Customized emotional disorder improving content providing system using knowledge tracing and reinforcement learning

Abstract

The present invention relates to a customized emotional disorder improving content providing system using knowledge tracing and reinforcement learning, capable of providing a therapeutic agent more suitable for a user as the user utilizes the system more, by using a knowledge tracing algorithm.

Description

A system that provides customized emotional disorder improvement content using knowledge tracking and reinforcement learning.

[0001] Research related to this patent was supported by the Development of Digital Treatment Device Technology for Depression in the Corona Blue Era (Project No.: 20014967) under the supervision of the Korea Evaluation Institute of Industrial Technology with funding from the Ministry of Trade, Industry and Energy, and by the Development of Brain-Body Interface Technology Using AI-Based Multi-Sensing (Project No.: S1601201041, Project Unique Number: 1711120116) under the supervision of the National IT Industry Promotion Agency with funding from the Ministry of Science and ICT.

[0002] The present invention relates to a system for providing customized emotional disorder improvement content using knowledge tracking and reinforcement learning, and more specifically, to a system capable of diagnosing emotional disorders and providing improvement content using various user data that can be collected from biometric information collected by sensors, self-questionnaires, and other user terminals.

[0003] In the case of physical illness, it's often necessary to collect, analyze, and assess the necessary data for a single condition. However, mental illnesses are often difficult to categorize under a single diagnosis, and many cases involve comorbid conditions. Therefore, treating mental illnesses like depression, anxiety, stress, and emotional disorders requires assessing complex illnesses.

[0004] In this regard, the diagnosis of a disease was previously determined through a questionnaire by a psychiatrist, but recently, technologies are being developed to determine the user's symptoms using artificial intelligence.

[0005] As a technology for diagnosing diseases using conventional artificial intelligence, Republic of Korea Publication Patent No. 10-2023-0011550 is disclosed.

[0006] However, these conventional technologies fail to provide efficient analysis algorithms tailored to the data being analyzed. Furthermore, providing personalized treatment requires algorithms that can clearly analyze the user's condition and provide effective treatment methods. However, these techniques are only described as utilizing "artificial intelligence." To accurately reflect the diverse data characteristics obtained from users and achieve the desired results, specific algorithms tailored to the specific needs are necessary.

[0007] Furthermore, traditionally, customized treatment for mental illness has been limited to counseling-based treatments based on cognitive behavioral therapy (CBT). While CBT is recognized as an effective non-pharmacological treatment, it can sometimes experience low compliance or require concurrent treatment. Therefore, a variety of treatment modalities, such as relaxation therapy and behavioral activation, along with CBT-based conversational therapy, are needed. Accurate diagnosis of the patient's condition and provision of tailored treatments are essential.

[0008] The purpose of the present invention is to provide a system for providing customized emotional disorder improvement content using knowledge tracking and reinforcement learning to solve the conventional problems described above.

[0009] As a means of solving the above problem, a system for providing customized emotional disorder improvement content using knowledge tracking and reinforcement learning can be provided, including a user data processing unit that receives user data including biometric information, environmental information, and in-app information from a terminal, a user status determination unit that analyzes the user data using an artificial intelligence model including a knowledge tracking and reinforcement learning algorithm in the user data processing unit and outputs a user status value, a content determination unit that selects one of a plurality of contents based on the user status value, and a content provision unit that loads content information from a database so that the selected content can be transmitted to the terminal.

[0010] Meanwhile, the user state determination unit may include a bidirectional long short-term memory (Bi-LSTM) model.

[0011] Additionally, in-app information can be updated each time a user performs a piece of content.

[0012] Meanwhile, user data may include biometric information and self-questionnaire questions received by the terminal input by the user, and in-app information may include content performance data obtained when the user performs content with the terminal.

[0013] Meanwhile, the user data processing unit may be configured to divide biometric information, environmental information, and in-app information into predetermined periods, perform noise removal and supplementation of missing data for user data, and perform grouping and vectorization of biometric information, environmental information, and in-app information.

