Digital health literacy-based content service device and method

Abstract

According to the present embodiments, provided are a digital health literacy-based content service device and method, wherein the device receives an input of a demographic factor, a health-related factor, and a digital information-related factor to predict a digital health literacy level, recommends user-customized health information on the basis of a prediction result, performs binary classification of a literacy level of a user by using a classification model and a recommendation model, and provides personalized health information on the basis of metadata and interaction data.

Description

Digital Health Literacy-Based Content Service Device and Method

[0001] Embodiments of the present disclosure relate to a digital health literacy-based content service device and method.

[0002] With the rapid aging of the global population, interest in senior health management is growing. In particular, older adults experience chronic conditions like high blood pressure and diabetes, as well as mental health issues like depression and sleep disorders. This leads to a growing demand for personalized health information to effectively manage these conditions. However, many older adults struggle to utilize digital health services due to their lack of digital technology proficiency. Consequently, a lack of digital health literacy negatively impacts access to, understanding of, and utilization of health information, further exacerbating health inequalities.

[0003] Reflecting this situation, efforts have been made to develop smartphone-based health information systems and personalized recommendation content. However, accessibility and usability remain limited for senior users with low digital literacy. In particular, a lack of data on new users has led to poor performance of personalized recommendation services, and recommendations are based solely on simple interaction data without specific disease classification, resulting in poor service quality.

[0004] Furthermore, most existing methods for assessing digital health literacy rely on self-reporting, making them vulnerable to social desirability bias and recall errors, potentially leading to over- or underestimation of assessment results. Furthermore, they are biased toward specific populations, making it difficult to accurately measure the digital health literacy of people from diverse cultural and social backgrounds. Furthermore, they fail to reflect the rapid advancement of digital health technologies.

[0005] Meanwhile, digital health technologies such as hybrid wearable sensors offer the advantage of continuously monitoring physiological changes in the elderly and detecting abnormal signs early, but the effective use of these technologies ultimately depends greatly on the user's level of digital health literacy.

[0006] Therefore, there is a growing need for new devices and methods that can accurately and objectively predict seniors' digital health literacy levels and effectively recommend personalized health information.

[0007] Embodiments of the present disclosure can provide digital health literacy-based content service technology.

[0008] In one aspect, the present embodiments can provide a digital health literacy-based content service device including an input unit that receives user information and interaction information from a user terminal, a user evaluation unit that evaluates the user's digital health literacy level using the user information received according to a first learning model that has been learned in advance, a first content provision unit that selects digital health literacy support content corresponding to the evaluation result of the user evaluation unit and provides the content to the user, a content extraction unit that extracts health information content suitable for the user using the user information and interaction information received according to a second learning model that has been learned in advance, and a second content provision unit that provides the extracted content to the user.

[0009] In another aspect, the present embodiments can provide a digital health literacy-based content service method including a receiving step of receiving user information and interaction information from a user terminal, an evaluation step of evaluating a user's digital health literacy level using the user's information received according to a first learning model learned in advance, a first content providing step of selecting digital health literacy support content corresponding to the evaluation result and providing the content to the user, an extraction step of extracting health information content to be recommended to the user using the user's information and interaction information received according to a second learning model learned in advance, and a second content providing step of providing the extracted health information content to the user.

[0010] In another aspect, the present embodiments may provide a digital health literacy-based content service device including an input unit for receiving metadata of any user, and a processing unit for recommending health information of a user for metadata of any user using a recommendation model learned to generate health information of the user for metadata of the user using learning data consisting of input data including metadata of the user and interaction data of the user, and output data including health information of the user corresponding to the metadata of the user and the interaction data of the user.

[0011] In another aspect, the present embodiments may provide a digital health literacy-based content service method including an input step for receiving metadata of any user, and a processing step for recommending health information of a user for metadata of any user using a recommendation model learned to generate health information of the user for metadata of the user using learning data consisting of input data including metadata of the user and interaction data of the user, and output data including health information of the user corresponding to the metadata of the user and the interaction data of the user.

[0012] In another aspect, the present embodiments may provide a digital health literacy-based content service device including an input unit that receives arbitrary demographic factors, health-related factors, and digital information-related factors as input values, a classification model that is trained using a learning data set that uses the demographic factors, health-related factors, and digital information-related factors as input data, and uses a binary-classified digital health literacy level as output data, and a processing unit that classifies a digital health literacy level from arbitrary demographic factors, health-related factors, and digital information-related factors.

[0013] In another aspect, the present embodiments may provide a digital health literacy-based content service method including a collection step of generating a learning data set using demographic factors, health-related factors, and digital information-related factors as input data and a binary-classified digital health literacy level as output data, a learning step of learning a classification model that predicts a digital health literacy level using the learning data set, an input step of inputting any demographic factors, health-related factors, and digital information-related factors as input values ​​to the classification model, and an output step of outputting a binary-classified digital health literacy level corresponding to the input value as a result value.

