User State Indicator-Based Cognitive Training Control Method

Abstract

The present invention relates to the field of artificial intelligence-based digital healthcare platform technology, and more specifically, to a user state indicator-based cognitive training control method that collects and analyzes multidimensional data including a user's cognitive state, psychological state, physiological state, and surrounding environment in real time, and dynamically provides cognitive training content of a difficulty level and type optimized for the individual based on this, thereby maximizing the effects of recovery and enhancement of brain cognitive function.

Description

Technology Field

[0001] The present invention relates to the field of artificial intelligence-based digital healthcare platform technology, and more specifically, to a user state indicator-based cognitive training control method that collects and analyzes multidimensional data including a user's cognitive state, psychological state, physiological state, and surrounding environment in real time, and dynamically provides cognitive training content of a difficulty level and type optimized for the individual based on this, thereby maximizing the effects of recovery and enhancement of brain cognitive function. Background Technology

[0002] With the recent entry into an aging society and the widespread use of digital devices, the demand for digital-based cognitive training programs targeting diverse users, including patients with mild cognitive impairment, the elderly, and children with Attention Deficit Hyperactivity Disorder (ADHD), is surging.

[0003] Conventional digital cognitive training systems have mostly adopted methods that follow fixed, pre-designed training plans or adjust the difficulty of the next training session based solely on fragmentary indicators of the user's previous training performance, such as accuracy rates or reaction times. However, this approach had limitations in that it failed to consider complex and variable factors, such as the user's actual condition on the day of training, accumulated fatigue, environmental factors that distract from concentration, or psychological receptiveness to new tasks.

[0004] As a result, users were provided with training that was excessively difficult or easy and did not match their current state, leading them to easily lose interest or feel frustrated. This increased the dropout rate and became a major cause of hindering long-term training effectiveness. In addition, there was a problem in that uniform types of content were provided to all users, failing to offer customized motivation strategies tailored to individual tendencies or emotional states.

[0005] Accordingly, there is an urgent technical need for an advanced personalized training delivery platform that can detect the multifaceted state of a user in real time, comprehensively analyze it, and dynamically optimize training plans. Prior art literature

[0006] Korean Registered Patent No. 10-1564168 The problem to be solved

[0007] The present invention was devised to solve the problems of the prior art as described above, and aims to solve the following technical problems.

[0008] The invention provides a training delivery method capable of precisely adjusting the difficulty of the next training session in real time by comprehensively considering not only the user's previous training performance but also psychological resilience after failure, accumulated cognitive fatigue, and the level of interference from the surrounding environment during training.

[0009] The goal is to provide a training delivery method that dynamically recommends the optimal content type capable of inducing the highest level of immersion and training effectiveness for the current user by comprehensively analyzing the user's real-time emotional state and individual adaptability to new types of training.

[0010] The goal is to provide an AI-based customized training platform that can reduce the dropout rate and maximize long-term training adherence and effectiveness by simultaneously optimizing difficulty and type based on the aforementioned multidimensional judgment criteria and organically combining them to provide users with a hyper-personalized training experience.

[0011] The problems of the present invention are not limited to those mentioned above, and other problems not mentioned will be clearly understood by those skilled in the art from the description below.

[0012] The problems of the present invention are not limited to those mentioned above, and other problems not mentioned will be clearly understood by those skilled in the art from the description below. means of solving the problem

[0014] In an embodiment of the present invention for solving the above problem,

[0015] A performance data collection module that receives basic user information and multiple heterogeneous data generated during cognitive training from a user terminal;

[0016] A data processing and indicator calculation module that analyzes the received plurality of heterogeneous data to calculate a plurality of quantitative indicators representing the user's cognitive state, psychological state, physiological state, and surrounding environment state;

[0017] A decision-making and scheduling module that dynamically determines the difficulty and type of training content for the user by applying predefined multi-stage logic rules based on a plurality of quantitative indicators calculated above; and

[0018] A training content providing module that retrieves training content corresponding to the above-determined difficulty and type from a database and provides it to the user terminal;

[0019] Includes,

[0021] The above decision-making and scheduling module is,

[0022] Regarding the basic difficulty level primarily determined based on the user's previous training performance,

[0023] A first correction rule that raises the basic difficulty level when the failure recovery rate indicator, which represents the degree of performance recovery immediately after an incorrect answer occurs and is calculated by the data processing and indicator calculation module, exceeds a first threshold; and

[0024] A second correction rule for lowering the basic difficulty level or the upwardly adjusted difficulty level when the combined value of the cognitive fatigue residual amount indicator, which indicates the user's cumulative fatigue, and the environmental cognitive disturbance index, which indicates the degree of disturbance in the surrounding environment, calculated by the data processing and indicator calculation module, exceeds a second threshold;

[0025] It is characterized by determining the final difficulty level by applying sequentially,

[0026] The above failure recovery rate indicator is,

[0027] It is calculated by comparing the average correct answer rate during a preset number of times (2 to 5 times) immediately after an incorrect answer occurs in the user's training log with the overall average correct answer rate of the corresponding training session,

[0028] The above residual cognitive fatigue indicator is,

[0029] Heart rate variability data received from a wearable device linked to the above-mentioned user terminal is calculated as the degree of deviation from an individual's stable state baseline, and

[0030] The above environmental perception disturbance index is,

[0031] The ambient noise level measured by the microphone of the user terminal is characterized by being calculated based on the duration during which the level exceeds a preset decibel value.

[0032] The above decision-making and scheduling module is,

[0033] A first filtering step for primarily selecting a first candidate content group that is pre-matched with the emotional state from the entire list of training content, based on an emotional variability factor representing the user's current emotional state and the range of change thereof, calculated by the data processing and indicator calculation module; and

[0034] A second sorting step for rearranging the group by assigning a higher priority to content that matches the user's tendencies within the selected first candidate content group, based on a novelty adaptation factor representing the user's past adaptation speed to a new type of training calculated by the data processing and indicator calculation module;

[0035] It is characterized by performing the above steps sequentially and determining the content with the highest priority in the reordered group as the final recommended content type,

[0036] The above emotional volatility factor is,

[0037] An expression recognition algorithm is performed on a user's expression image obtained through the front camera of the user terminal, and the ratio of negative emotional expressions and the frequency of changes in emotional state are combined to be calculated.

[0038] The above novel adaptation factor is,

[0039] It is characterized by being calculated based on the gap between the average performance of a specific number of initial attempts when the user first encounters a new type of training content, the average performance at a point where performance stabilizes thereafter, and the number of training attempts required until the point of stabilization.

[0040] The above decision-making and scheduling module is,

[0041] First, determine the final recommended content type through the procedure of Paragraph 4 above, and

[0042] The final difficulty level is determined by performing the procedure of Paragraph 2 above for the above-determined content type, and

[0043] If the conditions of the first correction rule and the second correction rule of the above-mentioned second paragraph are simultaneously satisfied, the downward adjustment according to the second correction rule may be applied preferentially. Effects of the invention

[0045] The AI-based customized training platform for the recovery and enhancement of brain cognitive function according to the present invention provides the following effects.

[0046] It maximizes training effectiveness by providing hyper-personalized training paths. This invention adjusts training difficulty in real time by synthesizing multidimensional indicators that were not considered in conventional technology, such as failure recovery rate, residual cognitive fatigue, and environmental cognitive disturbance index. This ensures that users always receive cognitive challenges best suited to their current state, and maintains an optimal learning curve by preventing frustration caused by excessive load or boredom caused by tasks that are too easy.

[0047] This invention enhances training immersion by recommending content that aligns with the user's internal state. It dynamically recommends content types by analyzing the user's real-time emotional state and individual propensity for new stimuli. This increases the user's emotional stability and induces a positive attitude toward training, while leading the user to voluntarily participate in training by providing interest and motivation for the training itself through content tailored to their individual tendencies.

[0048] It ensures the long-term continuity of training, significantly reducing the dropout rate. By proactively detecting negative factors such as user fatigue or the surrounding environment and reflecting them in the training plan, this invention prevents users from becoming exhausted or giving up due to excessive training. Such an adaptive system, which carefully considers the user's condition, enhances trust in the platform and, consequently, secures long-term training adherence, thereby increasing the likelihood of achieving the ultimate goal of cognitive function improvement.

[0049] Data-driven, objective, and reliable training management is possible. This invention adjusts training according to clear and reproducible logical procedures based on multiple objective data collected in real time, without relying on the subjective judgment of the training provider or fixed protocols. This ensures the transparency and reliability of the training process and provides a foundation for systematically tracking and managing changes in the user's status through the collected data.