[0014] Meanwhile, the user data processing unit can delete personal information from user data and integrate biometric information, environmental information, and in-app information into a single dimension and input it into the user status determination unit.

[0015] In addition, the user status determination unit may be improved to output the user status value based on the first data consisting of the user's age, gender, and weather information as the first learning, to output the user status value based on the second data consisting of the first data and biometric information as the second learning, and to output the user status value based on the third data consisting of the second data and in-app information as the third learning.

[0016] Meanwhile, biometric information can be transmitted from a wearable device worn by the user to a user data processing unit.

[0017] Meanwhile, biometric information may include the user's heart rate, heart rate variability, and activity level.

[0018] Furthermore, the database includes multiple treatment contents, and the multiple treatment contents may have at least one purpose among cognitive behavioral therapy, behavioral therapy, meditation based on relaxation effect, deep breathing training based on relaxation effect, concentration enhancement, and daily conversation training.

[0019] The system for providing customized emotional disorder improvement content using knowledge tracking and reinforcement learning according to the present invention can be used efficiently and at a low cost anytime and anywhere to treat emotional disorders that are difficult to access due to social gaze, cost, and distance.

[0020] Additionally, for emotional disorders that are difficult to classify as a single specific disease and have various comorbidities, the type and severity of the disease can be identified through analysis of various data collected from users, and treatment methods tailored to each group can be provided.

[0021] Furthermore, by utilizing a knowledge-tracking algorithm, the more a user utilizes the system, the more personalized treatments can be provided. This model can learn about individual users and provide treatments that maximize the effectiveness of treatment for their actual symptoms, rather than simply providing treatment content that the user will enjoy.

[0022] Figure 1 is a block diagram showing the configuration of a system for providing customized emotional disorder improvement content using knowledge tracking and reinforcement learning according to the present disclosure.

[0023] FIG. 2 is a diagram illustrating a data processing concept of an artificial intelligence unit in a system for providing customized emotional disorder improvement content using knowledge tracking and reinforcement learning according to the present disclosure.

[0024] Figure 3 is a flowchart illustrating a data processing process in a user data processing unit in the present disclosure.

[0025] Figure 4 is a flowchart illustrating a process for determining content based on user data in the present disclosure.

[0026] Figure 5 is a conceptual diagram illustrating the process from collecting user data to providing content for improvement according to the present disclosure.

[0027] Hereinafter, a system for providing customized emotional disorder improvement content using knowledge tracking and reinforcement learning according to an embodiment of the present invention will be described in detail with reference to the attached drawings. In the following description of the embodiments, the names of each component may be referred to by different names in the art. However, if they are functionally similar and identical, even if a modified embodiment is adopted, it can be considered an equivalent configuration. Furthermore, the symbols assigned to each component are described for convenience of explanation. However, the content depicted in the drawings in which these symbols are described does not limit each component to the scope of the drawings. Similarly, even if an embodiment with some modifications to the configurations depicted in the drawings is adopted, it can be considered an equivalent configuration if functional similarity and identity are present. Furthermore, if a component is recognized as something that should be included based on the general level of a person skilled in the art, its description will be omitted.

[0028] Figure 1 is a block diagram showing the configuration of a system for providing customized emotional disorder improvement content using knowledge tracking and reinforcement learning according to the present disclosure.

[0029] Referring to FIG. 1, a system (1) for providing customized emotional disorder improvement content using knowledge tracking and reinforcement learning according to the present disclosure may be configured to include a user terminal (10) and a server (20).

[0030] Here, the user terminal (10) may refer to a device that can provide simulated content to the user and receive at least one piece of information from the user. The user terminal (10) may be a device that the user selectively uses, such as a PC, laptop, tablet, mobile terminal (10), mobile phone, or smartphone. In addition, the user terminal (10) may be a wearable device, such as a smart watch, smart glasses, HMD (Head Mounted Display), or smart ring.

