The ratio of logical conjunctions and emotional adjectives relative to the total number of sentences is calculated as a percentage. As such, lectures with a high ratio of logical vocabulary are primarily recommended to Type C and Type D individuals who pursue structural completeness, while lectures with a high ratio of emotional vocabulary are matched as suitable content for Type I and Type S individuals who value empathy and storytelling.

[0061] As such, lectures with a high proportion of logical vocabulary are primarily recommended to Types C and D who pursue structural completeness, while lectures with a high proportion of emotional vocabulary are matched as suitable content for Types I and S who value empathy and storytelling.

[0063] A lecture content style analysis unit (134) according to one embodiment of the present invention does not simply rely on keyword matching, but calculates a DISC suitability index by quantitatively calculating the distance between the physical characteristics of the video itself and the user's personality preference. This process is performed according to the following specific temporal sequence and judgment criteria.

[0065] The lecture content style analysis unit (134) defines the following three key factors for precise matching and uses them in calculations.

[0067] - Lecture style attribute value (F): A value normalized between 0 and 1 for features extracted from individual lecture videos.

[0068] The reason for setting the above lecture style attribute values ​​is that since the length of each video varies, they are converted into ratios rather than absolute values ​​to enable mutual comparison. For example, 'speech speed' is measured based on syllables per minute (SPM), with the top 10% of the lectures mapped to 1.0 and the bottom 10% to 0.0.

[0069] The above lecture style attribute values ​​are derived through a process of converting the absolute values ​​of individual lectures into relative position values ​​by comparing them with the distribution of the entire database.

[0070] First, measurable physical data is extracted from the lecture video to be analyzed. For example, for the 'speech rate' attribute, the number of syllables in the entire text is counted via Speech-to-Text (STT), and the 'syllables per minute (SPM)' is obtained as raw data by dividing the count by the pure speech time (minutes) excluding silent periods.

[0071] Next, the SPM distribution for over tens of thousands of total lecture contents stored in the system is loaded to determine where the SPM of the corresponding lecture is located within the entire population. At this time, to prevent distortion caused by outliers, a valid range is set by excluding the top 5% and bottom 5% of the total data.

[0072] Finally, the percentile of the SPM value of the corresponding lecture within the valid range is converted to a decimal point between 0.00 and 1.00.

[0073] For example, when the average SPM of all lectures is 300 and the lecture under analysis has 450 SPM, if this falls within the top 10%, the speed attribute value (F_speed) of that lecture is assigned as 0.90. If it is significantly slower than the average and falls within the bottom 20%, it is assigned as 0.20.

[0075] - User Preference Optimum Value (P): This is the ideal value based on the user's DISC type, indicating the level of comfort they feel regarding the corresponding course attributes.

[0076] The reason for setting the optimal value based on user preference is that even at the same speed, Type D (Dominant) prefers a fast speed (e.g., 0.8), while Type S (Stable) prefers a slow speed (e.g., 0.3). In other words, since the 'correct answer' differs by type, this is set as the reference point.

[0078] The above user preference optimal value (P) is precisely tuned for each individual through a two-stage hybrid method of 'initial profiling' and 'cumulative learning'.

[0079] First, when a user completes the DISC test after initial registration, an initial value is set by referring to a predefined 'standard preference table by type' based on personality psychology theories. For example, if the user is identified as 'Type D (Dominant)', the standard values ​​for Type D, [Speed: 0.8, Visual: 0.4, Logical: 0.6], are assigned as the default optimal value ($P$).

[0080] Next, the style attribute values ​​(F) of courses that the user actually completed (progress rate of 90% or more) or gave high ratings to are collected and weighted averaged. If a tendency is repeatedly observed where the user, despite being of type D, actually listens to courses that are slow (F=0.3) to the end, the system gradually lowers the initial value of 0.8 (e.g., 0.8->0.75->0.6) to update to a unique optimal value (P) based on the user's actual behavioral data.

[0082] - Sensitivity weight by type (w): The intensity of stress felt by the user when a specific attribute is inconsistent.

[0083] Type C (Prudent) is highly likely to give up learning if there is a lack of logical structure (mismatch in logical attributes), so the weight is set high (e.g., 1.5 times), and Type I (Social) is more sensitive to visual stimuli than logic, so the weight of the visual attribute is set high to enhance discrimination.

[0085] The above type-specific sensitivity weights (w) are not merely preferences, but rather quantify the threshold resistance to the question, "Will learning be abandoned if this attribute is not a good fit?" This is calculated through correlation coefficient analysis with the dropout rate.

[0086] To this end, first, if a specific DISC type of user group turns off a video or drops out during training, we backtrack to identify which attribute (F) of the video showed the largest difference (Gap) from the user's preferred optimal value (P). If a positive correlation is confirmed where the dropout rate increases sharply when the gap of a specific attribute is large, that attribute is classified as a learning determinant and its weight is set to 1.0 or higher (e.g., 1.5–2.0). On the other hand, if there is no significant change in the dropout rate even with a large gap, this is considered a factor that users do not care much about, so the weight is lowered to less than 1.0 (e.g., 0.5–0.8).

[0088] For example, data analysis results show that Type C learners can tolerate poor video quality (visual elements), but they immediately drop out if there is a severe logical leap (logical elements). Accordingly, the system automatically sets the visual attribute weight (w_visual) to a low 0.7 and the logical attribute weight (w_logic) to a high 1.8 in the profiles of Type C users and reflects this in the recommendation algorithm.

[0091] This system determines whether to make a final recommendation through the following sequential steps.

[0093] Step 1 (Attribute Extraction and Normalization)

[0094] The system performs image processing and speech recognition on newly registered lecture videos to extract data on speech speed, screen transition rate, and logical vocabulary usage frequency. The system generates a histogram by sorting the corresponding attribute values ​​(e.g., speech speed) of all lecture videos accumulated in the database in ascending order, calculates the percentile rank of the new video's attribute value within the entire population, and converts it into a lecture style attribute value by mapping it to a decimal value between 0.00 (lower 0%) and 1.00 (upper 100%).

[0096] Specifically, the system loads attribute data of all previously stored lecture content to update a normal distribution curve or frequency distribution table, and performs clipping to fix the ranges exceeding the top 5% and the ranges below the bottom 5% of the distribution to a maximum value (1.0) and a minimum value (0.0), respectively, to ensure data reliability and prevent data distortion caused by extreme outliers. Then, if new data is located within the valid range (bottom 5% to top 95%), the relative position ratio within that range is calculated and a value is assigned.

[0097] For example, when the average speech rate of the entire lecture is 300 syllables / minute and the maximum effective rate is 500 syllables / minute, if the input lecture is 400 syllables / minute, this corresponds to the 75th percentile point, which is the middle position within the effective interval, so an attribute value of 0.75 is assigned.

[0099] Step 2 (Calculation of individual deviations)

[0100] Load the DISC type information of the connected user, and calculate the difference (absolute value) between the user's 'optimal user preference value' and the corresponding lecture's 'lecture style attribute value' for each attribute.

[0102] Stage 3 (Weighting Application and Penalty Accumulation)

[0103] The difference value of each calculated attribute is multiplied by the user's 'type-specific sensitivity weight', squared, and summed. The reason for performing 'squaring' rather than simple difference is to allow for minor differences, but to drastically increase the penalty and exclude from recommendations if even one attribute deviates significantly from the threshold.

[0105] Step 4 (Inversion of Fit Index and Final Decision)

[0106] Since the summed total penalty value represents the 'disagreement', an operation to take the reciprocal is performed to convert it into 'fit'. However, if the reciprocal is taken simply, there is a problem where a division by zero error occurs when the total penalty value is '0' (all attributes match perfectly) and the denominator becomes zero. To prevent this and to ensure that the fitness index is normalized to the maximum value of '1' when the penalty is zero, a 'correction constant 1' is added to the summed total penalty value and then the reciprocal is calculated.

[0107] Accordingly, if the total penalty is 0, the index is 1 (maximum fit), and as the total penalty increases, the index follows a non-linear decreasing curve that converges to 0, thereby ensuring the discriminative power to sensitively lower the recommendation score when elements unpopular with the user are included. Only when the final calculated index is above a preset threshold (e.g., 0.75) is it classified and displayed as a 'Recommended Lecture' on the user dashboard.