[0050] The effects according to the present invention are not limited to those exemplified above, and a wider variety of effects are included within the present invention. Brief explanation of the drawing

[0052] Figure 1 illustrates an overall relationship diagram according to the present invention. Figure 2 illustrates the organic operation flowchart between all components. Figure 3 illustrates a flowchart of dynamic training difficulty adjustment. Figure 4 illustrates a flowchart of personalized content recommendation. Figure 5 illustrates a flowchart of emergency schedule adjustment. Figure 6 illustrates a flowchart of the optimal training time zone recommendation. Specific details for implementing the invention

[0053] Hereinafter, various embodiments are described in more detail with reference to the attached drawings. The embodiments described in this specification may be modified in various ways. Specific embodiments may be depicted in the drawings and described in detail in the detailed description. However, specific embodiments disclosed in the attached drawings are intended only to facilitate understanding of various embodiments. Accordingly, the technical concept is not limited by specific embodiments disclosed in the attached drawings, and it should be understood that it includes all equivalents or substitutions that fall within the spirit and scope of the invention.

[0054] Terms including ordinal numbers, such as first, second, etc., may be used to describe various components, but these components are not limited by the aforementioned terms. The aforementioned terms are used solely for the purpose of distinguishing one component from another.

[0055] Functions related to artificial intelligence according to the present disclosure are operated through a processor and memory. The processor may be composed of one or more processors. In this case, the one or more processors may be general-purpose processors such as CPUs, APs, and DSPs (Digital Signal Processors), graphics-dedicated processors such as GPUs and VPUs (Vision Processing Units), or artificial intelligence-dedicated processors such as NPUs. The one or more processors control the processing of input data according to predefined operation rules or artificial intelligence models stored in memory. Alternatively, if the one or more processors are artificial intelligence-dedicated processors, the artificial intelligence-dedicated processors may be designed with a hardware structure specialized for processing a specific artificial intelligence model.

[0056] The predefined operating rules or artificial intelligence models are characterized by being created through learning. Here, being created through learning means that a predefined operating rules or artificial intelligence models configured to perform a desired characteristic (or objective) are created by a basic artificial intelligence model being trained using a number of training data by a learning algorithm. Such learning may be performed on the device itself where the artificial intelligence according to the present disclosure is executed, or it may be performed through a separate server and / or system. Examples of learning algorithms include supervised learning, unsupervised learning, semi-supervised learning, or reinforcement learning, but are not limited to the examples described above.

[0057] An artificial intelligence model can be composed of multiple neural network layers. Each of the multiple neural network layers has multiple nodes and weight values, and performs neural network operations through calculations between the results of previous layers and the multiple weights. The multiple weights possessed by the multiple neural network layers can be optimized based on the learning results of the artificial intelligence model. For example, multiple weights can be updated so that the loss value or cost value obtained by the artificial intelligence model during the learning process is reduced or minimized. Additionally, to minimize the loss value or cost value, multiple weights can be updated in a direction that minimizes the gradient associated with the loss value or cost value. Artificial neural networks may include deep neural networks (DNNs), such as Convolutional Neural Networks (CNNs), Deep Neural Networks (DNNs), Recurrent Neural Networks (RNNs), Restricted Boltzmann Machines (RBMs), Deep Belief Networks (DBNs), Bidirectional Recurrent Deep Neural Networks (BRDNNs), or Deep Q-Networks, but are not limited to the examples mentioned above.

[0058] 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).

[0059] Terminals can be implemented in various forms. For example, the terminals described in this specification may include mobile terminals such as smartphones, tablet PCs, PDAs, portable multimedia players, and MP3 players, as well as fixed terminals such as smart TVs and desktop computers.

[0060] In this specification, terms such as “comprising” or “having” are intended to specify the existence of the features, numbers, steps, actions, components, parts, or combinations thereof described in the specification, and should be understood as not precluding the existence or addition of one or more other features, numbers, steps, actions, components, parts, or combinations thereof. When a component is described as being “connected” or “connected” to another component, it should be understood that it may be directly connected to or connected to that other component, or that there may be other components in between. On the other hand, when a component is described as being “directly connected” or “directly connected” to another component, it should be understood that there are no other components in between.

[0061] Meanwhile, a "module" or "part" for a component as used in this specification performs at least one function or operation. Furthermore, a "module" or "part" may perform a function or operation by hardware, software, or a combination of hardware and software. Additionally, a plurality of "modules" or a plurality of "parts," excluding a "module" or "part" that must be performed on specific hardware or on at least one processor, may be integrated into at least one module. A singular expression includes a plural expression unless the context clearly indicates otherwise.

[0062] In addition, power, power transmission, and control therefor for the following assembly configurations and embodiments, including "by control," follow conventional technology including terminals, applications, hardware control modules, etc., so they are omitted to avoid redundancy.

[0063] In addition, the operation embodiments and configurations described in a general manner without being explained in detail below follow the prior art and are omitted in order to focus on describing the purpose of the present invention and the resulting effects.

[0064] Furthermore, in describing the present invention, if it is determined that a detailed description of related known functions or configurations may unnecessarily obscure the essence of the invention, such detailed description is abbreviated or omitted.

[0065] The user state indicator-based cognitive training control method of the present invention comprises a user terminal (200) that directly interacts with the user, a platform server (100) that performs core data processing and artificial intelligence operations, and a database (300) that systematically stores and manages all data.

[0066] First, the user accesses the platform through a user terminal (200) and inputs basic information such as their age, gender, and existing diseases. The input information is transmitted to the platform server (100) via a communication network.

[0067] The user information acquisition module (110) within the platform server (100) is responsible for receiving the transmitted user's basic information and storing it in the database (300).

[0068] Subsequently, the training content provision module (120) retrieves the user's basic information stored in the database (300) and, based thereon, provides the user terminal (200) with initial cognitive training content that is pre-set to be suitable for the individual.

[0069] More specifically, the process of selecting the aforementioned 'pre-set initial cognitive training content' is based on an 'initial content recommendation rule table' constructed by synthesizing clinical data and expert opinions in advance. The table takes the user's age group, diagnosed disease name, and training goal selected upon initial registration (e.g., 'improvement of memory concentration,' 'improvement of daily living performance ability') as input conditions, and defines the optimal initial training content type and difficulty level corresponding to these as output values. For example, if conditions such as '60s, mild cognitive impairment, improvement of memory concentration' are input, the table designates 'memory recall training, Level 3' as the initial content. Furthermore, the table is not static; instead, it continuously improves the accuracy of recommendations by retraining the initial training performance data of all users accumulated on the platform every three months and dynamically updating the mapping relationship to the content with the highest average success rate for each condition.

[0070] [Initial Content Recommendation Rule Table]

[0072] *

[0073] The above initial content recommendation rule table represents an example of a rule table for providing a personalized starting point based on the user's initial information. For example, if a user in their 60s with mild cognitive impairment in one example selects "memory concentration improvement" as a goal (row in bold in the table), the system refers to this table and automatically selects "memory recall training, level 3" as the initial training content.

[0074] When a user performs the provided training content on a user terminal (200), various performance data such as reaction speed, accuracy, and number of attempts are generated during this process. The performance data collection module (130) of the platform server (100) collects the performance data in real time and accumulates and stores it in the database (300) in the form of time-series data.

[0075] The cognitive state analysis module (140) operates an artificial intelligence model based on user performance data accumulated in the database (300). The artificial intelligence model analyzes the trends in changes to the user's cognitive function over time and unique response patterns to determine the current user's cognitive state as an objective indicator. Here, the 'artificial intelligence model' refers to a combination of multiple models and algorithms that perform different purposes, rather than a single model. Specifically, to analyze facial expression and voice data, a pre-trained Convolutional Neural Network (CNN)-based emotion classification model is used; to detect abnormal patterns in biosignals, a statistical outlier detection algorithm (e.g., 3-sigma rule, Z-score) is used; and to calculate performance-based indicators such as failure recovery rates, an explicitly defined stepwise computation algorithm is used. A set of quantitative values, such as multiple individual indicators calculated through various models and algorithms, such as 'failure recovery rate: 0.85' and 'remaining cognitive fatigue amount: 0.6', corresponds to the aforementioned 'objective indicators' and is used as input values ​​for the decision-making and scheduling module described later.

[0076] Finally, the training adjustment module (150) determines the difficulty and type of the next training in real time to enhance the user's cognitive recovery and enhancement effects, based on the analysis results received from the cognitive state analysis module (140). The determined information is transmitted to the training content provision module (120), and the training content provision module (120) provides new training content suitable for the adjusted difficulty and type to the user terminal (200).