[0031] The terminal (10) can generate user data based on images, and can also generate user data by receiving voice signals and motion signals. The terminal (10) can include a camera, and the camera can track the user's gaze or analyze the user's movements by analyzing images. Voice receiving devices such as microphones can receive information such as the user's speech or voice, and information such as the level of ambient noise. In addition, the user terminal (10) can include sensors for directly receiving dynamic information, such as a gyro sensor or an acceleration sensor. Accordingly, the terminal (10) can receive information related to movement when the user holds and operates the handpiece. In addition, the terminal (10) can receive information related to body movement and information about head movement while the user is wearing the terminal (10).

[0032] Meanwhile, the user terminal (10) can receive and display various content information stored in a database, as described below, to the user. For example, the user terminal (10) can implement a simulation and generate signals for conveying to human senses such as vision, hearing, and touch.

[0033] The server (20) may include a user data processing unit (21), a user status determination unit (22), a customized content determination unit (23), a database, and a communication unit.

[0034] The communication unit can transmit and receive data by communicating with the user terminal (10) via wired / wireless communication. For example, the communication unit can receive user data from the user terminal (10). Additionally, the communication unit can transmit content data to the user terminal (10).

[0035] The user data processing unit (21) is configured to perform preprocessing of received user data for input into the user status determination unit (22), which will be described later. For example, the data processing unit (21) can delete personal information from user data and integrate complex data into a single dimension through a concatenation process. Furthermore, it can perform noise filtering on the user data and fill in data for uncollected portions using interpolation. This will be described in detail with reference to FIG. 3.

[0036] The user status determination unit (22) is configured to include an artificial intelligence model and is configured to receive preprocessed user data and output a classification value for the user's status. The user status determination unit (22) may be configured to include a knowledge tracking and reinforcement learning model. Accordingly, the user status determination unit (22) continuously tracks the user's status based on user data collected as the user performs simulation content over time, and also performs reinforcement learning so as to output a classification value appropriate for the current status.

[0037] The personalized content determination unit (23) is configured to determine content based on the values ​​output from the user status determination unit (22). The personalized content determination unit (23) can determine the type and difficulty of content intended to improve the user's status. In this case, content predicted to improve the user's symptoms may be selected.

[0038] The customized content provision unit (24) may be configured to receive content-related information from a database (25) to the terminal (10). The customized content provision unit (24) transmits content data to the terminal (10) based on information determined by the customized content determination unit (24).

[0039] Simulation information can be stored in the database (25). This simulation information can be stored in different formats depending on the type of user terminal (10). Specifically, the simulation information can be created with different designs, taking into account the interactive elements of a user using a smartphone versus smart glasses. Examples of simulations include interactive content that asks the user questions, provides content that creates a specific situation, or presents specific requirements.

[0040] Simulations can be created differently based on various environments and requirements. Furthermore, the difficulty level can be adjusted based on the user's current status during simulation creation.

[0041] The database (25) stores user data received from the communication unit (26), and may also store data related to content. Since this database may be configured using a widely used storage device, a detailed description of its configuration will be omitted.

[0042] However, in the server (20) described above, the data processing unit (21), user status determination unit (22), customized content determination unit (23), customized content provision unit (24), and database (25) are functionally distinct and can be implemented by a widely distributed computing device, or can be implemented in an integrated manner by a computing processing device developed exclusively for artificial intelligence.

[0043] Hereinafter, the processing of data in the present disclosure will be described with reference to FIG. 2.

[0044] FIG. 2 is a diagram illustrating a data processing concept of an artificial intelligence unit in a system for providing customized emotional disorder improvement content using knowledge tracking and reinforcement learning according to the present disclosure.

[0045] Referring to FIG. 2, the system for providing customized emotional disorder improvement content using knowledge tracking and reinforcement learning according to the present disclosure can input user data received from a user terminal into a user status determination unit (22). The customized content determination unit (23) can determine content for improving the user status based on values ​​related to the user status output from the user status determination unit (22).

[0046] User data received from a user terminal may be stored in a database (25). The user data may include personal information, biometric information, activity information, conversation information, environmental information, and in-app information.

[0047] Personal information, activity information, conversation information, environmental information, and in-app information can be collected through applications that execute content on the user's terminal. Meanwhile, biometric information can be received from wearable devices that include sensors capable of collecting the user's physical information. A wearable device, for example, may be a bioband.