[0109] According to the above calculation method, the penalty value in the denominator increases as the difference between the lecture style attribute value and the optimal user preference value increases (i.e., as the gap widens) and as the weight for that attribute increases. Consequently, this becomes a factor that lowers the 'DISC fit index.' In other words, if the lecture style does not match a specific element that the user considers important (an element with a high weight), the recommendation ranking is designed to drop sharply, even if other elements are excellent, thereby ensuring the reliability of personalized recommendations.

[0111] To aid in a concrete understanding, we assume a situation where a 'Cautious (Type C)' user receives a suitability assessment for 'Lecture A'. In this embodiment, for the sake of convenience of explanation, the core factors (F, P, w) are used to calculate only the two representative attributes: transmission speed and logical density.

[0113] [Input Data and Parameter Settings]

[0114] - User conditions (Type C profile)

[0115] User preference optimal value (P): Since the user prefers calm and accurate delivery, the speed optimal value (P_{peed}) is set to 0.4, and since the user wants logical completeness, the logic optimal value (P_logic) is set to 0.9.

[0116] Sensitivity weight (w): It tolerates speed mismatch to some extent (default weight w_speed = 1.0), but gives up learning if there is no logic, so a weight of 2 (w_logic = 2.0) is assigned to the logic attribute.

[0118] - Comparison Course A (Emotional and fast style)

[0119] Lecture style attribute value (F): Analysis results showed that the speech is very fast (top 10%, F_speed = 0.9) and the logical structure is weak because it is emotion-oriented (F_logic = 0.2).

[0121] [Step-by-step output process]

[0122] First, the difference between the lecture attribute value (F=0.9) and the user optimal value (P=0.4), which is 0.5, is squared (0.25). This is then multiplied by the speed weight (w=1.0) to calculate the speed penalty value as 0.25.

[0123] The difference between the lecture attribute value (F=0.2) and the user optimal value (P=0.9), which is 0.7, is squared (0.49). This is multiplied by the core weight of type C (w=2.0) to calculate a logical penalty value of 0.98.

[0124] The total penalty sum is as follows.

[0125] Total Penalty = Speed ​​Penalty (0.25) + Logical Penalty (0.98) + Other Correction Values ​​(0) = 1.23

[0127] Finally, the total penalty (1.23) is combined with a correction constant (1) to make 2.23, and the reciprocal value of 1 divided by 2.23 is determined as the final exponent (1 / 2.23 = approximately 0.44).

[0129] The calculated index of 0.44 falls short of the recommendation threshold (e.g., 0.75), so the course is excluded from the recommendation list.

[0131] As such, the present invention applies a quantitative computational model combining multiple attribute variables and user weights, rather than a simple conditional statement (IF-THEN), thereby having the effect of significantly reducing the dropout rate caused by the same keywords but mismatched styles. In addition, while the threshold value uses the statistical average value by personality type during the initial stages of system operation, it subsequently receives feedback on whether the user watched the lecture to the end (course completion rate) and automatically adjusts the weights and optimal preference values, thus having a virtuous cycle structure in which the accuracy of recommendations improves with continued use.

[0133] In addition, the updating of the user preference optimal value (P) and type-specific sensitivity weight (w) as described above is performed through the organic interaction between the growth tracking and feedback unit (150) and the artificial intelligence lecture recommendation unit (130) of the present invention.

[0135] Specifically, the growth tracking and feedback unit (150) collects learning behavior logs such as 'completion', 'dropping out', and 'repeated listening' that occur while the user is taking a lecture in real time and transmits them to the artificial intelligence lecture recommendation unit (130). Accordingly, the artificial intelligence lecture recommendation unit (130) periodically executes the following self-learning process based on the received data.

[0136] To dynamically correct the preferred optimal value (P), style attribute values ​​(F) of lectures for which the user has achieved a progress rate of 90% or more are extracted, and the user's preferred optimal value (P) is updated by applying a time-series weighted average method that assigns higher weights to lectures taken more recently. This allows for the rapid reflection of the user's changing learning preferences.

[0137] To fine-tune the sensitivity weight (w), if a pattern of repeated user drop-off at a specific point is detected, the attribute with the largest gap between the lecture attribute value (F) and the user preference value (P) at that point is identified as the 'drop-off trigger factor.' The system adjusts the sensitivity weight (w) for that attribute upward (e.g., +0.2), thereby ensuring that discrepancies in that attribute are filtered more strictly during future recommendations.

[0139] This feedback loop structure is not a fixed algorithm, but rather provides a technical effect where the accuracy of recommendations evolves and becomes optimized for the individual as the user utilizes the system.

[0141] The user interface section (140) is intended to provide a learning dashboard for checking recommended lecture information and profiles.

[0143] The above user interface section (140) is configured to include a personalized dashboard section (141) and a profile update section (142).

[0145] The personalized dashboard section (141) is intended to display the examiner's profile information and recommended lecture information.

[0147] The profile update unit (142) is intended to reflect the inspection execution information and analysis result information provided through the disk type analysis unit (120) into the profile information.

[0149] The above profile information may include the examiner's gender, age, analysis result information including personality types analyzed through test performance information, recommended lecture information, and enrolled lecture information.

[0151] In addition, when the analysis result information is updated through the profile update unit (142), the AI ​​lecture recommendation unit (130) can automatically analyze the analysis result information to provide new recommended lecture information in order to provide recommended lectures accordingly.

[0153] The system of the present invention further includes a dynamic UI / UX generation unit (143) that enhances the user interface unit (140) to not simply list recommendation results but variably reconfigures the layout, color theme, and data display method of the dashboard according to the user's DISC personality type.

[0154] Generally, the preferred forms of information differ by type, such as Type D (Dominant) prioritizing 'results and goals' and Type C (Prudent) prioritizing 'process and evidence'. Therefore, this component does not use a fixed template, but quantifies the utility value that each UI component (widget) provides to the user and renders the screen in real time.

[0156] To this end, the dynamic UI / UX generation unit (143) is composed of a style attribute mapping module, a UI component priority calculation module, and a variable layout rendering module, and the specific operating principles through each module are as follows.

[0158] The Style Attribute Mapping Module receives the user's DISC personality type information, determines visual design attributes optimized for the cognitive preferences of that type, and transmits them to the system's front-end rendering engine. To this end, the module references a pre-established type-specific design system database (DB) and dynamically maps four core UI attribute values—color, typography, information density, and interaction—based on the user's primary and secondary type scores. In other words, it maps design system attributes, including pre-configured color palettes, typography, information density, and interaction response speed, according to the user's DISC type information. The specific processing logic is as follows.

[0160] First, the system maps color codes capable of maximizing psychological response for each DISC type into the form of CSS variables.

[0161] - Type D (Dominance): Considering the goal-oriented and challenging nature, high-saturation red tones that induce arousal and action are designated as the primary color, and high-contrast black is mapped as the secondary color to enhance intuition.

[0162] - Type I (Social): Considering the optimistic and active nature, bright yellow or orange tones that provide joy and vitality are designated as the main colors, and pastel-toned secondary colors are used to create a cheerful atmosphere.

[0163] - Type S (Stable): Considering the tendency to seek peace and stability, low-saturation colors in green or beige tones are mapped to minimize eye strain and provide a sense of psychological stability.

[0164] - Type C (Cautious): Considering the logical and analytical nature, data readability is prioritized by mapping dark blue or neutral colors that symbolize trust and composure.

[0166] Second, the layout density of on-screen content is adjusted according to the user's preferred information processing method.

[0167] - Type D and Type I (Intuitive Processing Preference): Apply a low-density layout that reduces the amount of text and allocates ample whitespace. Place key figures or conclusions at the top using large fonts, and hide detailed explanations behind a 'See More' button to facilitate quick decision-making.

[0168] - Type C and Type S (Preference for analytical / procedural processing): Apply a high-density layout that can fit as much information as possible on a single screen. Divide the grid system into fine sections to arrange detailed data tables, references, graphs, etc., so that they can be grasped at a glance without scrolling.

[0170] Third, differentiate the fonts and graphic elements that determine the tone and manner of information delivery.

[0171] - Headline Emphasis Type (Type D Target): The font size of the title is increased by 1.5 times compared to the standard and made bold to clearly establish hierarchy.

