Abstract: A method of learning a task model and a language model according to an embodiment of the present application includes: acquiring log data for each user for a content set consumed by a plurality of users and content included in the content set; acquiring text embedding for the content included in the content set through a pretrained language model (PLM); acquiring an output value related to a task from the text embedding and the log data through a task model; updating a weight of the task model based on the output value and a target value of the task; and acquiring back-propagation information for training the language model from the task model, and training the language model based on the back-propagation information.

Abstract: A method in which a server recommends a word to a user according to the present specification, includes receiving training data from a network and training an AI model by using the training data; inputting (1) a user vector and (2) a word vector to the AI model, and generating (1) a user embedding vector and (2) a word embedding vector for determining whether the user knows a word related to the word vector, on the basis of the trained AI model; inputting (1) the user embedding vector and (2) the word embedding vector to a function for determining whether the user knows a word related to the word vector; and outputting a result value for predicting whether the user knows a word related to the word vector from the function.

Abstract: The present disclosure relates to a system capable of increasing the accuracy of a prediction result when predicting a user drop-out rate or predicting a correct answer probability of a user in an online learning environment, and further relates to an operation method of the system. A system according to the present disclosure may include a plurality of encoder neural networks and a plurality of decoders, and may input question information to a kth encoder neural network and input response information to kth decoder neural network to learn the system, thereby predicting the user drop-out rate information and the users correct answer probability information with higher accuracy on the basis of the learned system.

Abstract: Provided is an apparatus for training a knowledge tracking model, which is an apparatus for predicting a correct answer probability of a user on the basis of data augmentation, the apparatus including: a problem-solving data storage unit configured to store problem-solving data in which a problem solved by a user and a response of the user to the problem are mapped; a data augmentation performing unit configured to receive the problem-solving data from the problem-solving storage unit and convert the problem-solving data to generate augmented data; a regularization performing unit configured to receive the augmented data from the data augmentation performing unit and perform a regularization operation using a regularization loss function determined on the basis of a data augmentation method that is performed; and a model training unit configured to input the augmented data to a knowledge tracking model, allow the knowledge tracking model to learn a weight representing a relationship between a problem-solving

Abstract: A method of tracking a dialogue state according to an embodiment of the present application includes: acquiring a trained dialogue state tracking model; acquiring target dialogue data; acquiring dialogue summary data from the target dialogue data using the dialogue state tracking model; and generating a dialogue state template from the dialogue summary data, in which the dialogue state tracking model includes an input layer for receiving the target dialogue data, an output layer for outputting the dialogue summary data, and a hidden layer having a plurality of nodes connecting the input layer and the output layer, and is trained using a training set that includes dialogue data and a dialogue summary sentence generated from dialogue state data related to the dialogue data.

Abstract: According to an embodiment of the present disclosure, a pre-training modeling system for predicting an educational element using a pre-trained AI model includes: a pre-training unit configured to pre-train an AI model, through an evaluation element that can be automatically collected every time a user solves an individual question by receiving an interaction element according to an offline user's operation from a user terminal, classifying an AI element of educational content into the interaction element and the educational element, and intersecting the interaction element and the educational element; a fine-tuning unit configured to perform an additional training operation with respect to the pre-trained AI model, according to an educational element to be predicted; and an educational element prediction unit configured to predict, on the basis of the fine-tuned AI model, one or more pieces of information among test scores, grades, qualifications, and review information indicating whether the user got a corre

Abstract: Provided are a method of recommending preceding educational content and an educational content recommendation device for recommending preceding educational content. The method includes acquiring a language model of which training has been completed, updating the language model by tuning the language model to acquire a target language model, and determining preceding educational content through the target language model. The acquirement of the target language model includes acquiring an educational data set including first clustering data and second clustering data and updating the language model to predict a probability that the first clustering data is a next token of the second clustering data.

Abstract: A method of training a sequence encoder according to an embodiment of the present application includes: acquiring interaction data for each user; extracting an embedding vector from log data of the interaction data for each user; acquiring a reference vector from the embedding vector; acquiring a hidden vector from the reference vector through the sequence encoder; acquiring a projection embedding vector from the hidden vector; and updating a parameter of the sequence encoder using a contrastive learning technique.

Abstract: According to an embodiment of the present disclosure, a pre-training modeling system for predicting an educational element using a pre-trained AI model includes: a pre-training unit configured to pre-train an AI model, through an evaluation element that can be automatically collected every time a user solves an individual question by receiving an interaction element according to an offline user's operation from a user terminal, classifying an AI element of educational content into the interaction element and the educational element, and intersecting the interaction element and the educational element; a fine-tuning unit configured to perform an additional training operation with respect to the pre-trained AI model, according to an educational element to be predicted; and an educational element prediction unit configured to predict, on the basis of the fine-tuned AI model, one or more pieces of information among test scores, grades, qualifications, and review information indicating whether the user got a corre

Abstract: A method for pre-training artificial intelligence models to predict a score of a user by a server, comprises: generating a first sequence for training a first model, wherein the first sequence includes a masked element related to an exercise for predicting the score of the user; inputting the first sequence to the first model to train the first model; and inputting a second sequence to a second model predicted by the first model on the basis of the first sequence to train the second model, wherein the second model is trained through comparison between the first sequence and a third sequence predicted through the second model on the basis of the second sequence.