[0048] Personal information can be determined through pre-determined questions. Personal information can include information about personality traits, gender, and age.

[0049] Biometric information may include at least one of the user's blood pressure (AMBP, Ambulatory Blood Pressure Monitoring), pulse, and electrocardiogram (ECG, Electrocardiogram).

[0050] Activity information may include information about the user's activity level, sleep time, and meals.

[0051] Conversation information may include the user's voice and text information.

[0052] Although not urban, environmental information may include weather-related information such as temperature, precipitation, and fine dust concentration.

[0053] In-app information refers to data obtained from users while engaging with content. This information includes content preference, content satisfaction, content frequency, content score, changes in mental health indices (PHQ-9, GAD-7, BADS-9, RSES-10, WHO-5), and scores for mitigation effects. Changes in mental health indices can be included in the content engaged by users as questions and determined by user input.

[0054] The Patient Health Questionnaire-9 (PHQ-9) is a questionnaire used to assess the presence and severity of depression. It consists of nine questions, each reflecting a key symptom of depression. Respondents are asked to rate how frequently each symptom occurred over the past two weeks.

[0055] The General Anxiety Disorder-7 (GAD-7) is used to diagnose and assess the severity of generalized anxiety disorder (GAD). It consists of seven items and assesses various aspects of anxiety. This assessment method provides a simple measure of anxiety levels and is useful for identifying potential anxiety disorders.

[0056] The BADS-9 (Behavioral Activation for Depression Scale) is a tool based on behavioral activation theory to assess a patient's level of activation and treatment response. Consisting of nine items, respondents assess their activities over the past week. The BADS is primarily used to assess the behaviors patients with depression exhibit to overcome their depression.

[0057] The Rosenberg Self-Esteem Scale (RSES-10) is a tool used to measure an individual's self-esteem. It consists of 10 items, and respondents can respond to questions that reflect their attitudes and self-evaluations toward themselves. The RSES is widely used to measure self-esteem in psychological research and clinical assessments.

[0058] The World Health Organization-Five Well-Being Index (WHO-5) is a simple questionnaire designed to assess overall psychological well-being and quality of life. Consisting of five items, it measures the frequency of positive experiences related to mood and activity over the past two weeks. The WHO-5 can be used across a wide range of age groups and cultures, and can also be used as a screening tool for depression.

[0059] Once user data is acquired, the user data processing unit (21) performs preprocessing of the user data. Data preprocessing is essential for improving data quality and optimizing model performance.

[0060] The user data processing unit (21) performs data integration, data refinement, and data transformation processes.

[0061] Data integration stores received data in a database and classifies the data according to the mission (included in the content).

[0062] Data refinement includes the processes of outlier deletion and missing value deletion.

[0063] Data transformation involves converting data into a format that can be processed by AI models, including text embedding and image embedding. Text embedding is the process of converting words or sentences into fixed-size vectors. Image embedding refers to the process of mapping images from a high-dimensional space to a low-dimensional vector space. Image embedding can be performed by a convolutional neural network (CNN), allowing data characteristics to be standardized.

[0064] Preprocessed user data is input to the user state determination unit (22). The user state determination unit (22) may include a neural network model. The user state determination unit (22) may include a knowledge tracking model. The neural network used here may include a Bidirectional Long Short-Term Memory (Bi-LSTM) model. In addition to the advantages of LSTM, BiLSTM is designed to richer understand the context of the entire sequence by processing the sequence in both directions (from past to future and from future to past). Specifically, Bi-LSTM uses two separate LSTM layers, one for forward processing the sequence and the other for backward processing. The outputs of these two layers are combined at a specific point in time to provide information reflecting both the bidirectional context at that point in time. Therefore, the user state determination unit (22) can understand the factors that have influenced the user in the long term and the status that has changed accordingly.