[0172] -Graphic-centric type (Type I target): Instead of reducing text descriptions, actively incorporate non-verbal graphic elements such as emojis, badges, and character avatars that express emotions into buttons and menus.

[0173] - Readability-Oriented Type (Type C / Type S Target): Applies serif fonts suitable for reading long texts or highly readable sans-serif fonts, and provides a comfortable reading environment by setting generous line spacing.

[0175] Fourth, it controls the speed and method of the system's response to user operations.

[0176] - Immediate Feedback (D / I Type): Set the screen transition speed to fast (within 0.2 seconds) upon button click, and apply flashy animation effects to provide fun and a sense of speed in operation.

[0177] - Stable Feedback (S / C Type): Applies smooth fade-in / fade-out effects during screen transitions (over 0.5 seconds) and displays detailed tooltip descriptions when hovering the mouse over buttons, reducing the possibility of operation errors and providing a sense of stability.

[0179] As such, unlike conventional methods that provide a single design template, this module maximizes the convenience of using the system that users unconsciously feel by replacing the user's personality traits with four dimensions of design variables (color, density, shape, and response) and rendering them in real time.

[0181] The UI component priority calculation module determines how important each functional unit (hereinafter 'UI component') to be placed on the screen is to the currently connected user, calculates a placement weight, and derives a UI visibility score (S) based on this. Rather than providing the same screen to all users, the present invention determines the optimal layout order for each individual by combining the DISC propensity score and the personality affinity of each component using a vector operation method. The definition of specific calculation factors and the calculation procedure are as follows.

[0183] This module defines the following four key factors for precise priority calculation and utilizes them in the operation.

[0184] 1) UI Displayability Score

[0185] The UI visibility score is a final priority value indicating how conspicuously a specific UI component should be placed on the user's screen. The higher this score, the larger the component is placed in the screen's 'prime zone (top or center); conversely, the lower the score, the more it is pushed to the bottom or hidden within a menu. The above UI visibility score is calculated through the following computational process. The system obtains four median values ​​by multiplying the 'user personality vector' and 'component attribute weight' (described later) by each tendency (D, I, S, C), and sums them all to derive the 'propensity fit raw score.' Subsequently, the UI visibility score is finalized through a final correction step in which this raw score is multiplied by a 'contextuality coefficient' determined based on the user's device environment. In other words, this value is the final output that reflects environmental variables in the result of the vector dot product of user tendencies and component characteristics.

[0187] 2) User Personality Vector

[0188] The user personality vector is a value normalized between 0 and 1 by adjusting the score ratio for each attribute (D, I, S, C) extracted from the user's DISC test results. To comprehensively consider not only the user's primary tendency but also their inherent secondary tendency, a 4-dimensional vector value is used instead of a single type code. The above user personality vector is calculated through the following normalization process. First, the raw score of the user's DISC test results is received from the DISC type analysis unit (120). The system calculates the 'total score' by adding all the raw scores for each item of D, I, S, and C. Then, the raw score of each item is divided by this total score to convert it into a ratio in decimal units. For example, if the raw score is [D=10, I=5, S=5, C=20], the total score becomes 40 points, and the normalized user personality vector is calculated as [0.25, 0.125, 0.125, 0.5].

[0190] 3) Component property weights

[0191] The component attribute weight is a pre-set constant used to indicate how a specific UI component is affinity with each attribute of DISC. It has a value between -1.0 and 1.0. For example, a 'Ranking Board' is attractive (positive) to Type D users who enjoy competition, but it can be a stress factor (negative) to Type S users who seek stability. This value is set to automatically place stress-inducing components in a lower priority, reflecting the correlation between 'preference' and 'avoidance'. The above component attribute weight is calculated (loaded) through the following database mapping process. The system references a pre-established UI attribute database for each UI component. This database stores affinity values ​​for each DISC attribute assigned through psychological analysis during the system design phase. Specifically, the system queries the database using the ID of the target component to be rendered (e.g., Ranking_Widget_01) as the key value and loads the 4-dimensional weight vector assigned to that component (e.g., D=0.9, I=0.7, S=-0.5, C=0.0).

[0193] 4) Situational Context Coefficient

[0194] The contextual coefficient is a correction value used to adjust UI complexity based on the user's access environment (mobile, tablet, PC). While complex charts are useful on PC screens, they can impair readability in mobile environments with narrow screens. Therefore, this value is set to optimize UX by forcibly lowering the score of complex components in environments with narrow resolution widths (e.g., less than 768px). The above contextual coefficient is calculated through the following real-time environment detection logic. When a user accesses the system, the dynamic UI / UX generation unit (143) receives display resolution width information from the user's browser or application. The system compares the received width value with a pre-set threshold resolution (e.g., 768px, 1024px). If the width is greater than or equal to the threshold (PC environment), the coefficient is set to the reference value of 1.0. On the other hand, if the width is less than the threshold (mobile environment) and a high-density tag is attached to the metadata of the component (e.g., complex chart), the coefficient is calculated by lowering it to a value less than 1.0 (e.g., 0.8) to prevent a decrease in readability.

[0196] This system utilizes the factors defined above to derive a final score through the following time-series three-step process.

[0198] Step 1) The system loads the DISC test result data of the connected user and generates a user personality vector by converting the scores of each item (D, I, S, C) into ratios relative to the sum score. At the same time, for candidate UI components to be displayed on the screen (e.g., ranking, chat, graph, to-do list, etc.), component attribute weights stored in the database are loaded.

[0200] Step 2) For each UI component, the user's D, I, S, and C propensity scores are individually multiplied by the corresponding component attribute weights. That is, multiplication is performed on the four attributes, such as 'User's D score × Component's D weight' and 'User's I score × Component's I weight'. Subsequently, all four calculated multiplication results are summed to derive the propensity fit raw score. According to this process, for a propensity in which the user has a high score (e.g., high D), if the component has a high positive weight (e.g., D-friendly), the result increases significantly, raising its priority. Conversely, if the component has a negative weight (e.g., D-avoidant), the result decreases, suppressing its placement.

[0202] Step 3) Analyze the user's connected device information to determine the contextual coefficient. In PC environments where the screen width is greater than a threshold (e.g., 1024px), the coefficient is set to 1.0, and in mobile environments, the coefficient is lowered to 0.8 for components with a large amount of information (charts, etc.). Finally, the UI presentation score is determined by multiplying the raw propensity fit score calculated in Step 2 by the contextual coefficient, and the sorting order of all components is determined in descending order of this score.

[0204] We will explain, through specific numerical calculations, how the dashboard configuration changes depending on the user type while holding the same data.

[0206] [Input Data Settings]

[0207] - Target component

[0208] (A) Competitive Ranking Widget: Favorable to Type D and disadvantageous to Type S (Weights: D=0.9, I=0.7, S=-0.2, C=0.1)

[0209] (B) Detailed Learning History Table: Favorable to Type C and disadvantageous to Type I (Weights: D=0.1, I=-0.1, S=0.4, C=0.9)

[0210] - Context: PC environment (Context coefficient = 1.0)

[0212] [CASE 1: Type D User (Personality Vector: D=0.8, I=0.2, S=0.1, C=0.1)]

[0213] - Ranking Widget (A) Calculation: (0.8×0.9) + (0.2×0.7) + (0.1×-0.2) + (0.1×0.1) = 0.72 + 0.14 - 0.02 + 0.01 = 0.85 (Final Score)

[0214] - History Table (B) Calculation: (0.8×0.1) + (0.2×-0.1) + (0.1×0.4) + (0.1×0.9) = 0.08 - 0.02 + 0.04 + 0.09 = 0.19 (Final Score)

[0215] -Result: The ranking widget with a significantly higher score (0.85) is placed in the prime zone at the top of the dashboard, and the history table is placed in a small size at the bottom.

[0217] [CASE 2: Type C User (Personality Vector: D=0.1, I=0.1, S=0.3, C=0.8)]

[0218] - Ranking Widget (A) Calculation: (0.1×0.9) + (0.1×0.7) + (0.3×-0.2) + (0.8×0.1) = 0.09 + 0.07 - 0.06 + 0.08 = 0.18 (Final Score)

[0219] - History Table (B) Calculation: (0.1×0.1) + (0.1×-0.1) + (0.3×0.4) + (0.8×0.9) = 0.01 - 0.01 + 0.12 + 0.72 = 0.84 (Final Score)

[0220] -Result: The detailed learning history table with a high score (0.84) is placed as the main center of the dashboard, and the ranking widget with a low score (0.18) is moved out of view.