[0077] By repeating this series of processes, the platform of the present invention can continuously provide customized cognitive training that actively responds to changes in the individual user's state.

[0078] More specifically, the user information acquisition module (110) is configured to collect and process basic data necessary for the personalization of cognitive training when a user first joins the platform or updates information, and receives from the user not only static information such as age, gender, and diagnosed disease name, but also dynamic 'recent status' information such as condition tags subjectively selected by the user and resting heart rate data measured by an external wearable device. The module receives the basic data through an input unit of the user terminal (200) using an encrypted protocol, processes the received data according to a predetermined format, and transmits it to the user information storage unit of the database (300).

[0079] The above-described performance data collection module (130) is configured to collect all data generated while the user performs cognitive training in real time and transmit it for subsequent analysis. It collects not only reaction speed, accuracy, and number of attempts, which are direct performance indicators of the training, but also collects heterogeneous data in combination to calculate the core judgment criteria of the present invention. Specifically, it performs the function of collecting data in real time, including biosignal data such as voice, facial expressions, surrounding environment, and heart rate variability, from a microphone, camera, and light sensor embedded in the user terminal (200) and an external wearable device. The module receives the heterogeneous data from the user terminal (200) in a streaming or periodic batch form, attaches a timestamp, and transmits it to the data processing and indicator calculation module described later.

[0080] The above data processing and indicator calculation module is configured to process and analyze original data received from the above execution data collection module (130) and to calculate quantitative key indicators necessary for the decision-making and scheduling module to make a judgment, which will be described later. The module includes a plurality of sub-components internally to independently calculate each indicator, and the sub-components include a failure recovery rate calculation unit, a cognitive fatigue residual amount calculation unit, an emotional variability factor calculation unit, and a biological state risk index calculation unit. The failure recovery rate calculation unit calculates (1) the correct answer rate (A) for a preset number of times immediately after an incorrect answer occurs in the user's training log, and (2) the overall average correct answer rate (B) of the corresponding training session, and then calculates the value of (A - B) as the final 'failure recovery rate' index. The above cognitive fatigue residual amount calculation unit calculates (1) the percentage difference (C) between the real-time heart rate variability value and the individual resting state baseline, and (2) the percentage (D) of the touchscreen input delay time delayed relative to the individual average, and then applies a predefined weight (e.g., 0.7C + 0.3D) to calculate the final 'cognitive fatigue residual amount' index. Through the above specific step-by-step algorithm, the reproducibility of the 'analysis' and 'calculation' processes is ensured. The above failure recovery rate calculation unit calculates the 'failure recovery rate' by analyzing the user's training log and comparing the performance immediately after an incorrect answer occurs with the overall average performance; the above cognitive fatigue residual amount calculation unit calculates the 'cognitive fatigue residual amount' by analyzing the change patterns of heart rate variability data and touchscreen input latency; the above emotional variability factor calculation unit calculates the 'emotional variability factor' through facial expression recognition and voice tone analysis algorithms; and the above bio-state risk index calculation unit calculates the 'bi-state risk index' by calculating how statistically the real-time bio-signal deviates from the individual stability baseline.The above data processing and indicator calculation module receives original data from the execution data collection module (130) and transmits a plurality of calculated key indicators in the form of structured data to the decision and scheduling module.

[0081] The above 'rule-based engine' and 'decision matrix' are implemented as a set of specific 'if-then' rules. For example, the engine includes explicit rules such as "IF 'failure recovery rate' > 0.8 AND 'remaining cognitive fatigue amount' < 0.4 THEN current difficulty level = current difficulty level + 1". In addition, the engine incorporates a priority processing rule that, when an 'emergency action command' occurs, immediately stops processing all other decision information, such as 'difficulty level' and 'recommendation content identifier', and transmits only the 'emergency action command' to the training content provision module described later, thereby preventing conflicts between decision information and ensuring stable operation of the system.

[0082] The above decision-making and scheduling module is configured to serve as the core brain of the present invention and synthesizes multiple indicators received from the data processing and indicator calculation module to finally determine the difficulty level, content type, time zone, and emergency response measures for the next training. The module incorporates the 'rule-based engine' and 'judgment matrix' of processes 1 to 4 described in the detailed description of the present invention, and the engine derives an optimal training plan by sequentially performing predefined multi-stage logic rules and filtering procedures using indicators such as the received 'failure recovery rate', 'remaining cognitive fatigue amount', and 'emotional variability factor' as input values. The module receives key indicators from the data processing and indicator calculation module and transmits information such as the finally determined 'difficulty level', 'recommended content identifier', 'recommendation time', and 'emergency action command' to the training content provision module (120) to be described later.

[0083] The above training content provision module (120) is an execution configuration that retrieves appropriate training content from the database (300) and provides it to the user terminal (200) according to the finally determined training plan. The module receives a 'difficulty level' and a 'recommended content identifier' from the decision-making and scheduling module, and based on this, searches for and retrieves training content that exactly matches the conditions from the content storage unit of the database (300). Upon initial access, initial content is selected by comparing the user's age group and diagnosed disease name with a pre-established 'initial content recommendation rule table'; for example, in the table, 'memory recall training level 3' is mapped to the condition '60s, mild cognitive impairment'. The module receives a training command from the decision-making and scheduling module, retrieves content data from the database (300), and finally transmits the content to the display unit of the user terminal (200).

[0084] As an example of an organic operation between the components of the present invention, first, a user accesses the platform through a user terminal (200), and a user information acquisition module (110) receives the user's basic information and condition tag and stores it in a database (300). Subsequently, a training content provision module (120) transmits 'Memory Training Level 5' to the user terminal (200) according to an initial content recommendation rule based on the stored information. While the user is performing training, a performance data collection module (130) collects the user's high accuracy data, along with 'smile' facial expressions through a camera and 'stable heart rate variability' data through a wearable device in real time, and transmits them to a data processing and indicator calculation module. The data processing and indicator calculation module analyzes the received data to calculate a plurality of key indicators, such as 'Failure Recovery Rate: Positive', 'Residual Cognitive Fatigue: Low', and 'Emotional Variability Factor: Stable-Positive', and transmits them to a decision-making and scheduling module. The decision-making and scheduling module drives a built-in rule engine. To apply the rule of Process 1, the calculated 'failure recovery rate' indicator value (0.92) exceeds the upward adjustment threshold (0.8), and the sum of the 'cognitive fatigue residual amount' indicator value (0.3) and the 'environmental cognitive disturbance index' indicator value (0.2) (0.5) is less than the downward adjustment threshold (0.7). Therefore, the "current difficulty level + 1" rule is finally applied, and the next difficulty level is determined to be 'Level 6'. At the same time, according to the rule of Process 2, 'new type of problem-solving training' is finally determined as the next recommended content by considering the 'stable-positive' emotional state and the user's 'high novelty adaptation factor'. Finally, the command determined above, 'Difficulty: Level 6', 'Content: Problem-solving training', is transmitted to the training content provision module, and the module retrieves content with the corresponding conditions from the database and transmits it to the user terminal, thereby completing one complete personalized training cycle.

[0085] The judgment criteria used to dynamically adjust training difficulty in the present invention provide hyper-personalized training by comprehensively evaluating the multifaceted state of the user, and the judgment criteria include a failure recovery rate, residual cognitive fatigue, and an environmental cognitive disturbance index. The failure recovery rate functions as a quantitative indicator of how effectively a user recovers performance in the next attempt immediately after experiencing an incorrect answer or failure during training. Since an incorrect answer is a major event that triggers cognitive conflict, and the change in performance immediately following it most sensitively reflects the user's immediate cognitive and emotional response to that conflict, the present invention uses it as an objective proxy indicator to measure psychological resilience and engagement with the task. The residual cognitive fatigue represents the user's physiological and cognitive fatigue level accumulated from previous training sessions or daily activities that affects current training performance ability; it prevents mistaking a decline in performance caused by a temporary deterioration in the user's condition for a lack of training ability and prevents burnout caused by excessive training, thereby ensuring the long-term sustainability of training. The aforementioned environmental cognitive disturbance index is a quantified indicator of the negative impact that noise, lighting, and visual complexity of the surrounding environment where training is performed have on the user's concentration. By clearly separating and analyzing changes in the user's own cognitive abilities from performance changes caused by external environmental factors, it enables more accurate difficulty adjustment.