[0065] Knowledge Tracing (KT) is a technology primarily used in education and learning. It is an algorithm that tracks and predicts a learner's knowledge status. The goal of this model is to determine whether a learner has acquired specific knowledge or skills and how they change over time during their learning process. Using KT allows for a better understanding of a user's current status and the improvements made after performing specific content, enabling the selection of customized content based on the user's changed status. For example, the user status determination unit, Deep Knowledge Tracing (DKT), may include a recurrent neural network (RNN) capable of processing sequential learning data.

[0066] Additionally, the user status determination unit (22) can be subjected to reinforcement learning.

[0067] At this point, model improvement will proceed in at least three stages. The first stage of AI model improvement will be based on analysis of basic user data. The second stage will utilize additional user biometric data. The third stage will utilize additional in-app data collected when users use application content.

[0068] In reinforcement learning, the primary data used is user information (age, gender): basic demographic information is used to identify general trends. Weather information is also considered, along with external environmental factors that may influence the user's mood or activity level. At this stage, the user's status is predicted based on relatively simple information. This process helps the user status determination unit identify basic patterns and trends.

[0069] Utilizing secondary data adds biometric information, such as heart rate, sleep patterns, and activity levels, to more precisely track a user's physical condition and its changes. This step adds biometric information to primary data to better understand the connection between a user's physical and mental state, enabling more precise predictions.

[0070] Tertiary data utilization involves leveraging in-app data obtained from user activities within the app (e.g., playing games, completing missions). This allows for a more detailed understanding of user preferences, behavioral patterns, cognitive and emotional responses. In addition to primary and secondary data, this step leverages information gleaned from users' in-app behavior to maximize understanding of their mental state. This can enable highly personalized and precise predictions.

[0071] The user status determination unit (22) also includes a fully connected layer, and the output values ​​are input to the customized content determination unit. The customized content determination unit (23) matches customized content to improve the user's status according to the user's status classification value based on user data. Meanwhile, the functions of this user status determination unit will be described in detail later with reference to FIG. 4.

[0072] Afterwards, the customized content provision unit (23) can load the customized improved content determined by the customized content determination unit and transmit it to the user terminal.

[0073] As described above, the user status determination unit (22) according to the present disclosure can provide customized and improved content that better suits the user the more it performs content stored in the system through knowledge tracking and reinforcement learning. Therefore, rather than simply providing content that the user will enjoy, it can independently learn about individual users and provide treatment to maximize the improvement effect on actual symptoms.

[0074] Figure 3 is a flowchart illustrating a data processing process in a user data processing unit in the present disclosure.

[0075] Referring to Figure 3, the preprocessing process in the user data processing unit is described in chronological order as follows.

[0076] First, a data collection step (S11) necessary for analysis can be performed from the user terminal, such as biosignals, self-questionnaires, and content performance data.

[0077] A step (S12) of performing data segmentation by setting an analysis period may then be performed. That is, the acquired data may be segmented into multiple sections based on a predetermined period or the time at which a specific event occurred.

[0078] Afterwards, a data verification and preprocessing step (S13) can be performed. Noise removal and missing data supplementation can be performed in this step.

[0079] Afterwards, a data grouping and vectorization step (S14) can be performed to input the user data analysis unit.

[0080] Figure 4 is a flowchart illustrating a process for determining content based on user data in the present disclosure.

[0081] Referring to FIG. 4, the user status determination unit may perform a step (S21) of receiving grouped and vectorized data and inputting it into a model composed of knowledge tracking and reinforcement learning, a step (S22) of analyzing user data using the model composed of knowledge tracking and reinforcement learning, a step (S23) of determining optimal improvement content based on the analysis results among improvement content, and a step (S24) of transmitting the selected improvement content to a mobile application and providing it to the user.

[0082] The step (S21) of receiving grouped and vectorized data and inputting it into a model composed of knowledge tracking and reinforcement learning corresponds to the step of inputting preprocessed data from the user data processing unit into the user status determination unit.

[0083] The user status determination unit can define state, action, and reward. Specifically, state represents the user's current status. Action refers to the content provided. Here, reward can be defined by evaluating whether the user's status has improved.

[0084] As described above, the user state determination unit includes a bidirectional LSTM model, which receives preprocessed user data, learns complex patterns of time series data, and selects optimal content for improvement by considering the user's previous state and current response.