[0222] As such, unlike conventional methods that use fixed templates, this component predicts and visually highlights the information that the user considers most important through the multiplication of personality vectors and weights. In particular, by introducing negative (-) weights, it automatically filters out or excludes information that may cause stress to the user (e.g., information that is excessively competitive for Type S types) during the screen composition stage, thereby providing a technical effect that minimizes the user's cognitive load and maximizes the efficiency of information acquisition.

[0224] The variable layout rendering module receives the UI visibility score (S) of each component and the user personality vector calculated by the UI component priority calculation module, dynamically allocates the screen grid system based on this, and performs polymorphic rendering that varies the visualization method according to tendencies even for the same data. The specific operation process is as follows.

[0225] This module divides the entire screen into N×M virtual grids and executes dynamic grid mapping logic that determines the position and size of each component based on the descending sort result of the UI visibility score (S).

[0227] Highest priority components with a UI visibility score in the top 20% are placed in the 'Prime Zone' (top-left or center on PC), where the user's gaze rests first. At this time, to maximize the visibility of the component, the grid occupancy area is allocated by expanding it to more than twice the standard.

[0228] Components with mid-range scores are arranged sequentially in standard sizes, and components with scores below the bottom 40% are placed in a 'sub-zone' near the bottom of the screen or moved into the 'More' menu to reduce the complexity of the initial screen.

[0230] This module transforms not only the position of the component but also the presentation method of the information contained within it in real time according to the user's personality vector. The system holds multiple rendering templates (Text type, Chart type, Graphic type) for a single data source (e.g., learning progress rate) and selects the most suitable template based on the user's tendency to render.

[0231] If the user has a high D (Dominant) or I (Influential) score, a template is applied that boldly omits complex numerical data and emphasizes only the key conclusion. For example, data such as "Progress Rate 75%" is rendered in the form of a circular gauge along with the phrase "25% remaining until goal achievement!", allowing the user to intuitively recognize the goal.

[0232] If the user's C (Cautious) or S (Stable) propensity score is high, a detailed template is applied to grasp the context and trends of the data. For example, even with the same "75% progress rate" data, it is rendered in the form of a line chart or a detailed table that includes "a 5% increase from the previous day and a graph showing the trend of learning volume over the past week," enabling the user to precisely analyze the learning status.

[0234] The above-determined layout coordinates and visualization template information are finally passed to a frontend rendering engine (e.g., Virtual DOM such as React, Vue.js, etc.). The rendering engine detects only the changed parts and updates the screen, thereby minimizing system resource consumption and screen flickering that may occur during the process of real-time layout changes.

[0235] In this way, this module breaks away from a uniform screen configuration and maximizes UX satisfaction by providing the information that users unconsciously consider most important (high visibility score) in the most prominent location and form (prime zone & customized visualization).

[0237] The growth tracking and feedback unit (150) is intended to track the examiner's learning progress when taking a lecture based on the recommended lecture information recommended through the artificial intelligence lecture recommendation unit (130) and to receive feedback information about the lecture from the examiner terminal.

[0239] The system according to the present invention can construct a neural network that extracts contextual information for input data by training training data with the Word2Vec algorithm to understand or estimate the meaning of the data.

[0240] The Word2Vec algorithm may include a Neural Network Language Model (NNLM). A Neural Network Language Model is fundamentally a neural network composed of an Input Layer, a Projection Layer, a Hidden Layer, and an Output Layer. The Neural Network Language Model is used as a method for vectorizing words. Since the Neural Network Language Model is a well-known technology, a more detailed explanation will be omitted.

[0241] The Word2vec algorithm is designed for text mining and determines proximity based on the preceding and succeeding relationships between words. It is an unsupervised learning algorithm. As its name suggests, Word2vec is a quantitative technique that represents the meaning of words in vector form. The Word2vec algorithm can represent each word as a vector in a space of approximately 200 dimensions. By utilizing the Word2vec algorithm, a vector corresponding to each word can be obtained.

[0242] The Word2vec algorithm can enable a dramatic improvement in precision in the field of natural language processing compared to other conventional algorithms. Word2vec learns the meaning of words by utilizing the relationships between words and adjacent words within sentences of an input corpus. Based on artificial neural networks, the Word2vec algorithm starts from the premise that words with the same context carry similar meanings. The algorithm learns through text documents, training the neural network to identify related words by identifying other words that appear nearby (approximately 5 to 10 words before and after) a given word. Since words with related meanings are highly likely to appear close together within a document, the two words can gradually acquire closer vectors as the learning process is repeated.

[0243] There are two training methods for the Word2vec algorithm: CBOW (Continuous Bag Of Words) and skip-gram. The CBOW method predicts a target word by utilizing the context created by surrounding words. The skip-gram method predicts words that may follow a single word. The skip-gram method is known to be more accurate in large-scale datasets.

[0244] Accordingly, in the embodiments of the present invention, a Word2vec algorithm using the skip-gram method is used. For example, if training is successfully completed through the Word2vec algorithm, similar words can be located nearby in a high-dimensional space. According to the Word2vec algorithm described above, the closer the distribution of surrounding words within a training document, the more similar the calculated vector values ​​can be, and words with similar calculated vector values ​​can be considered similar. Since the Word2vec algorithm is a known technology, a more detailed explanation regarding the calculation of vector values ​​will be omitted.

[0245] The server can input collected data into a neural network to extract an evaluation result vector value representing contextual information.

[0246] The server calculates the similarity between the evaluation result vector value and each of the multiple reference vector values, and can extract the reference vector value among the multiple reference vector values ​​that has the highest similarity to the evaluation result vector value. In this case, Euclidean distance, cosine similarity, Tanimoto coefficient, etc., may be adopted as the similarity calculation method.

[0247] The management server can extract the word corresponding to the reference vector value with the highest similarity to the evaluation result vector value as the word corresponding to the recognized text.

[0248] Furthermore, the management server can train artificial neural networks and utilize completed artificial neural networks. The processor can train or execute artificial neural networks stored in memory, and memory can store completed artificial neural networks. The electronic device that trains the artificial neural network and the electronic device that utilizes it may be identical or separate. Artificial intelligence is a computer system that partially implements the functions of the human brain, capable of learning, speculating, and making judgments on its own. As learning progresses, the probability of extracting the correct answer may increase. Artificial intelligence can be composed of learning and elemental technologies that utilize it. The learning of artificial intelligence is an algorithmic technology that classifies and learns features based on input data, and the elemental technologies may be technologies that partially implement the functions of the human brain by utilizing learning algorithms.

[0249] Artificial intelligence is a technology that facilitates the approach to problems where multiple probabilistic answers are possible, enabling it to logically and probabilistically infer optimal cycles, methods, and plans based on input data. AI inference techniques can include evaluating input data, optimization prediction, knowledge and probability-based reasoning, and preference-based planning.

[0250] Artificial neural networks are learning algorithms in the field of machine learning that programmatically implement the connections between neurons and synapses in the brain. By creating a neural network structure through programming and then training it, artificial neural networks can acquire desired functions. Although errors may exist, they can learn from massive datasets to produce appropriate output data from input data. They have the advantage of being able to obtain output data that has yielded statistically good results and are similar to human reasoning.

[0251] The server can construct query / metric datasets required for learning using artificial intelligence algorithms built on big data, and to this end, it may include multiple pre-trained artificial neural networks.

[0252] The system according to the present invention may include a plurality of pre-trained artificial neural networks for performing machine learning algorithms. Through machine learning, it can output output data based on input data and learn autonomously using the results, thereby improving its data processing capabilities. The artificial neural network can extract features and predict regularities based on input data to output result data, and as this process accumulates, the reliability of the result data increases.

[0253] In this embodiment, the artificial neural network may be an algorithm that outputs text data from at least one feature data among the shape, length, number, and height difference of an object recognized as text. The artificial neural network can infer the best output data by using big data as input data as is, or by using it as input data after undergoing a processing step to clean up unnecessary data.