[0014] In another aspect, the present embodiments may provide a computer device including a memory storing any demographic factor, health-related factor and digital information-related factor, a classification model, the classification model being a model that performs machine learning by using any demographic factor, health-related factor and digital information-related factor as input data and a digital health literacy level according to any demographic factor, health-related factor and digital information-related factor as output data, and a processor that executes the classification model that performed machine learning stored in the memory and outputs a digital health literacy level from any demographic factor, health-related factor and digital information-related factor when a task of predicting a digital health literacy level is requested.

[0015] According to embodiments of the present disclosure, a digital health literacy-based content service technology can be provided.

[0016] FIG. 1 is a block diagram of a health information recommendation device according to one embodiment.

[0017] Figure 2 illustrates the learning dataset of the recommendation model of Figure 1.

[0018] Figure 3 illustrates the process of providing health information through a learned recommendation model.

[0019] Figure 4 illustrates the types of metadata and interaction data of Figure 1.

[0020] Figure 5 illustrates metadata including demographic information and disease and lifestyle information among the metadata of Figure 4.

[0021] Figure 6 illustrates health information content according to one embodiment.

[0022] Figure 7 is a block diagram of the recommended model of Figure 1.

[0023] Figure 8 illustrates an example of a detailed internal configuration participating in the learning process of the recommendation model of Figure 7.

[0024] Figure 9 illustrates the relationship between variables of each component during the learning process of the recommendation model of Figure 8.

[0025] Figure 10 illustrates an example of detailed internal components participating in the recommendation process of the recommendation model of Figure 7.

[0026] Fig. 11 is a display screen of a user terminal displaying the health information of Fig. 1.

[0027] Figures 12 and 13 are configuration diagrams of computing systems according to other embodiments.

[0028] Fig. 14 is a block diagram of a digital health literacy level prediction device according to another embodiment.

[0029] Figure 15 is a diagram illustrating the learning process of a classification model according to another embodiment.

[0030] Figure 16 is a diagram exemplarily representing the learning data set of Figure 15.

[0031] Figure 17 is another diagram exemplarily representing the learning data set of Figure 15.

[0032] Figure 18 is another diagram exemplarily representing the learning data set of Figure 15.

[0033] Figure 19 is a diagram illustrating the processing unit of Figure 14.

[0034] Figure 20 is a drawing for explaining the importance of each characteristic according to another embodiment.

[0035] Figure 21 is a diagram illustrating a customized program according to the level of digital health literacy according to another embodiment.

[0036] Figure 22 is a flowchart of a method for predicting a digital health literacy level according to another embodiment.

[0037] Figure 23 is a configuration diagram of a computing system according to another embodiment.

[0038] Figure 24 is a configuration diagram of a client-server computer system according to another embodiment.

[0039] Figure 25 is a block diagram of a digital health literacy-based content service device according to another embodiment.

[0040] Figure 26 is a flowchart of a digital health literacy-based content service method according to another embodiment.

[0041] Figure 27 illustrates a digital health literacy-based content service of Figure 26.

[0042] Embodiments of the present disclosure are described in detail with reference to the accompanying drawings. Specific structural or functional descriptions of the embodiments are provided as examples to illustrate the concepts disclosed herein. Embodiments or examples according to the concepts of the present disclosure may be implemented in various forms, and the scope of the present disclosure is not limited to the embodiments or examples described herein.

[0043] Identical hatching shown across the drawings indicates corresponding or identical areas in the drawings and does not indicate material associated with those areas.

[0044] When one element is described as being "connected" or "coupled" to another element, the elements may be directly connected or coupled, or may be connected or coupled through an intermediate element between the elements. When two elements are described as being "directly connected" or "directly coupled," one element is directly connected or coupled to the other, without any intermediate element between them.

[0045] When an element is described as being placed "above" or "below" another element, the elements may be in direct contact with each other or there may be intermediate elements placed between the elements.

[0046] Terms such as "vertical," "horizontal," "upper," "lower," "above," "below," "front," "rear," "side," "column," "row," and other relative spatial relationships or directions are used solely for the purpose of facilitating description or reference to the drawings and are not limiting. Other spatial relationships or directions not shown in the drawings or described in the specification are also possible within the scope of this specification.

[0047] Terms such as "first" and "second" are used to distinguish different elements and do not imply size, order, priority, quantity, or importance of the elements. For example, in some embodiments, a first element may be referred to as a second element, and in other embodiments, a second element may be referred to as a first element.

[0048] If an element included in the embodiments in this specification is described in singular form, the element may be interpreted to include a plurality of elements that perform the same or similar function.

[0049] Figure 1 is a block diagram of a health information recommendation device according to one embodiment. Figure 2 illustrates the training dataset for the recommendation model of Figure 1. Figure 3 illustrates the process of providing health information using the trained recommendation model.

[0050] Referring to FIG. 1, a health information recommendation device (10) according to one embodiment includes an input unit (20) for receiving metadata of an arbitrary user, and a processing unit (30) for recommending health information of a user for metadata of an arbitrary user using a recommendation model (32) trained to generate health information of the user for metadata of the user using learning data consisting of input data including metadata of the user and interaction data of the user, and output data including health information of the user corresponding to the metadata of the user and the interaction data of the user.