[0085] In the user status determination unit, changes in the user's mental health index (based on the PHQ-9, GAD-7, BADS-9, RSES-10, WHO-5, etc.) can be defined as a reward for reinforcement learning. These rewards can be optimized through trial and error as the user uses the system. In other words, the system can be optimized to select content that improves different symptoms depending on the user's status.

[0086] The step (S22) of analyzing user data using a model comprised of knowledge tracking and reinforcement learning corresponds to the step where the user status determination unit continuously learns using the input user data and tracks the user's responses and improvement processes. By tracking user responses and improvement processes, content selection can be improved to optimize rewards (user improvement).

[0087] The step (S23) of determining the optimal content for improvement based on the analysis results among the content for improvement determines the content and difficulty based on the classification value output from the user status determination unit.

[0088] The step (S24) of delivering the selected content for improvement to the mobile application and providing it to the user corresponds to the step of loading the content for improvement from the database based on the determined content and difficulty and transmitting it to the user terminal.

[0089] Afterwards, the user terminal executes customized improvement content, and the user performs the customized improvement content accordingly. The customized improvement content executed on the terminal may be, for example, cognitive behavioral therapy, behavioral therapy, meditation based on relaxation effect, deep breathing training based on relaxation effect, concentration enhancement mission, and interactive AI chatbot for daily conversation training. As a specific example, the content may be determined as 'Deep Breathing Training Mission 3' as deep breathing training content based on relaxation effect. Furthermore, the difficulty of the content may be determined according to the patient's condition, such as difficulty level 'medium'.

[0090] Figure 5 is a conceptual diagram illustrating the process from collecting user data to providing content for improvement according to the present disclosure.

[0091] Referring to FIG. 5, the system according to the present disclosure may include a scenario modeling generation unit (2) and a virtual simulation performance model (3).

[0092] The scenario modeling generation unit (2) can generate agents that group the characteristics of subjects with emotional and behavioral disorders. It can also be used to collect virtual simulation data.

[0093] Here, the scenario modeling generation unit can generate virtual scenarios and agents by assuming voice, text, psychological, environmental, and social factors, aiming for medical categorization consistent with the DSM-5 classification.

[0094] As voice factors, we can define changes in the user's voice tone, inaccurate changes in intensity, lack of tone consistency, reduction in speed, and limited changes in pitch.

[0095] As character factors, we can define changes in sentence length by users, the balance of positive and negative expressions, emphasis on the connection between cause and effect, frequently used types of negative words, reduction in the number and variety of parts of speech, and restrictions on emotional expression.

[0096] Psychological factors can be defined as high stress levels, irregular sleep patterns, high depression levels, high heart rate and breathing rates, changes in eating patterns, and associations with diseases.

[0097] Environmental factors can be defined as high ambient noise, strong lighting and sensitivity, predictability and comfort of the environment, importance of natural environment, diversity of individual environmental preferences, and reduced adaptability to environmental changes.

[0098] Social factors can be defined as differences in age and gender, differences in national and cultural influences, factors related to occupation and stress, personal personality traits, and stress related to social status.

[0099] The virtual simulation execution model (3) repeats virtual simulations using agents generated in the scenario modeling generation unit (2) and collects data according to the virtual simulation execution.

[0100] Data generated from the virtual simulation execution model (3) are input to the user data processing unit (21), the user status determination unit (22), and the customized content determination unit (23) and can be tested.

[0101] Afterwards, the actual collected data from the user is input into the user data processing unit (21), and the user status determination unit (22) and the customized content determination unit (23) can determine improved content according to the user's status.

[0102] As mentioned above, when determining improvement content based on user status, an AI model can be applied that predicts the user's mission performance results and recommends the optimal mission by utilizing Medical State Tracing, a knowledge tracing technique that predicts future progress by looking at the user's performance record and applying it to the digital healthcare field.

[0103] To this end, additional models such as a bidirectional long short-term memory (Bi-LSTM) model suitable for extracting time series features of data and a convolutional neural network (CNN) for extracting unique features of each data are connected to convert the data in each area into an embedding format, and data with standardized characteristics can be used.