[0254] Artificial intelligence machine learning models are classified into Supervisory Learning, Unsupervisory Learning, Semi-supervised Learning, and Reinforcement Learning, depending on the type of learning. Additionally, machine learning algorithms that can be used include Decision Trees, K-Nearest Neighborhoods, Artificial Neural Networks, Support Vector Machines, Ensemble Learning, Gradient Descent, Hidden Markov Models, and K-Means Clustering.

[0255] Artificial neural networks can be pre-trained on various input values ​​that may be included in the input data. An artificial neural network can be trained using reinforcement learning, a learning method. Reinforcement learning is a method that gradually increases the probability of obtaining the correct result by setting rewards and constraints. Artificial neural networks can also be modeled based on Convectional Neural Networks (CNNs) or Recurrent Neural Networks (RNNs).

[0256] In this way, the system according to the present invention can estimate the meaning of text data using big data and artificial neural networks.

[0258] The customized lecture recommendation server (100) may further include a real-time concentration-based adaptive control unit (160) to maximize the learning efficiency of the learner. This configuration goes beyond simply playing lectures to analyze biosignals and behavioral patterns collected through the camera and input device of the examiner terminal (200), detects different concentration decline patterns for each DISC type, and performs customized intervention.

[0260] The above real-time concentration-based adaptive control unit (160) is largely composed of a multi-sensing module, a type-specific concentration calculation module, and an adaptive stimulus generation module.

[0262] The multi-sensing module operates by being divided into a video analysis unit that receives a video stream from a webcam or front camera equipped in the examiner terminal (200), and an input event analysis unit that detects mouse and keyboard input signals. Each unit quantifies the learner's biosignals and behavioral data in real time through a non-invasive method.

[0264] The video analysis unit calculates the accumulated time of gaze deviation (D_gaze) by performing the following stepwise processing on video frames input from the camera at a rate of 15 to 30 frames per second. First, a lightweight object detection algorithm (e.g., Haar Cascade or HOG+SVM) is applied to primarily identify the learner's face region (ROI: Region of Interest) within the video, and secondarily determine the coordinates of the left and right eyes within the face. Next, binarization and morphological operations are performed on the detected eye region to extract the 'pupil center coordinates,' which are the center points of the darkest pixel clusters that contrast with the surrounding skin color, in real time.

[0265] A 3D gaze vector is calculated by combining the learner's face angle and the relative position of the pupil center point, and this is projected onto a 2D monitor screen coordinate system (x, y). Finally, the monitor screen area is defined as the 'valid learning area,' and if the calculated gaze coordinates remain outside that area for a preset duration (e.g., 2 seconds) or longer, this is defined as a 'gaze loss event' and counted to calculate the accumulated gaze loss time (D_gaze).

[0267] In addition, the video analysis unit extracts 68 facial landmarks to derive the learner's physiological fatigue index (F_fatigue) and numerically calculates geometric changes in the eyelids and mouth.

[0268] Specifically, for EAR (Eye Aspect Ratio)-based drowsiness detection, the EAR value—the ratio of the vertical length to the horizontal length of the eye—is calculated frame by frame. If the state where the EAR value falls below a threshold (e.g., 0.2) (eyes closed) persists for a certain period, it is determined to be 'drowsiness' or 'microsleep'. In particular, the PERCLOS (Percentage of Eyelid Closure) metric, representing the proportion of time the eyes are closed by 80% or more, is calculated and utilized as a key measure of fatigue. Additionally, for MAR (Mouth Aspect Ratio)-based yawn detection, the MAR value is calculated by measuring the distance between the upper and lower lip landmarks. If a large MAR value exceeding the level of normal conversation or speech is detected, and this state persists for several seconds accompanied by a landmark change where the eyes are squinted, it is recognized as 'yawning', and the fatigue weight is increased. Finally, the above-calculated PERCLOS index (eye fatigue) and the number of yawn detections (mouth fatigue) are summed and normalized using pre-set weights to derive a physiological fatigue index (F_fatigue) with a value between 0 and 1.

[0270] The input event analysis unit serves as an auxiliary means for video analysis, analyzing the learner's mouse and keyboard usage patterns (A_input) to distinguish between 'active learning' and 'mere idleness'.

[0271] Specifically, to detect, mouse movement events (WM_MOUSEMOVE) and keyboard input events (WM_KEYDOWN) are monitored through system hooking at the operating system (OS) level. If no input signal is detected for a preset threshold time (e.g., 5 minutes), the concentration score is deducted because it is highly likely that the user is in a 'vacant seat' state with low concentration, regardless of the video analysis results.

[0272] In addition, by analyzing the movement path of the mouse cursor, if a 'jitter' pattern of meaninglessly repeating back and forth within a narrow radius in a short period of time, or a distracting movement of rapidly scanning the entire screen is detected rather than a click (button, scroll) for learning purposes, it is interpreted as a signal of boredom, especially for Type I (social) learners, and a weight is assigned.

[0274] The collected video frame data and event time-series data are synchronized by assigning timestamps based on the system's internal clock. Subsequently, they undergo noise filtering (e.g., Kalman filter) and are converted into input variables (D_gaze, F_fatigue, A_input) for calculating the real-time concentration index, which are then passed to the Real-time Concentration Index (RCI) calculation module.

[0277] The type-specific concentration calculation module integrates heterogeneous raw data collected from multiple sensing modules to calculate the Real-time Concentration Index (RCI), a quantitative indicator representing the learner's current state of immersion.

[0278] In this context, the core technical feature of the present invention is that, rather than applying the same formula to all learners, it takes into account that behavioral patterns exhibited when concentration decreases differ depending on the DISC personality type, and therefore dynamically varies the weights applied to each variable for summing. The specific calculation logic and procedure are as follows.

[0280] To calculate concentration precisely, this module defines the following three key factors and utilizes them in the calculation.

[0281] - Attention retention rate

[0282] The gaze retention rate is the ratio of the time the gaze remained within the monitor screen (valid area) during the most recent measurement interval (e.g., 10 seconds). This is derived by calculating the ratio of the 'cumulative time of gaze departure' to the total measurement interval time and subtracting this from 1. In other words, the longer the time of departure, the closer it approaches 0, and the more the gaze remains fixed, the closer it approaches 1. The reason for establishing this factor is that visual attention is the most fundamental stage of learning, so the act of the gaze leaving the screen is intended to serve as the most direct measure of loss of concentration.

[0283] - Biological Awakening

[0284] Biological alertness indicates the degree to which a learner remains awake without drowsiness or fatigue. This factor is derived by subtracting the 'fatigue index (0–1)' measured through facial analysis from 1. The fatigue index is a value normalized by the ratio of time the eyes are closed and the frequency of yawning, so the more fatigued a learner is, the lower the alertness value becomes. Additionally, the reason for setting this factor is to prevent mistaking a state of vacant staring or dozing off as being in the middle of learning, even if the gaze is directed toward the screen.

[0285] - Input responsiveness index

[0286] The input responsiveness index indicates whether mouse or keyboard operations are being performed in accordance with the learning context. This factor is normalized to a value between 0 and 1 by assigning bonus points for normal learning activities (writing, pausing, navigating sections) and deducting points when a 'idle state' with no input or a 'jitter state' involving meaningless mouse shaking is detected. The reason for setting this factor is to determine active participation in learning and, in particular, to filter out unconscious repetitive behaviors that occur when boredom is felt.

[0288] This system does not simply average the three factors defined above, but rather applies weights reflecting the behavioral characteristics of each user's DISC type to calculate the final index in the following sequential steps.

[0290] Step 1) The system checks the DISC type information of the logged-in learner and loads the type-specific behavior weight table stored in the database.

[0291] For Type I (Social) and Type D (Dominant), since there is a strong tendency to look away or move distractedly when bored, the gaze weight and input weight are set high (e.g., 0.4 and 0.4 respectively), and the 'arousal weight' is set relatively low (e.g., 0.2).

[0293] For S (Stable) and C (Conscientious) types, since they have a strong tendency to zone out or doze off quietly without significant movement even when concentration is broken, the arousal weight reflecting micro-expressions and eye blinking is set very high (e.g., 0.6), and the remaining weights are set low. (In this case, the three weights are normalized so that their sum is always 1.)

[0295] Step 2) For each of the previously calculated 'gaze retention rate', 'biological arousal level', and 'input responsiveness index', calculate the 'item contribution score' by multiplying by the weight of the corresponding loaded type.