Abstract: Provided are a device and method for recommending educational content. The method includes acquiring a user's search information, acquiring a candidate webpage set on the basis of the search information, classifying candidate webpages included in the candidate webpage set into a first webpage group and a second webpage group, determining a target webpage on the basis of classification results, and transmitting the determined target webpage.

Abstract: A method of tracking a dialogue state according to an embodiment of the present application includes: acquiring a trained dialogue state tracking model; acquiring target dialogue data; acquiring dialogue summary data from the target dialogue data using the dialogue state tracking model; and generating a dialogue state template from the dialogue summary data, in which the dialogue state tracking model includes an input layer for receiving the target dialogue data, an output layer for outputting the dialogue summary data, and a hidden layer having a plurality of nodes connecting the input layer and the output layer, and is trained using a training set that includes dialogue data and a dialogue summary sentence generated from dialogue state data related to the dialogue data.

Abstract: A method of learning a task model and a language model according to an embodiment of the present application includes: acquiring log data for each user for a content set consumed by a plurality of users and content included in the content set; acquiring text embedding for the content included in the content set through a pretrained language model (PLM); acquiring an output value related to a task from the text embedding and the log data through a task model; updating a weight of the task model based on the output value and a target value of the task; and acquiring back-propagation information for training the language model from the task model, and training the language model based on the back-propagation information.

Abstract: Provided is a method of training a neural network model for calculating an uncertainty index, the method including: obtaining a reference answering data set of a plurality of reference users, calculating expected score information of the reference user from the reference answering data set; obtaining actual score information of the reference user; obtaining a training set on the basis of the reference answering data set, the expected score information, and the actual score information, the training set including label information that is defined as a difference between the expected score information and the actual score information; and training a first neural network model for calculating an uncertainty index related to accuracy of the expected score information of the reference user from the reference answering data set using the training set.

Abstract: Provided are a method of recommending preceding educational content and an educational content recommendation device for recommending preceding educational content. The method includes acquiring a language model of which training has been completed, updating the language model by tuning the language model to acquire a target language model, and determining preceding educational content through the target language model. The acquirement of the target language model includes acquiring an educational data set including first clustering data and second clustering data and updating the language model to predict a probability that the first clustering data is a next token of the second clustering data.

Abstract: The present disclosure relates to a method of recommending a diagnostic test question for user evaluation by an electronic device, including: generating a first matrix indicating whether users answer all questions correctly; generating a second matrix based on the first matrix using knowledge tracing (KT); and selecting the diagnostic test question using Lasso regression based on the second matrix.

Abstract: The present disclosure relates to a method of reducing a size of an artificial intelligence model by an electronic device, including: inputting an input value for training to a first model; training the first model for performing a specific task based on the input value; inputting the input value to a second model; and training the second model based on an output value of the first model, in which the first model may be an artificial intelligence model larger in size than the second model.

Abstract: The present disclosure relates to a method of predicting a user's score on a question by an electronic device. The method includes: training a DP-multi tasking learning (DP-MTL) model; verifying the DP-MTL model; receiving choice selection information related to the question from the user through the terminal, and predicting 1) a probability that the user answers the question correctly and 2) the user's score related to the question using the verified DP-MTL model based on the choice selection information, and the DP-MTL model may be a model for predicting the user's score based on 1) information on whether the user answers the question correctly, 2) information on which incorrect answer is selected among choices of the question when the user selects an incorrect answer, and 3) a skill level of the user.

Abstract: In the present specification, a method of performing pre-training using an automated machine learning (AutoML) model by a server includes: using a first model for performing a first task to generate a second model for a second task; inputting, to the Auto ML model, a preset feature based on 1) components of the first model and the second model and 2) an element obtainable from training of the first model and generating of the second model as a state value; and changing the first model using the AutoML model.

Abstract: Provided are a device and method for assessing a user's learning ability. The method includes acquiring target assessment data of a target user and a reference user related to a target domain, wherein the target assessment data includes question data related to the target domain and answer data of each of the target user and the reference user for the question data, acquiring a target neural network model of which training has been completed, acquiring comparison information representing the target user's ability in relation to the reference user's ability in the target domain through the target neural network model, and calculating the target user's virtual score in the target domain on the basis of the comparison information.