[0086] The difficulty adjustment procedure of the present invention is composed of a three-stage sequential correction process rather than a batch calculation method based on a single formula. This ensures system stability and predictability by independently evaluating judgment criteria of different natures and applying clear priority rules that prioritize the prevention of user overload. In the first stage, a basic difficulty level is set based on the user's previous performance. In the second stage, the possibility of adjusting the difficulty upward is reviewed by evaluating the failure recovery rate, which represents the user's positive psychological state. In the third stage, the user's internal and external load factors, such as the residual amount of cognitive fatigue and the environmental cognitive disturbance index, are combined to make a final determination on whether to adjust the difficulty downward. Even if an upward adjustment is decided in the second stage, if the conditions for downward adjustment in the third stage are met, a command is transmitted to apply the downward adjustment preferentially in accordance with the user protection principle. In addition, the adjustment values ​​for changing the difficulty level at each stage are determined according to generalized rules: +1 level is added to the current difficulty level if the failure recovery rate satisfies the 'positive' state; -1 level is subtracted from the current difficulty level if only one of the conditions, either the residual cognitive fatigue or the environmental perception interference index, satisfies the 'high' state; and -2 levels are subtracted from the current difficulty level by applying weights if both the residual cognitive fatigue and the environmental perception interference index satisfies the 'high' state.

[0087] The value of each judgment criterion is obtained from the sensor or training log of the user terminal (200). The failure recovery rate is obtained by the execution data collection module (130) analyzing the user's training log and comparing the average correct answer rate for a specific number of attempts immediately after an incorrect answer occurs with the overall average correct answer rate of the session. At this time, the specific number corresponds to the minimum number of attempts that is statistically significant for evaluating the impact of short-term memory, and this number is variably set by the system within the range of 2 to 5 times depending on the complexity of the training type. For example, if the user's overall average correct answer rate was 90% but the correct answer rate for 3 attempts immediately after a specific incorrect answer occurred was higher at 95%, the failure recovery rate is determined to be in a 'positive' state. The cognitive fatigue remaining amount is obtained by the cognitive state analysis module (140) through the heart rate variability sensor of a wearable device linked to the user terminal (200) or the analysis of the terminal's touchscreen input pattern. For example, if the user's standard heart rate variability value in a stable state was an average of 55ms but the value measured in the current training session is 40ms, which is a decrease of more than the threshold ratio set individually relative to the standard, the residual cognitive fatigue is determined to be in a 'high' state, and the threshold ratio is not a fixed value but is dynamically determined through user-specific calibration. The environmental cognitive disturbance index is obtained by the cognitive state analysis module (140) measuring the ambient noise level through the microphone of the user terminal (200) and detecting a rapid change in light intensity through an illuminance sensor. For example, if the average noise level during training exceeds 60dB for more than 1 minute, or if an event occurs three or more times in which the illuminance changes rapidly to 500 Lux or more within 5 seconds, the environmental cognitive disturbance index is determined to be in a 'high' state.

[0088] The 'difficulty level' finally determined through the above three-step sequential correction process is transmitted to the training adjustment module (150), and the module uses this 'difficulty level' value as an identifier to immediately retrieve content that exactly matches the level from among the numerous training contents stored in the database (300) and sends a command to provide it to the user.

[0089] As an example, assume a situation in which a user with mild cognitive impairment begins cognitive training. First, the user recorded an accuracy of 96% and an average reaction time of 480ms in the previous session, and the cognitive state analysis module (140) compares this performance with a 'performance-difficulty mapping table' embedded in the system and sets the basic difficulty level to 'Level 7'. The mapping table consists of rows representing accuracy ranges and columns representing reaction time ranges, with the corresponding difficulty level listed in each cell, and is automatically updated weekly based on the user's performance data over the past month. Subsequently, the performance data collection module (130) analyzes the user's previous session log and confirms that the correct answer rate after an incorrect answer occurred was 98%, which is higher than the overall average. Since this failure recovery rate satisfies the 'positive' criterion, the training adjustment module (150) temporarily raises the difficulty level to 'Level 8' according to the upward adjustment rule. Next, the cognitive state analysis module (140) receives a residual cognitive fatigue signal in a 'high' state indicating that the heart rate variability has decreased by 30% compared to normal from the user's wearable device at the start of training, and simultaneously detects an environmental cognitive disturbance index signal in a 'high' state indicating that the surrounding construction site noise has persisted at 65dB or more from the terminal microphone, and finally determines that the downward adjustment rule 2 is satisfied. Finally, although the upward adjustment factor and the downward adjustment factor conflict, the system prioritizes the downward adjustment according to the highest priority rule of 'prevention of user overload', subtracts -2 levels from the temporarily set 'Level 8' according to the downward adjustment rule 2, determines the final training difficulty as 'Level 6', and controls the training content provision module (120) to provide training content corresponding to 'Level 6' to the user.

[0090] [Performance-Difficulty Mapping Table]

[0091]

[0092] The above performance-difficulty mapping table represents an example of a mapping table that determines the base difficulty based on the user's previous performance. For example, in one example, if the user records an accuracy of 96% and an average reaction speed of 480ms, the system finds the intersection point of the '95% or higher' row and the '400 to 599' column in this table and sets the base difficulty to 'Level 7' (refer to the bolded cell in the table). As another example, if the accuracy is 88% and the reaction speed is 900ms, the base difficulty is set to 'Level 5'. The above table is automatically updated weekly based on the distribution of the user's performance data over the past month, so that the difficulty standard can be adjusted upward in line with the user's growth.

[0093] The logic of the present invention ensures reliability in exceptional situations. If the connection with the wearable device is disconnected and the reception of the remaining perceived fatigue data becomes impossible, the judgment criterion is designed to be treated as a neutral value so as not to affect the final difficulty determination, thereby preventing the loss of some data from halting the operation of the entire system. In addition, if a discrepancy occurs three or more times consecutively between the fatigue level subjectively entered by the user and the objective remaining perceived fatigue signal, the system automatically transmits a command to guide the user through a sensor calibration procedure to continuously manage the reliability of the data. As described above, the 'calibration' function for dynamically setting thresholds and baselines is handled by the data processing and indicator calculation module (140). The module separately stores all heterogeneous data collected from the user during a specific period (e.g., 2 weeks) after the initial use of the system, and performs statistical analysis (calculation of mean, standard deviation, percentile, etc.) to set a personalized initial threshold for each indicator. Subsequently, the module periodically reanalyzes the accumulated data to update the thresholds, thereby continuously adapting to changes in the user's state.

[0094] The threshold values ​​used in the present invention are verified for their rationality and established through the following data-based examples. For example, to determine an upward adjustment threshold for the failure recovery rate indicator, training data of 100 virtual users was generated and the failure recovery rate index distribution was analyzed.

[0095] [Table 1]

[0096]

[0097] As shown in [Table 1] above, the user group with a failure recovery rate index of 0.8 or higher was found to exhibit a high success rate (88% or higher) and a low dropout rate (5% or lower) when the difficulty level was increased. On the other hand, for the group with an index below 0.8, particularly the group below 0.7, a pattern was observed in which the dropout rate increased sharply upon upward adjustment. Therefore, the present invention secures the technical feasibility of setting 0.8 as the threshold for upward adjustment of the failure recovery rate index, which can include the majority of potential growth groups while maintaining a significantly low dropout rate.

[0098] In addition, to verify the effects of the present invention, the results of an example were compared with the prior art having a fixed difficulty adjustment rule and the multidimensional adjustment rule of the present invention.

[0099] [Table 2]

[0100]

[0101] In [Table 2] above, the 'appropriate difficulty matching rate' refers to the ratio of difficulty levels provided that correspond to the user's actual condition (considering fatigue, environment, etc.). The present invention significantly increases the appropriate difficulty matching rate by precisely reflecting the multifaceted condition of the user compared to the prior art, and objective figures prove that this reduces the user's frustration and boredom, leading to a substantial reduction in the dropout rate and an increase in the average performance improvement rate.

[0102] The present invention demonstrates a distinct difference in effect compared to the prior art. For example, in a scenario where the user is in optimal condition, the prior art, which relies solely on accuracy, maintains the difficulty level; in contrast, the present invention provides challenging tasks that draw out the user's potential by detecting a high failure recovery rate and adjusting the difficulty upward. Furthermore, in a scenario where the user is in a state of accumulated fatigue, while the prior art merely lowers the difficulty slightly, the present invention further lowers the difficulty based on a high residual amount of cognitive fatigue to prevent burnout and ensure training continuity. Finally, in a scenario where performance is degraded due to ambient noise, the prior art lowers the difficulty by mistaking this for a decline in cognitive ability; however, the present invention prevents misjudgment caused by external factors by maintaining the difficulty at an appropriate level through the combined consideration of the environmental cognitive disturbance index and the high failure recovery rate.