[0296] This process is a step that adjusts the influence on the final result as each factor increases or decreases. For example, if the value of a high-weighted item (e.g., S-type arousal) decreases, the final result drops sharply, regardless of how high the scores of other items (gaze, input) are. This implies that the system sensitively detects signs of decreased concentration typically exhibited by learners of that specific type.

[0298] Step 3) The Real-time Attention Index (RCI) is derived by summing all three calculated contribution scores for each item. Finally, if the condition in which this RCI value falls below a preset 'intervention threshold (e.g., 0.6)' persists for a certain period of time (e.g., 5 seconds), the system confirms this as a 'decreased attention state' and triggers an adaptive stimulus generation module.

[0300] We will explain how the system's judgment differs depending on the DISC type even in the same behavioral situation by applying specific numerical values.

[0302] For example, the time the learner looked at the monitor for one minute was 54 seconds (gaze retention rate 0.9), which is good, but the eye closed frequently due to fatigue, indicating a high level of fatigue (biological arousal 0.3), and the mouse was not moving (input responsiveness 0.5).

[0304] [For Type I (Social) Learners]

[0305] - Weight setting: Since distraction detection is important, it is set to [Gaze 0.5 / Arousal 0.2 / Input 0.3].

[0306] -Calculation: (Gaze 0.9 × 0.5) + (Awakening 0.3 × 0.2) + (Input 0.5 × 0.3) = 0.45 + 0.06 + 0.15 = 0.66

[0307] - Result: Since it is above the threshold (0.6), it is determined that Type I learners are still focused, so no intervention is made. (It is judged that Type I learners are highly likely to be listening to the content if they are looking at the screen even when tired.)

[0309] [For S-type (Stable) learners]

[0310] - Weight setting: Since silent departure detection is important, it is set to [Gaze 0.2 / Arousal 0.6 / Input 0.2].

[0311] -Calculation: (Gaze 0.9 × 0.2) + (Awakening 0.3 × 0.6) + (Input 0.5 × 0.2) = 0.18 + 0.18 + 0.10 = 0.46

[0312] - Result: Since it is below the threshold (0.6), it is determined that the S-type learner has low concentration, so an immediate evocative stimulus (e.g., a stretching reminder) is provided. (This is because S-types are highly likely to be in a blank state when their arousal level drops, even while looking at the screen.)

[0314] Unlike conventional methods that judge concentration using only a single condition, the present invention adopts a multi-condition weighted summing method applying DISC type-specific weights, thereby enabling the accurate capture of 'signals of loss of concentration' that differ for each learner. As confirmed in the above embodiments, since the judgment results vary depending on personality type even for the same biosignal data, unnecessary interference (false positives) is reduced, and appropriate intervention is performed only at the necessary moments, resulting in improved overall completion rates and learning efficiency.

[0316] The adaptive stimulus generation module uses the real-time concentration index (RCI) and duration of attention drop received from the type-specific concentration calculation module as trigger signals to generate and output audiovisual intervention content optimized for the learner's current state and personality type.

[0317] The specific operation process of this module follows the sequential technical procedures of database construction, content matching and selection, and dynamic rendering and control.

[0319] 1) Structuring of the stimulus content database (DB) by type

[0320] Prior to generating stimuli, the system categorizes various intervention elements that induce learning motivation by DISC type and stores and manages them in advance in a database.

[0322] - Challenge-type DB [Type D Target]: This is content that stimulates a sense of achievement. It includes competitive and goal-oriented UI templates such as 'Time Attack Quiz,' 'Current Chapter Key Goal Achievement Rate,' and 'Short-term Mission (Summarize within 3 minutes).'

[0323] - Relational DB [Type I Target]: This is content that stimulates social interaction. It includes visual elements that induce interest and a sense of bonding, such as 'real-time cheering emoticons from fellow learners,' 'peer group progress rate rankings,' and 'video messages of encouragement from instructors.'

[0324] - Ventilation-type DB [S-type Target]: This is content designed to induce psychological stability. It includes features that relieve fatigue and provide comfort, such as an 'eye rest guide,' 'simple stretching timer,' 'background music (BGM) switching,' and 'rest notifications with warm colors.'

[0325] - Informational DB [Type C Target]: This content alleviates anxiety regarding learning gaps. It includes widgets that provide logical and analytical data, such as 'Review Missed Scripts,' 'Key Data Chart Popups,' and 'Shortcut to Incorrect Answer Notes.'

[0327] 2) Stimulating Content Matching and Selection Logic

[0328] When a signal of decreased concentration (RCI < threshold) is received, the system queries the database for the optimal stimulus using the learner's DISC type code as the key value. At this time, instead of a simple random call, the following situation-based selection algorithm is executed.

[0329] To this end, first, if the learner is Type D (Dominant), a challenge-type DB is loaded as the primary candidate, and if the learner is Type S (Stable), a revitalizing-type DB is loaded. If the primary cause of decreased concentration in the preceding computation module is identified as fatigue (decreased arousal), the priority of revitalizing content (e.g., stretching) is increased regardless of type. On the other hand, if the primary cause is identified as distraction (excessive input responsiveness), 'challenge-type (quiz)' or 'informative-type (data verification)' content that can fix attention on the screen is finally selected.

[0330] Since tolerance may develop if the same stimulus is repeated, a weak stimulus in the form of a small 'toast message' at the bottom of the screen is selected upon initial detection, and if concentration is not restored thereafter, a strong stimulus in the form of a modal pop-up that covers the center of the screen is selected.

[0332] The selected content is rendered on the top layer of the screen of the examiner terminal (200), and the playback environment is controlled as follows to induce an immediate response from the learner.

[0333] As soon as a strong stimulus (e.g., quiz, stretching) is displayed, the system forcibly pauses the lecture video currently playing. This is a technical device designed to prevent the learner from missing content while simultaneously inducing forced refocus by ensuring that progress cannot be made without responding to the stimulus. After the stimulus is displayed, the stimulus overlay is removed and the video resumes only when a valid interaction is detected, such as the user clicking the 'Take Quiz' button or pressing 'Confirm Stretching Complete'.

[0334] If there is no input for a preset time (e.g., 30 seconds) after the stimulus output, the system determines this as a 'sleep' or 'otolith' state and switches to a power-saving mode or automatically terminates the learning session to preserve the accuracy of the data.

[0336] An example of this is as follows.

[0337] Assume a situation where a Type D learner's RCI drops to 0.4 while taking a lecture, and 'decreased concentration' is detected.

[0338] The system analyzes the characteristics of Type D (goal orientation) and the cause of decline (eye distraction) to select a gamification quiz from the challenge-type DB. The lecture is paused, and a multiple-choice quiz pops up in the center of the screen with the message, "Wait! What was the key term that just passed by? (Experience +50 for correct answer)."

[0339] Type D learners are simultaneously driven by frustration over the interrupted lecture and a competitive spirit to answer the quiz, causing them to immediately turn their attention to the screen to solve the problems. Upon answering correctly, they feel a sense of accomplishment and resume the lecture with high concentration.

[0341] As such, the present invention is not merely a simple 'drowsiness alert,' but rather has the effect of minimizing the learner's discomfort while inducing active refocus by providing customized content utilizing psychological mechanisms based on DISC types in a timely manner.

[0343] The customized lecture recommendation server (100) may further include a job persona gap analysis unit (170), and the job persona gap analysis unit (170) is intended to analyze the discrepancy between the user's innate temperament and the job environment currently being performed (or targeted) to predict potential job stress and alleviate it.

[0344] The above job persona gap analysis unit (170) analyzes the user's natural personality (Natural DISC, N) graph and the job-related personality (Adaptive DISC, A) graph required in the job by overlapping them, and calculates a job stress index (JSI) that quantifies the discrepancy between the two indicators.

[0345] To this end, the job persona gap analysis unit (170) is composed of a job profiling module, a gap calculation module, and a solution matching module, and the specific operating principles of each module are as follows.

[0347] The job profiling module performs the function of converting qualitative personality and job characteristics into quantitative four-dimensional vector data. To this end, it operates by being subdivided into a natural propensity extraction unit and a job requirement propensity definition unit, and the specific processing process is as follows.