[0104] Additionally, this system automatically generates decision points that serve as the basis for grouping decisions by combining data embeddings based on various cases.

[0105] At this time, among the various combinations of decision points that produce virtual mission play data, a combination that is helpful for the final grouping is selected, and the data received from the virtual simulation execution model (3) is produced and reflected in the final grouping.

[0106] Ultimately, in this disclosure, the grouped emotional behavioral disorder scores utilizing the user agent are compared with the scores of users who have performed the mission, and the types and severity of emotional behavioral disorders are selected and diagnosed in real time, and customized improvement content is selected accordingly.

[0107] Afterwards, based on the user's evaluation results as medical state tracing, the performance results of the content are predicted (evaluation), and the content (Related Category a& Data) suitable for treatment is selected in a user-tailored manner.

[0108] Subsequently, the user status determination unit can be retrained based on additional data collected through reinforcement learning as the user continues to use the system (1). Therefore, as the user continues to use the system (1), the personalized content provision unit (24, recommender system) can provide more personalized and improved content.

Claims

1. A user data processing unit that receives user data including biometric information, environmental information, and in-app information from a terminal; A user status determination unit that analyzes the user data using an artificial intelligence model including a knowledge tracking and reinforcement learning algorithm in the user data processing unit and outputs the user status value; A content determination unit that selects one of multiple contents based on the status value of the user; and A system for providing customized emotional disorder improvement content using knowledge tracking and reinforcement learning, including a content providing unit that loads the content information from a database so that the selected content can be transmitted to the terminal.

2. In paragraph 1, The above user status determination unit, A system for providing customized emotional disorder improvement content using knowledge tracking and reinforcement learning, including a bidirectional long short-term memory (Bi-LSTM) model.

3. In paragraph 2, The above in-app information is a system that provides customized emotional disorder improvement content using knowledge tracking and reinforcement learning, which is updated each time the user performs a certain content.

4. In paragraph 3, The above user data is, Includes biometric information and self-questionnaire questions received by the terminal input by the user, The above in-app information is a system for providing customized emotional disorder improvement content using knowledge tracking and reinforcement learning, including content performance data obtained when the user performs the content using the terminal.

5. In paragraph 4, The above user data processing unit, The above biometric information, the above environmental information, and the above in-app information are divided into predetermined periods, Perform noise removal and missing data supplementation on the above user data, A system for providing customized emotional disorder improvement content using knowledge tracking and reinforcement learning, configured to group and vectorize the above biometric information, the above environmental information, and the above in-app information.

6. In paragraph 5, The above user data processing unit, A system for providing customized emotional disorder improvement content using knowledge tracking and reinforcement learning, which deletes personal information from the above user data and integrates the above biometric information, the above environmental information, and the above in-app information into a single dimension and inputs it into the above user status determination unit.

7. In paragraph 6, The above user status determination unit, As a first learning, it is improved to output the user's status value based on the first data consisting of the user's age, gender, and weather information. As a second learning, it is improved to output the user's status value based on the second data consisting of the first data and biometric information. A system for providing customized emotional disorder improvement content using knowledge tracking and reinforcement learning, which is improved to output the user's status value based on the tertiary data composed of the secondary data and in-app information as tertiary learning.

8. In paragraph 7, The above biometric information is, A system for providing customized emotional disorder improvement content using knowledge tracking and reinforcement learning transmitted from a wearable device worn by the user to the user data processing unit.

9. In paragraph 8, The above biometric information is, A system for providing customized emotional disorder improvement content using knowledge tracking and reinforcement learning, including the user's heart rate, heart rate variability, and activity level.

10. In paragraph 9, The above database contains multiple treatment contents, The above multiple treatment contents are a system for providing customized emotional disorder improvement contents using knowledge tracking and reinforcement learning, with at least one of the following purposes: cognitive behavioral therapy, behavioral therapy, relaxation effect-based meditation, relaxation effect-based deep breathing training, concentration enhancement, and daily conversation training.

Images