[0103] The judgment rules adopted in the present invention secure objectivity based on academic and technical necessity. The failure recovery rate identifies users who accept failure as positive feedback based on the 'growth mindset' theory of cognitive psychology, the cognitive fatigue residual amount is based on the fact that excessive load hinders learning effects according to the 'cognitive load theory,' and the environmental cognitive interference index reflects the principle that external stimuli deplete resources allocated to cognitive tasks according to the 'attention resource theory.'

[0104] The adoption of the aforementioned components and judgment rules provides a hyper-personalized training path that realizes a level of precise, personalized training previously unavailable in conventional technology. Furthermore, by proactively managing user fatigue and frustration and compensating for environmental factors, it maximizes training sustainability and significantly reduces the dropout rate. Moreover, by moving away from the subjective judgment of training providers or fixed protocols to determine difficulty levels based on objective data collected in real-time and clear logical procedures, it guarantees the reliability and reproducibility of the training process.

[0105] The terms and thresholds used in this specification are defined according to clear criteria. The failure recovery rate in a 'positive' state refers to a case where the average correct answer rate of a specific number of attempts immediately after an incorrect answer occurs is higher than the overall average correct answer rate of the session by a preset percentage. The residual cognitive fatigue amount in a 'high' state refers to a case where the measured biosignal deteriorates by more than a preset deviation relative to the individual's stable state reference value. All thresholds mentioned in this specification are not fixed constants; rather, they are dynamically set to values ​​reflecting the individual's unique characteristics by statistically analyzing the distribution of individual user data collected during a two-week calibration period upon initial system use, and are periodically updated.

[0106] The judgment criteria used in this invention to recommend training content optimized for the user maximize the immersion and effectiveness of training by comprehensively considering the user's real-time state and personal tendencies, and these criteria include an emotional variability factor and a novelty adaptation factor. The emotional variability factor functions as an indicator that quantifies the user's current emotional state and the range of its changes. Facial expressions and voice are widely used as objective indicators reflecting an individual's internal emotional state as major channels of non-verbal communication; this invention utilizes these known technologies to use detected external signals as input data for inferring the user's current emotional state, which is intended to consider factors that directly influence training acceptance. The novelty adaptation factor is an individual tendency indicator representing how quickly and positively a user adapts to a new type of training task they have not previously experienced. Rapid performance adaptation implies low cognitive resistance and high learning efficiency regarding the task; since this has a high correlation with the user's positive task experience, this invention utilizes performance data based on this relationship as an important basis for judging an individual's potential preferences and stimulus-seeking tendencies.

[0107] The content recommendation procedure of the present invention consists of a two-stage sequential filtering and sorting process rather than a ranking method based on a single score, which enhances the accuracy and level of personalization of recommendations by evaluating the user's current state and long-term tendencies separately. In the first stage, a group of candidate content that aligns with the training goal is initially selected based on the user's real-time emotional state, and in the second stage, the exposure priority of the content is finally reordered within the selected candidate group based on the novelty adaptation factor, which is the user's personal tendency. However, in exceptional cases where the user's novelty adaptation factor is measured to be very high above a specific threshold, the system may apply an auxiliary rule to relax the first-stage filtering and prioritize the inclusion of new content in the recommendation list even when the emotional state is at a 'mild instability' level. This configuration implements a flexible dual-adaptation structure that responds immediately to the user's temporary mood changes while respecting the individual's unique learning style.

[0108] The value of each judgment criterion is obtained from the sensor or log data of the user terminal (200) in a specific manner. The emotional variability factor is obtained by the cognitive state analysis module (140) by performing a facial expression recognition algorithm through the front camera of the user terminal (200) to analyze the ratio of positive, negative, and neutral emotions, or by analyzing changes in tone and pitch of voice through a microphone. For example, if negative emotional expressions account for 40% or more of the user's facial expressions measured for 1 minute before the start of a training session and the frequency of changes in the emotional state exceeds 5 times per minute, the emotional variability factor is determined to be in an 'unstable' state. The novelty adaptation factor is obtained by the performance data collection module (130) by analyzing the user's cumulative training history and calculating the gap and time to reach between the average performance at a specific initial number of times when the user first encountered a new type of training content and the average performance at a later point of stabilization. At this time, the aforementioned initial specific number corresponds to the section where the initial learning effect appears most rapidly in learning curve theory, and the stabilized point corresponds to the average point where performance enters the plateau phase, and these section settings can be dynamically adjusted by the system according to the difficulty of the training type. For example, if a user repeatedly shows a pattern of reaching 90% of their overall average performance level within 5 attempts when exposed to a new type of training, the novelty adaptation factor is judged to be 'high'.

[0109] The training content list, with its priority finally determined through the above two-step process, is transmitted to the training adjustment module (150), and the module transmits a command to finally select the content at the top of the list as the recommended training for the session and provide it to the user.

[0110] As an example, assume a situation in which a user with a tendency to enjoy new challenges begins training in a fatigued but satisfied state after completing important tasks. First, the cognitive state analysis module (140) determines the emotional variability factor as a 'stable-positive' state, characterized by a calm but high frequency of positive word usage, through the analysis of the user's voice tone. Based on this determination result, the training content provision module (120) refers to an 'emotion-content mapping table' embedded in the system to select a first candidate content group consisting of training content tagged with 'cognitive expansion' and 'challenge'. The mapping table defines the relationship between the emotional state and the optimal training type based on cognitive psychology theory. Subsequently, the training adjustment module (150) confirms that the user's accumulated novelty adaptation factor is 'high' and rearranges the list to give the highest priority to 'completely new' types of content that the user has never performed before within the selected first candidate content group. Finally, the training content provision module (120) controls the selection of the ‘multiple task simultaneous processing’ training at the top of the reordered list as the customized recommendation content for this session and provides it to the user.

[0111] [Emotion-Content Mapping Table]

[0112]

[0113] The above emotion-content mapping table represents an embodiment of a mapping table for selecting a primary candidate content group based on a user's real-time emotional state. For example, in one embodiment, if the user's voice tone analysis result determines a 'stable-positive' state (rows highlighted in bold in the table), the system refers to this table to select content tagged with '#cognitive expansion', '#challenge', and '#problem solving' from the total content in the database as the first candidate content group. Conversely, if the user is determined to be in an 'unstable-negative' state due to stress, the system prioritizes selecting and providing deep breathing induction training or simple sensory stimulation training tagged with '#stabilization'.

[0114] The logic of the present invention ensures reliability in exceptional situations. If camera access of the user terminal (200) is blocked and facial expression recognition is impossible, the system prioritizes the use of alternative means such as voice analysis; if even this is impossible, the system processes the emotional variability factor into a 'neutral' state to omit emotion-based filtering during the recommendation process and performs only novelty-based sorting. Additionally, in cases where the novelty adaptation factor cannot be determined due to a lack of accumulated data, such as with a new user, the system initially applies a 'search mode' that operates according to a 'minimum frequency algorithm' which prioritizes recommending content from the category the user has performed the least among all training categories, thereby collecting user propensity data.

[0115] The threshold value used in the content recommendation logic of the present invention is also set through a data-based embodiment. For example, to set a criterion for determining the novelty adaptation factor as 'high', the novel training adaptation patterns of 100 virtual users were analyzed.

[0116] [Table 3]

[0117]

[0118] According to [Table 3] above, the user group that adapts quickly to a new type of training within 5 attempts showed high satisfaction and maintained a low dropout rate when new content was repeatedly recommended. However, when new content was repeatedly recommended to the group that adapts after more than 5 attempts, satisfaction dropped sharply and the dropout rate increased significantly. Through this, the present invention provides a reasonable basis for setting 'within 5 attempts' as the threshold for judging the novelty adaptation factor 'high' as ​​a criterion for identifying a core group that positively accepts new stimuli without performance degradation.

[0119] In addition, to verify the effectiveness of the recommendation method of the present invention, a simulation was performed comparing the prior art using a single criterion (past performance) based recommendation with the two-stage filtering and sorting method of the present invention.

[0120] [Table 4]

[0121]

[0122] The results of [Table 4] above show that the present invention significantly increases the probability of a user selecting recommended content (content selection rate) by recommending content while considering the user's real-time emotional state and personal tendencies. This demonstrates that it increases immersion in training, extends the average training time per session, and ultimately leads to increased satisfaction with the platform, resulting in an improved weekly revisit rate.

[0123] The present invention demonstrates a distinct difference in effect compared to the prior art. For example, unlike the prior art which simply recommends high-difficulty content based on past performance when a user is in an unstable emotional state due to stress, the present invention detects 'unstable' emotions and prioritizes recommending low-intensity stabilization content that does not require attention, thereby preventing training dropout. Furthermore, for users who have become accustomed to specific training and are experiencing a sense of stagnation, while the prior art repeatedly presents the same content, the present invention provides motivation and enhances cognitive flexibility by presenting new types of training based on high novelty adaptability factors.