[0348] First, the DISC test results performed by the user in the most psychologically comfortable state are received from the DISC type analysis unit (120). Instead of simply obtaining a result value of "Type D," raw scores for each of the D, I, S, and C items are extracted and converted into a ratio relative to the total sum score to normalize them to a value between 0 and 100. Through this, a natural disposition vector N = [N_d, N_i, N_s, N_c] representing the user's unique temperament is generated.

[0349] Since job environment requirements vary by company or department, this module derives personalized job propensity vectors through a hybrid approach that combines reference to a standard job database with user experience surveys.

[0351] Step 1: Load Standard Job Template

[0352] When a user selects their job category (e.g., Sales, R&D, Management Support, etc.), the system retrieves the standard DISC model for each job function already stored in the database to set the primary threshold values. (e.g., when 'Sales' is selected, the standard vector A_std = [80, 90, 30, 20], which represents high sociability and initiative, is loaded.)

[0354] Stage 2: Perceived Difficulty Adjustment

[0355] Since the standard model alone cannot reflect the specific culture of individual companies, a brief survey (5–10 questions) is conducted to reflect the users' perceived difficulty level.

[0356] Survey Example: "How frequently does your job require immediate response to unexpected situations?" (Measurement of D propensity), "Is precise review necessary to avoid data errors?" (Measurement of C propensity)

[0357] The user's response value (1-5 point scale) is converted into a weight to fine-tune the previously loaded primary threshold value (A_std). Through this, the final work-related tendency vector A = [A_d, A_i, A_s, A_c] is determined, which reflects the pressure the user feels in the actual field.

[0359] To verify whether the two extracted vectors (N and A) are suitable for gap analysis, the magnitude of the vectors is verified. If the value of a specific vector falls outside the valid range (e.g., insufficient sum score) due to insincere user response, an error handling logic is executed to request re-entry, thereby ensuring the reliability of the data. The two 4-dimensional vectors (N, A) generated in this way are passed as input variables to the gap calculation module described later, and are used as basic data for calculating job stress.

[0361] The gap calculation module compares the natural propensity vector (N) and the job propensity vector (A) extracted through the job profiling module to calculate the total amount of adaptive energy that the user must consume while performing the job, and calculates this as a quantified job stress index (JSI).

[0362] This invention adopts a method of summing not by simply adding the differences between two tendencies, but by reflecting the importance of each personality factor in the job and the difficulty based on the direction of tendency change as weights. The definitions of specific calculation factors and the calculation procedure are as follows.

[0364] This module defines the following three key factors and utilizes them in calculations for precise stress calculation.

[0365] 1) Absolute value of tendency discrepancy

[0366] The absolute disparity value is the positive difference between the natural disparity score and the work disparity score for each of the four DISC attributes (D, I, S, C). This factor was established because the greater the disparity, the higher the intensity with which the user must act against their nature, which is the most fundamental cause of job stress.

[0367] The absolute value of the above-mentioned disparity is calculated as follows. The system converts the natural disparity (N) and work disparity (A) data input by the user into a 4-dimensional vector (D, I, S, C) and calculates the difference in scalar values ​​for each dimension. At this time, each disparity score is input normalized on a scale from 0 to 100 points, and the system derives a simple difference value as a positive integer using the formula |(N_k) - (A_k)| (where k is the D, I, S, C attribute).

[0369] 2) Job Importance Weighting

[0370] The job importance weight is a coefficient indicating how essential a specific DISC attribute is in performing the job. The reason for establishing this factor is that, even with gaps of the same size, a gap arising from core competencies causes much greater psychological pressure than a gap arising from secondary competencies.

[0371] The above job importance weights are calculated as follows. These values ​​are derived by referring to the job competency database within the system. When a user selects a job group, the system loads the required level for each DISC attribute of that job group. The system classifies attributes with a loaded required level within the top 30% of all attributes (or an absolute score of 70 points or more) as 'core competencies' and assigns a first weight (e.g., 1.5), and in other cases, assigns a basic weight (1.0) or a second weight (0.8), thereby determining the value through a differential allocation method based on intervals.

[0373] 3) Adaptation Difficulty Coefficient

[0374] The adaptation difficulty coefficient is a correction factor applied differently depending on whether the adaptation should aim to increase or decrease one's disposition. This factor was established to reflect the fact, from a psychological perspective, that forcibly raising a naturally low disposition (upward adaptation) consumes more energy and intensifies stress compared to suppressing or enduring an overflowing disposition (downward adaptation).

[0375] The above adaptation difficulty coefficient is calculated as follows. This coefficient is determined through a logic that compares the relative relationship between the natural propensity score (N) and the work propensity score (A). The system evaluates the condition (A > N) for each attribute. If this condition is true, that is, if the work requirements are higher than the original propensity and energy must be input, it is determined to be in an 'upward adaptation' state, and a high load coefficient exceeding 1.0 (e.g., 1.2) is assigned. On the other hand, if the condition is false (A≤N), it is determined to be in a 'downward adaptation' state where the propensity is suppressed or maintained, and a normal coefficient (e.g., 1.0) is assigned.

[0377] This system utilizes the factors defined above to derive the final index through the following sequential four-step process over time.

[0379] Step 1) The system compares the relationship between the natural propensity score (N) and the occupational propensity score (A) for each of the attributes D, I, S, and C. Specifically, if 'natural propensity score < occupational propensity score' (upward adaptation), a high load factor (e.g., 1.2) exceeding 1.0 is assigned as the adaptation difficulty factor for the corresponding attribute. Conversely, if 'natural propensity score ≥ occupational propensity score' (downward adaptation or maintenance), a standard value 'general factor (e.g., 1.0)' is assigned.

[0381] Step 2) The 'absolute value of disposition mismatch' is calculated for each attribute, and the individual attribute stress load is calculated by sequentially multiplying this by the job importance weight loaded from the corresponding job database and the adaptation difficulty coefficient assigned in Step 1. According to this process, if a disposition is both critically required for the job and simultaneously one that the user must forcibly cultivate, all three factors are multiplied, resulting in an exponential amplification of the stress load.

[0383] Step 3) Add up the individual stress loads of the four calculated attributes (D, I, S, C) to obtain the total sum. This total sum becomes the user's job stress index (JSI).

[0385] Step 4) The calculated JSI value is compared with a pre-set risk threshold. At this time, the risk threshold is initially set to the value corresponding to the top 30% of the JSI distribution of all user data (e.g., 70 points), and is automatically corrected by reflecting the statistical mean and standard deviation as operational data accumulates. If the JSI exceeds the risk threshold, the system finally classifies it as a burnout risk group and generates a trigger signal to activate a 'mental care solution' rather than simple job training.

[0387] To aid in a concrete understanding, the calculation process is explained by assuming a scenario where a user with an introverted and cautious nature, a 'developer background (Natural Type I Score: 20),' performs the role of 'Sales Team Leader (Work Type I Score: 80),' which requires sociability and outgoingness. (In this case, it is assumed that the importance weight for Type I sales roles is set high at 1.5.)

[0389] [Operations on Type I (Social) Attributes]

[0390] - Absolute value of disposition mismatch: |20(Natural) - 80(Work)| = 60 (Very large gap occurs)

[0391] - Directional analysis: Since nature (20) < work (80), it is an 'upward adaptation' situation where one must force out non-existent social skills. Accordingly, a high load factor (1.2) is applied.

[0392] - Job Importance: Since sociability is essential for sales positions, a weight of 1.5 is applied.

[0393] - Final Load: 60 (Difference) × 1.5 (Importance) × 1.2 (Difficulty) = 108

[0395] [Comparison: Operation on the opposite situation (Type C attribute)]

[0396] If the same user has high meticulousness (Type C) (80) and sales positions require less meticulousness (30), the difference is 50, but since this is a situation where one can 'endure it (downward adaptation),' the coefficient is applied as low as 1.0 and the importance weight is 0.8.

[0397] - Final Load: 50 (Difference) × 0.8 (Importance) × 1.0 (Difficulty) = 40

[0399] The load (108) of the Type I attribute is calculated to be significantly higher than the load (40) of the Type C attribute. This is a much larger gap than a simple score difference (60 vs 50), and it means that the system has accurately quantified the psychological phenomenon that "one experiences the greatest stress when forced to act out important tendencies in the job." Unlike conventional methods that simply compare the difference in the area of ​​the tendency graph, the present invention applies a calculation logic that comprehensively considers job importance and the direction of adaptation (upward / downward), thereby distinguishing the severity of fatigue actually felt by the user even with the same score difference. Through this, the system prevents job abandonment and achieves the effect of increasing long-term work efficiency by prioritizing a recovery program that allows users to return to their natural personality and recharge their energy, rather than pressuring users whose stress index exceeds a threshold by recommending unnecessary job skill lectures (e.g., "how to sell harder").

[0401] The solution matching module receives the Job Stress Index (JSI) and maximum gap attribute information calculated from the gap calculation module, diagnoses the user's condition by comparing it with a pre-set care threshold, and then selects and provides customized solution content optimized for that condition. The core operating principle of this module is to provide solutions by branching them into dual tracks of 'stress care mode' and 'behavior correction mode' depending on the severity of the JSI.

[0403] If the calculated JSI exceeds a threshold (e.g., 70 points), the system determines that the user is at risk of burnout. In this situation, since learning new job skills may actually increase the burden, the system prioritizes matching solutions focused on recovery. Specifically, it provides guidance to help the user shed the mask of their forced work-oriented disposition (A) and return to their natural disposition (N) to rest and feel most comfortable. For instance, if a user with an I (Social) temperament is stressed by isolated analytical work, the system recommends "methods for casual tea time conversations with colleagues" or a "social networking guide to recharge energy after work." It also provides psychological care content to prevent energy leakage originating from the attribute with the largest gap. Specifically, content such as meditation, breathing techniques, or community posts sharing the struggles of people with that disposition are displayed at the top of the recommendation list.

[0405] If the calculated JSI is below the threshold, the system determines that the user has the capacity to perform their duties while enduring stress. In this case, it matches adaptation and skill enhancement solutions that can reduce the gap by increasing work efficiency. First, it trains users on psychological techniques to effectively deflect their work-oriented disposition (A) only during work hours and immediately switch it off after work is finished. This includes efficient deflection strategies that reduce unnecessary emotional drain. Additionally, it proposes methods to compensate for deficiencies in work-oriented disposition through technology rather than personality modification. For example, for a user who makes frequent mistakes due to a lack of Type C (cautious) disposition (natural score < work demands), instead of advising them to "change their personality to be meticulous," it recommends instrumental solutions such as "how to create checklists to prevent mistakes at the source" or "how to use automated review tools."

[0407] The solution matching module intervenes in the ranking algorithm of the AI ​​lecture recommendation unit (130) according to the determined mode. Normally, the keyword 'job relevance' is the first priority criterion for recommendation, but when the stress care mode is activated, the weight of content with the 'psychological stability' tag is increased by 2.0 times and the recommendation list is rearranged. Conversely, in the behavior correction mode, the weight of content with the 'practical tips' and 'know-how' tags is increased.

[0409] Specific examples of this are as follows.

[0410] - Situation: An introverted developer (low Type I score) performs a sales role (high Type I demand) and has a JSI of 85 (risk group).

[0411] - Judgment: 'Stress Care Mode' activated as the threshold (70 points) has been exceeded.

[0412] -Solution:

[0413] The "How to Sell Well" course (previously ranked 1st) was moved down to a lower rank in the recommendation list.

[0414] Instead, mental care content such as "Energy Recharge Methods for Introverts" and "Building Mental Resilience to Avoid Rejection" is displayed as a pop-up at the top.

[0415] Displays a customized message saying, "Take off your work mask for a moment and recover with some time alone."

[0417] As such, this module does not stop at simply displaying the degree of job mismatch in a numerical value, but actively decides whether to provide 'rest' or 'weapons' based on the psychological risk inherent in that value, thereby preventing mental breakdown in learners and supporting sustainable job performance.

[0419] The inspector terminal (200) is a terminal held by an inspector to perform disk inspection and communicates with the customized lecture recommendation server (100) via a network.

[0421] In this way, by providing personalized educational content through disk type analysis via disk inspection according to the present invention, lectures optimized for the learner (examiner) can be recommended, thereby enabling the learner to achieve self-development more effectively through lectures tailored to their personality and learning needs.

[0423] FIG. 3 is an exemplary diagram showing the overall configuration of the present invention. When a user (examiner) takes a DISC test, the user receives a test result regarding a personality type through a DISC type analysis module (a), receives a recommendation for a lecture based on the test result through an artificial intelligence recommendation module (b), receives a user profile, a list of recommended lectures, etc. through a user interface module (c), and when taking a recommended lecture, the user (examiner) receives a learning evaluation and feedback, etc., and the accuracy of the artificial intelligence recommendation module can be improved by retraining the artificial intelligence model with this information (d).

[0425] The embodiments described in this specification and the configurations illustrated in the drawings are merely the most preferred embodiments of the present invention and do not represent all of the technical ideas of the present invention; therefore, it should be understood that there may be various equivalents and modifications that can replace them. Explanation of the symbols

[0426] 100 Customized Lecture Recommendation Server 110 Disk Check Database Section 120 Disk Type Analysis Department 130 AI Lecture Recommendation Department 131 Lecture Database Department 132 Personalized Recommendation AI Algorithm Study 133 Recommendation Result Update Section 140 User Interface Section 141 Personalized Dashboard Section 142 Profile Update Department 150 Growth Tracking and Feedback Section 151 Learning Outcome Tracking Department 152 Feedback System Department 200 inspector terminals

Claims

Claim 1 The system includes a customized lecture recommendation server (100) for providing customized recommended lectures by receiving test performance information based on disk analysis test information from a tester terminal (200) and analyzing the provided test performance information based on artificial intelligence. The customized lecture recommendation server (100) includes a disk test database unit (110) for storing disk analysis test information including predetermined questions, a disk type analysis unit (120) for providing analysis result information including personality types analyzed through the provided test performance information received via the tester terminal (200) after performing disk analysis test information, an artificial intelligence lecture recommendation unit (130) for providing recommended lecture information by analyzing the analysis result information provided via the disk type analysis unit (120) through a customized recommendation AI model for recommending customized lectures, and a user interface unit (140) for providing a learning dashboard for checking recommended lecture information and profiles. The artificial intelligence lecture recommendation unit (130) includes lecture content style The analysis unit (134) is further included, and the lecture content style analysis unit (134) includes a voice speed and tone analysis module that calculates the effective speech rate excluding silent intervals through VAD in the audio track of the lecture video and quantifies the instructor's tone by analyzing the frequency fluctuation range of the voice waveform, a visual stimulus density analysis module that calculates the number of scene transitions through the color histogram difference value between video frames and calculates the ratio of text and image occupancy through OCR and object recognition, and a linguistic structure analysis module that quantifies the linguistic structure of the lecture by analyzing the frequency of occurrence of logical conjunctions and emotional adjectives through natural language processing (NLP) in the text converted into STT (Speech-to-Text), and the lecture content style analysis unit (134) derives a lecture style attribute value (F) normalized between 0 and 1 by calculating the percentile rank that the analysis data of an individual lecture occupies within the entire lecture data population.An AI-based DISC type analysis and customized lecture recommendation system characterized by calculating the difference between the above lecture style attribute value (F) and the optimal user preference value (P) according to the user's DISC type, calculating a total penalty value by multiplying the difference value by the user's type-specific sensitivity weight (w) for the corresponding attribute, squaring the result, and summing the result, and calculating a DISC fit index through an operation in which a correction constant is added to the total penalty value and the reciprocal is taken, and selecting lectures in which the DISC fit index is greater than or equal to a preset threshold as a recommendation list. Claim 2 In claim 1, the AI ​​lecture recommendation unit (130) updates the user preference optimal value (P) and the type-specific sensitivity weight (w) based on the user's learning history, and corrects the user preference optimal value (P) to a personalized value by weighting the lecture style attribute value (F) of the lectures completed by the user, and for lectures that the user dropped out of midway, detects an attribute showing a positive correlation in which the dropout rate increases as the gap between the user preference optimal value (P) and the lecture style attribute value (F) increases, and adjusts the type-specific sensitivity weight (w) of the corresponding attribute upward. This characterizes an AI-based disk type analysis and customized lecture recommendation system. Claim 3 delete Claim 4 delete Claim 5 delete Claim 6 delete

Images