[0124] The judgment rules adopted in this invention are based on objectivity and necessity. The use of the emotional variability factor is based on the premise that an individual's emotional state and information processing method are closely related according to the theory of the 'mood synchronization effect.' Furthermore, the incorporation of the novelty adaptation factor into recommendations applies the principle that the level of stimulation at which an individual exhibits optimal performance differs according to the 'optimal arousal theory.'

[0125] The adoption of the aforementioned components and judgment rules provides a technical effect that maximizes training immersion by responding in real-time to changes in the user's internal state. Furthermore, through recommendation logic that respects an individual's unique tendencies, it increases long-term training participation rates and consequently enhances cognitive function improvement. Moreover, by separating the recommendation process into distinct stages based on clear criteria of 'emotion' and 'tendency,' it reduces bias in recommendation results and ensures the predictability and stability of the system.

[0126] The terms and thresholds used in this specification are defined according to clear criteria. The emotional variability factor in the 'unstable' state refers to the case where the measured rate of negative emotional expression exceeds a preset threshold. The novelty adaptation factor in the 'high' state refers to the case where a specific percentage of the individual average performance is reached within a preset number of attempts when exposed to a new type of training. All thresholds mentioned in this specification are not fixed constants but are dynamic values ​​that are periodically updated by statistically analyzing the total user data or individual data accumulated during the use of the system.

[0127] In this invention, the judgment criteria used to dynamically adjust the training schedule in response to changes in the user's emergency state are defined as follows, aiming to ensure the user's safety and well-being as the top priority, beyond the effectiveness of the training. The judgment criteria include a bio-state risk index, a state type adjuster, scheduled training importance, and cognitive resilience. The bio-state risk index functions as an indicator that statistically quantifies how far the user's current real-time bio-signals deviate from a baseline of a stable state set for each individual, and objectively measures the severity of changes occurring in the user's health state. The state type adjuster is a weight assigned based on whether the detected emergency state type is physiological or psychological; it assigns a higher priority to physiological crisis signals, such as a sudden increase in heart rate, than to signals of psychological changes, thereby ensuring a priority response to immediate physical danger. The scheduled training importance is a value indicating how important the currently scheduled training session is for achieving the user's long-term cognitive function improvement goals, and serves as the basis for deciding whether to completely cancel the training or switch to low-intensity alternative training in the event of an emergency. The aforementioned cognitive resilience is an individual recovery ability indicator representing how quickly a user recovers normal training performance after experiencing poor condition or training failure in the past, thereby enabling differentiated measures that take the user's tolerance into account.

[0128] The emergency schedule adjustment procedure of the present invention is composed of a multi-stage process based on a rule-based emergency response matrix, rather than a batch suspension method based on a single threshold. This comprehensively assesses the severity and context of the emergency situation to prevent both over-response and lack of response. In the first stage, the risk level of the current state is immediately determined as one of 'Caution', 'Warning', or 'Danger' by combining the biological state risk index and the state type adjuster. In the second stage, only if the determined risk level is not 'Danger', the final response strategy is determined by additionally considering contextual information, such as the importance of scheduled training and cognitive resilience. This configuration ensures that the highest level of response is executed immediately, disregarding all other factors, in the event of a severe crisis signal that could be directly related to the user's life, while taking more flexible and appropriate measures in other situations.

[0129] The value of each judgment criterion is obtained from the user terminal (200) or a connected external device in a specific manner. The above bio-state risk index is calculated by the cognitive state analysis module (140) by determining how many times the user's heart rate and heart rate variability data received from the wearable device deviates from the standard deviation of the pre-set individual stable state baseline. For example, if the current heart rate is measured at 95 bpm while the user's average heart rate is set to 70 bpm and the standard deviation to 5 bpm, this is judged as a very high risk index because it is a deviation exceeding 5 standard deviations. The above state type adjuster is a value pre-defined in the system, and a high weight is assigned to physiological signals such as heart rate and heart rate variability, while a relatively low weight is assigned to psychological signals such as voice tone and input patterns. The above scheduled training importance is obtained through metadata tags pre-assigned to training content within the database (300), and sessions such as initial diagnostic evaluation are classified as 'high', while daily repetitive training is classified as 'low'. The above cognitive recovery resilience is calculated by the performance data collection module (130) analyzing the user's past training logs and calculating the average number of sessions required to return to a normal performance level after a session in which the residual amount of cognitive fatigue was measured to be high.

[0130] The response strategy finally determined through the above multi-stage process is transmitted to the training coordination module (150), and the module transmits a command to immediately change the existing training schedule according to the strategy. The response strategy is configured differentially according to the risk level, such as 'proposing a short-term postponement,' 'providing low-intensity alternative training,' and 'forcibly suspending training and calling an emergency manager.'

[0131] As an example, assume a situation in which a sudden drop in heart rate variability is detected in a user's wearable device while the user is performing training. First, the cognitive state analysis module (140) recognizes that the signal is of the 'physiological' type and applies a high state type modifier weight, and simultaneously calculates a high biological state risk index by calculating that the signal has deviated by more than 4 standard deviations from the individual stable baseline. Subsequently, the system immediately determines the combination of the two indicators as a 'risk' level according to the built-in 'risk judgment matrix'. Finally, the training adjustment module (150) skips the procedure of considering additional contextual information, such as the importance of the training session or the user's resilience, and immediately forcibly stops the training currently in progress according to a priority response strategy predefined for the 'risk' level, and automatically sends an emergency notification to a pre-registered guardian's terminal.

[0132] The logic of the present invention ensures reliability in exceptional situations. If a signal at a 'danger' level is detected but returns to a stable baseline in a series of measurements within 30 seconds thereafter, the system determines this as a possibility of a temporary sensor error and sends a non-urgent log requesting an equipment inspection to the system administrator instead of an emergency notification. Additionally, if a user refuses a 'warning' level response measure (e.g., a rest recommendation) two or more times within a single session, the system considers this as a possibility of concealing danger and raises the danger level by one step to strengthen the response measure.

[0133] The risk level judgment threshold used in the emergency schedule adjustment of the present invention is established through embodiments based on actual emergency medical data and clinical studies. The vital status risk index is defined as the level of deviation from the individual stable baseline by a standard deviation.

[0134] [Table 5]

[0135]

[0136] Table 5 above shows that as the level of deviation from the standard deviation increases, the probability of an actual risk event occurring increases exponentially. In particular, in the range exceeding 4 standard deviations, the probability of a false positive is very low while the probability of an actual risk event occurring is very high, proving the validity of defining this as a 'risk' level and implementing the immediate highest level of response (stopping training, notifying guardians). Since the possibility of false positives still exists in the range of 2 to 4 standard deviations, it supports the need for a multi-stage response that defines this as a 'caution / warning' level and additionally considers contextual information.

[0137] In addition, to verify the effectiveness of the safety management system of the present invention, a simulation was performed comparing the system of the present invention with the prior art that fails to detect danger signals.

[0138] [Table 6]

[0139]

[0140] Table 6 above shows that the present invention can successfully detect most potentially dangerous biosignals and prevent them from developing into serious health problems with an 80% probability through preemptive responses (stopping training, recommending rest, etc.). This objectively demonstrates that the present invention has the effect of substantially protecting user safety, going beyond a simple training platform.

[0141] The present invention demonstrates a distinct difference in effect compared to the prior art. For example, when a user's heart rate rises sharply, the prior art, which detects only a decline in training performance, merely reduces the difficulty level; however, the present invention identifies this as a 'danger' signal, stops the training, and notifies a guardian, thereby preventing potential health problems. Furthermore, unlike the prior art, which fails to detect when a user's voice becomes agitated due to stress, the present invention identifies this as a 'caution' level and helps the user maintain emotional stability by replacing stimulating training with stabilization exercises such as deep breathing.

[0142] The judgment rules adopted in this invention are based on objectivity and necessity. Using heart rate variability as a key biological risk index is based on the scientific fact that heart rate variability is a medically recognized indicator of the balance of the autonomic nervous system, and that a sharp decline can be a precursor to severe stress or health abnormalities. Furthermore, the rule of prioritizing physiological signals over psychological signals follows standard crisis response protocols that prioritize physical stability in emergency situations.

[0143] The adoption of the aforementioned components and judgment rules has the technical effect of expanding the cognitive training platform beyond a simple educational tool into a safety management system that monitors user safety in real time. Furthermore, by detecting early signs of user health abnormalities and responding preemptively, it prevents the occurrence of negative health-related incidents, thereby maximizing the trust of users and guardians in the platform.

[0144] Terms and thresholds used in this specification are defined according to clear criteria. The risk level 'Caution' means when a biosignal deviates by 2 to 3 standard deviations from an individual baseline, 'Warning' means when it deviates by 3 to 4 standard deviations, and 'Danger' means when it deviates by more than 4 standard deviations. All baseline and standard deviation thresholds mentioned in this specification are not fixed constants, but are continuously learned and updated to reflect each user's unique health patterns based on data collected during the calibration period set upon initial system use and subsequent accumulated data.

[0145] The judgment criteria used in this invention to recommend the optimal training time slot to a user are defined as follows. The purpose of this is to induce the user to train during the "golden time," when cognitive performance is highest, rather than at random times, thereby achieving maximum training effects with minimal effort. The judgment criteria include a daily routine compatibility score, a macroscopic stress index, and predicted cognitive fatigue. The daily routine compatibility score functions as a quantitative indicator of how well a specific time slot aligns with the user's unique lifestyle patterns, such as sleep, meals, and work, and serves as the basis for establishing a training schedule that conforms to the individual's biological rhythm. This invention establishes a basic framework for the individual's biological rhythm by prioritizing the user's sleep-wake time data, and utilizes calendar activity data as auxiliary information to identify time slots within this basic framework where cognitive resources are consumed. In cases where sleep patterns are atypical, such as for night shift workers, the system redefines the activity cycle based on sleep time to calculate a score aligned with the individual's actual wakefulness cycle rather than social activity time. The aforementioned macroscopic stress index is an indicator representing the potential level of social stress during a given time period, based on major news affecting society as a whole. It is used to preemptively prevent a decline in concentration caused by significant external events, regardless of an individual's physical condition. Furthermore, the system learns patterns in which users skip training or experience performance declines in response to news of specific categories. By calculating 'individual macroscopic stress sensitivity' based on this data and applying it as a weight, the system differentially reflects the impact of social issues on individuals. The aforementioned predicted cognitive fatigue indicates the level of cognitive fatigue expected during a specific time period, based on the user's past activity data and general circadian rhythms. By scheduling training during times when fatigue has already accumulated, it prevents the expending of inefficient effort.

[0146] The optimal training time slot recommendation procedure of the present invention is composed of a two-stage process that selects candidate groups based on positive factors and filters them based on negative factors, rather than a recommendation method based on a single indicator. This adopts an efficient and logical approach that first identifies the most ideal time slot and then removes exceptionally disruptive time slots from among them. In the first stage, multiple optimal candidate time slot groups most suitable for training within the user's 24-hour day are primarily selected based on the positive factor known as the daily routine compatibility score. In the second stage, negative factors known as the macroscopic stress index and the predicted cognitive fatigue are applied to the selected candidate time slot groups to finally exclude time slots that do not meet the conditions. However, the invention includes a flexible filtering rule that provides the user with the option to select a specific time slot as a 'recommended time requiring caution' when the daily routine compatibility score of that specific candidate time slot exceeds the highest threshold, but the macroscopic stress index or predicted cognitive fatigue is at a mild level.

[0147] The value of each judgment criterion is obtained from various internal and external data sources in a specific manner. The daily routine conformity score is calculated by the training adjustment module (150) receiving data such as sleep start and end times, meal times, and major meeting times from the user's smartwatch or calendar application, and assigning a higher score the further the time period is from these major activity times. For example, if the user's average wake-up time is 7:00 AM and sleep time is 11:00 PM, time periods with high alertness levels, such as 10:00 AM or 3:00 PM, receive a high conformity score. The macroscopic stress index is calculated by the module collecting major news in the social and economic fields in real time through an application programming interface of an external news provider, and analyzing the frequency and intensity of predefined negative keywords to obtain a value between 0 and 1. The 'intensity' is measured based on the negative sentiment score assigned to each keyword by referring to a pre-established 'sentiment vocabulary dictionary'. For example, negative scores are assigned differentially, such as -0.9 for 'crash' and -0.4 for 'concern,' and the system calculates the final index by weighting the average of the scores of all negative keywords within the article. The aforementioned predictive cognitive fatigue is calculated by the module analyzing the user's past activity logs to learn the time-of-day patterns where training performance is lowest on average, and combining this with general circadian rhythms.

[0148] The 'final recommended time zone group' that is finally filtered through the above two-step process is generated as a list by the training adjustment module (150), and the system sends a notification to the user presenting the list in the form of "Today's recommended training time."

[0149] As an example, assume a user who is an office worker and primarily works on weekday mornings. First, the training adjustment module (150) analyzes the user's calendar data to identify that the time from 9:00 AM to 12:00 PM and from 1:00 PM to 6:00 PM is for meetings and work, and from 11:00 PM onwards is for sleep. Among the time slots excluding these, it selects 8:00 AM, 7:00 PM, and 9:00 PM as the first candidate time slot group with high daily routine alignment scores. Subsequently, while collecting data through an external news API, the module detects that the macroscopic stress index has risen sharply due to a breaking news report related to an emergency disaster text message occurring at 7:00 PM. Additionally, based on the analysis of the user's past training patterns, it confirms that 9:00 PM is a time slot with high predicted cognitive fatigue, where the reaction speed is statistically slowest. Finally, the training adjustment module (150) filters out and removes 7:00 PM and 9:00 PM from the first candidate time slot group according to a negative factor exclusion rule, and finally determines the remaining 8:00 AM as the only optimal training time slot, sending a notification to the user recommending it.

[0150] The logic of the present invention ensures reliability in exceptional situations. "Cases where it is difficult to extract meaningful daily routines" are defined as cases where the standard deviation of the user's average sleep start and end times over the past four weeks exceeds two hours. When this condition is met, the system automatically reduces the influence of the daily routine compatibility score and instead switches to an auxiliary logic that selects the time slot where the user showed the highest performance in the past as the optimal candidate time slot. Additionally, if no recommended time slots remain as a result of the final filtering, the system sends a user-friendly message that does not force excessive training, such as "We recommend comfortable rest today."

[0151] The negative factor filtering threshold used in the time zone recommendation logic of the present invention is set through an example that analyzes the impact of each factor on training performance.

[0152] [Table 7]

[0153]

[0154] According to [Table 7] above, it was observed that the user's training performance (accuracy, reaction speed) significantly deteriorated during the time period when the macroscopic stress index exceeded 0.7. This suggests that performing training during that time period is inefficient. Therefore, the present invention establishes the validity of setting 0.7, which is the interval where a rapid decline in training effectiveness begins, as the filtering threshold of the macroscopic stress index.

[0155] In addition, to verify the effectiveness of the time zone recommendation system of the present invention, an example was performed to compare the performance when training at random time zones with when training according to the recommendation of the present invention.

[0156] [Table 8]

[0157]

[0158] Table 8 above shows that when training is performed at the optimal time according to the recommendations of the present invention, it not only improves the average accuracy per session but also has the effect of shortening the total time required to reach the same target difficulty level. Furthermore, it demonstrates that the voluntary training continuation rate is significantly improved as training is perceived as an efficient activity rather than a daily burden.

[0159] The present invention demonstrates a distinct difference in effect compared to the prior art. For example, unlike the prior art which recommends training simply because the user is free during the evening hours when major social issues occur, the present invention prevents inefficient training caused by decreased concentration by detecting a high macroscopic stress index and excluding that time period from the recommendation list. Furthermore, while the prior art may recommend training during late-night hours when concentration drops due to the user's biological rhythm, the present invention filters out that time period based on high predicted cognitive fatigue, thereby ensuring the user's rest and maintaining their condition for the following day.

[0160] The judgment rules adopted in this invention are based on objectivity and necessity. The use of daily routine conformity scores is based on the scientific fact that human cognitive abilities peak at specific times of the day according to the 'Circadian Rhythm' theory. In addition, the consideration of macroscopic stress indices reflects the principle that emotional factors such as anxiety or worry impede the efficiency of working memory.

[0161] The adoption of the aforementioned components and judgment rules has a technical effect of maximizing training outcomes by inducing users to train during the time when they can utilize their cognitive resources most efficiently. Furthermore, through a recommendation method that respects both individual lifestyle patterns and social contexts, it supports training in becoming a positive habit rather than a daily burden, which contributes to increasing long-term training adherence.

[0162] Terms and thresholds used in this specification are defined according to clear criteria. A 'high' daily routine compliance score refers to a time period corresponding to the top 25% of the score distribution calculated over the user's 24 hours. A 'high' macroscopic stress index refers to a case where an index normalized to a value between 0 and 1 exceeds a preset threshold (e.g., 0.7). All thresholds mentioned in this specification are not fixed constants and may be periodically updated according to the statistical distribution of the entire user data or system operation policies.

[0163] To specifically explain how the system of the present invention actually operates, the process of each component and four core processes interacting organically is described in detail step by step through a scenario in which a 68-year-old user diagnosed with mild cognitive impairment uses the platform of the present invention. The user wears a smartwatch to track daily activities, which is linked to a user terminal (200).

[0164] First, when the user first joins the platform, they enter their age and diagnosis name through the screen of the user terminal (200), and the user information acquisition module (110) receives this information and stores it in the database (300). The system requests access rights to the smartwatch and calendar, and if the user agrees, data linkage is completed. The next morning, the decision-making and scheduling module initiates a procedure to recommend the optimal training time slot for the user by running Process 4. The module calculates a daily routine compatibility score from the linked calendar and smartwatch, calculates a macroscopic stress index through an external news API, and calculates a predicted cognitive fatigue based on the user's general biological rhythm. The module applies the two-stage filtering logic of Process 4 to select a time slot group with a high daily routine compatibility score, and then determines 10:00 AM, which has a low macroscopic stress index and predicted cognitive fatigue, as the final recommended time. Based on the determined information, the training content provision module (120) sends a push notification to the user terminal (200) suggesting the start of training.

[0165] Subsequently, when the user starts training via a notification at 10:00 AM, the decision-making and scheduling module simultaneously executes processes 1 and 2 to determine the training content and difficulty level. According to process 2, the performance data collection module (130) analyzes the user's facial expression through the camera of the user terminal (200), and the data processing and indicator calculation module determines this as an emotional variability factor in a 'stable' state and analyzes past training history to confirm that the user's novelty adaptation factor is 'high'. Based on these two indicators, the decision-making and scheduling module determines 'novel visuospatial reasoning training', which has never been performed before, as the final content type among the 'challenging' content group suitable for a 'stable' state, in accordance with the tendency to prefer novelty. At the same time, according to process 1, since the user's previous training performance is good, the data processing and indicator calculation module sets the basic difficulty level to 'Level 5', and the performance data collection module (130) collects data on a quiet indoor environment, stable heart rate variability, and positive failure recovery rate. When the data processing and indicator calculation module calculates these indicators, the decision-making and scheduling module applies the 3-step correction rule of process 1 to increase the difficulty level by '+1' according to the positive failure recovery rate and determines the final difficulty level as 'Level 6'. The training content provision module (120) retrieves the determined 'new visual-spatial reasoning training, Level 6' content from the database (300) and displays it on the screen of the user terminal (200).

[0166] Next, while the user is performing 'Level 6' training, an emergency situation occurs in which the user feels sudden dizziness and the heart rate rapidly increases. The performance data collection module (130) detects this abnormal bio-signal in real time through the smartwatch, and the data processing and indicator calculation module analyzes it to calculate a 'dangerous' level bio-state risk index that deviates from the individual's stable baseline by more than 4 standard deviations. As soon as the decision and scheduling module receives the 'dangerous' level indicator, it activates the priority emergency response matrix of Process 3, ignoring all other factors such as the importance of the current training or the user's resilience, and simultaneously sends a command to the training content provision module (120) to immediately forcibly stop the training and display a safety message on the user terminal (200) screen, and a command to send an emergency text message to a pre-registered guardian.

[0167] Finally, if the user recovers their condition the next day and attempts training again at the cafe in the afternoon, the decision-making and scheduling module performs adaptive readjustment by considering a different context than before. Reflecting yesterday's emergency event log, the system, according to Process 2, decides to recommend familiar and stable 'memory recall training' as recommended content rather than presenting an unreasonable challenge. At the same time, the performance data collection module (130) detects noise in the cafe, i.e., a high environmental cognitive disturbance index, and due to yesterday's event, the system temporarily increases the sensitivity of the residual cognitive fatigue. According to the rule of Process 1, '-1' is deducted from the basic difficulty level 'Level 5' due to the high environmental disturbance index, and the final difficulty level is lowered to 'Level 4'. The training content provision module (120) provides 'memory recall training, Level 4' to the user according to the above decision, thereby demonstrating perfect adaptation to the user's changed state and environment. Through this series of processes, each component of the present invention organically executes four core processes and responds in real time to changes in the user's microscopic and macroscopic state to realize a hyper-personalized adaptive cognitive health management service.

[0168] Although preferred embodiments of the present invention have been illustrated and described above, the present invention is not limited to the specific embodiments described above. Various modifications are possible by those skilled in the art without departing from the essence of the invention as claimed in the claims, and such modifications should not be understood individually from the technical spirit or perspective of the present invention. Explanation of the symbols

[0170] Platform server (100) User information acquisition module (110) Training content provision module (120) Execution data collection module (130) Cognitive state analysis module (140) Training adjustment module (150) User terminal (200) Database (300)

Claims

Claim 1 A method for controlling cognitive training based on user state indicators comprises: a step in which a performance data collection module receives basic user information and a plurality of heterogeneous data generated during the performance of cognitive training from a user terminal; a step in which a data processing and indicator calculation module analyzes the received plurality of heterogeneous data to calculate a plurality of quantitative indicators representing the user's cognitive state, psychological state, physiological state, and surrounding environment state; a step in which a decision-making and scheduling module determines the timing of the provision of training content to be provided to the user by applying a plurality of predefined condition rules based on the plurality of quantitative indicators; a first filtering step in which the decision-making and scheduling module primarily selects a first candidate content group that is pre-matched with the emotional state from the entire list of training content based on an emotional variability factor representing the user's current emotional state and the range of change calculated by the data processing and indicator calculation module; a second sorting step in which the decision-making and scheduling module rearranges the group by assigning a higher priority to content that matches the user's tendencies within the selected first candidate content group based on a novelty adaptation factor representing the user's past adaptation speed to a new type of training calculated by the data processing and indicator calculation module; and the decision-making and A step in which a scheduling module determines the content with the highest priority in the aforementioned reordered group as the final recommended content type; a step of applying a search mode that operates according to a minimum frequency algorithm, wherein if camera access of the user terminal is blocked and facial expression recognition is impossible, voice analysis is used preferentially, and if voice analysis is also impossible, the emotion variability factor is treated as a neutral state to omit emotion-based filtering in the recommendation process and perform only novelty-based sorting, and if the novelty adaptation factor cannot be determined due to insufficient accumulated data, content from the category performed least by the user among all training categories is recommended preferentially.A user state indicator-based cognitive training control method comprising: a step in which, regarding a basic difficulty level primarily determined based on the user's previous training performance, the decision-making and scheduling module assigns +1 level to the basic difficulty level if the failure recovery rate, which indicates the degree of performance recovery immediately after an incorrect answer occurs, satisfies a positive state; subtracts -1 level from the current difficulty level if only one of the conditions—the residual amount of cognitive fatigue indicating the user's accumulated fatigue or the environmental cognitive disturbance index indicating the degree of disturbance from the surrounding environment—satisfies a high state; and subtracts -2 levels from the current difficulty level by applying a weight if both the residual amount of cognitive fatigue and the environmental cognitive disturbance index satisfy a high state, wherein if the upward adjustment condition and the downward adjustment condition are simultaneously satisfied, the downward adjustment is applied preferentially to determine the final difficulty level; and a step in which the training content provision module retrieves training content corresponding to the final recommended content type and the final difficulty level from the database at the determined provision time and provides it to the user terminal. Claim 2 A user state indicator-based cognitive training control method according to claim 1, wherein the emotional variability factor is calculated by performing an expression recognition algorithm on a user's expression image obtained through the front camera of the user terminal and combining the ratio of negative emotional expressions and the frequency of changes in the emotional state; the novelty adaptation factor is calculated based on the gap between the average performance of a specific number of initial times when the user first encounters a new type of training content and the average performance at a point where performance stabilizes thereafter, and the number of training sessions taken until the point of stabilization; the failure recovery rate is calculated by comparing the average correct answer rate during a set number of times within a range of 2 to 5 times immediately after an incorrect answer occurs in the user's training log with the overall average correct answer rate of the corresponding training session; the residual cognitive fatigue amount is calculated by applying a predefined weight to the percentage difference between the real-time heart rate variability value and the individual stable state baseline and the percentage of the touchscreen input delay time delayed relative to the individual average; and the environmental cognitive disturbance index is calculated based on the ambient noise level measured through the microphone of the user terminal and the rapid change in light intensity detected through the illuminance sensor.

Images