[0051] As illustrated in FIG. 2, a recommendation model (32a) that has not been pre-trained is trained to generate user health information for the user's metadata using training data consisting of input data including the user's metadata and the user's interaction data and output data including the user's health information corresponding to the user's metadata and the user's interaction data.

[0052] As illustrated in Fig. 3, the learned recommendation model (32) recommends health information for a user based on metadata of any user. The learned recommendation model (32) can recommend health information by receiving only the user's metadata as input data through the input unit (20) without using the user's interaction data used in the learning process.

[0053] Figure 4 illustrates the types of metadata and interaction data of Figure 1. Figure 5 illustrates metadata of Figure 4, including demographic information and disease and lifestyle information.

[0054] As illustrated in Figure 4, the user's metadata includes at least one of demographic information including at least one of gender, age, place of residence, marital status, and subjective economic status, and information regarding a disease or lifestyle including at least one of diagnosed disease, smoking, drinking, exercise, snacking, and preference for salty foods. The user metadata may include, but is not limited to, both demographic information and disease and lifestyle information.

[0055] User interaction data includes health information content including preferences.

[0056] A user's health information includes at least one of the following: disease-related information, health promotion information, hospital usage information, and medical service policy information. The disease-related information includes at least one of the following: high blood pressure, diabetes, depression, or sleep disorders.

[0057] The global population is aging rapidly. The proportion of the global population aged 60 and older is projected to nearly double from 12% in 2015 to 22% by 2050. At the same time, health needs become more complex with age, with an increase in chronic and multimorbid conditions. This shift is driving a growing need for healthcare systems worldwide to provide personalized health information that enables personalized care and preventative measures tailored to the specific needs of older adults.

[0058] The digital age has led to an explosive increase in the volume and variety of health-related information. However, this vast amount of data may not be tailored to the specific conditions or health concerns of older adults, and the exponential growth of health information can hinder providing them with practical and actionable health information. Because older adults have the lowest digital literacy levels among the four information-vulnerable groups (people with disabilities, low-income families, farmers and fishermen, and the elderly), their ability to access, understand, and utilize digital health information is limited. In this context, services that automatically recommend and provide relevant information, rather than requiring individuals to manually search for it, are needed.

[0059] Existing health information recommendation services have limitations, offering most users either general health information or very limited customization. These services often fail to fully utilize information about senior users' medical conditions, individual health conditions, and lifestyle habits, making them unsuitable for health management given individual differences. Therefore, personalized recommendations tailored to a variety of health conditions are essential.

[0060] Moreover, seniors face significant challenges accessing essential services like hospitals. When visiting a hospital for treatment, they face challenges navigating the hospital system, which requires multiple digital procedures like QR code authentication and the use of self-service check-ins. Therefore, rather than focusing solely on the knowledge-based aspects of personal health management, such as "understanding and treating diseases" and "health promotion and prevention methods," we need to provide content related to "hospital information and usage guides" and "medical services and policies," which are difficult to access and understand.

[0061] To address these issues, we can provide health information-related content optimized for senior users by leveraging deep learning technology and learning and recommendation algorithms based on demographic data, lifestyle data, and content preference data related to senior users.

[0062] Through the DIDIM-S (Senior Customized Digital Health Literacy Capacity Building Education and User-Centered Health Information Delivery Service) app running on the terminal, senior users are provided with customized health information content for each disease (hypertension, diabetes, depression, sleep disorder) in four categories (disease understanding and treatment-related information, health promotion and prevention-related information, hospital information and usage guide-related information, medical service and policy-related information).

[0063] Existing recommendation systems can be broadly categorized into content-based filtering, collaborative filtering, and hybrid recommendation systems. Content-based filtering analyzes the characteristics (content) of items previously rated or preferred by users to recommend new items with similar characteristics, while collaborative filtering makes recommendations based on user interactions (e.g., ratings, purchase history). Finally, hybrid recommendation systems combine the strengths of content-based filtering and collaborative filtering. In a single system, content-based filtering provides early item recommendations to users, and collaborative filtering refines or supplements these recommendations.

[0064] However, the user-item matrices handled by recommender systems are often extremely sparse. Most users only rate or purchase a very small fraction of items. Traditional recommender models struggle to effectively handle this sparseness. Furthermore, when a user is new to the platform or a new product is released, the recommender system faces the cold-start problem, which makes it difficult to provide accurate recommendations due to insufficient information about the user or item.

[0065] Additionally, user metadata was used to address the cold-start problem in recommendation systems. User metadata includes demographic, medical, and lifestyle data, such as age, gender, and occupation, provided by the user during the membership registration process or collected from the user's device and preferences. When a new user joins the platform, metadata provides an initial point in the latent space for predicting the user's preferences in the absence of information about their behavioral patterns or preferences. Park et al. (2006) reported that they used user age and gender to establish initial estimates for the recommendation algorithm, generating more accurate recommendations and improving the performance of a movie recommendation system.

[0066] The recommendation model (32) used in the health information recommendation device (10) according to one embodiment is a model that combines a probabilistic graphical model and deep learning, and is used to learn the latent representation of data. Since the health information recommendation device (10) according to one embodiment has the advantage of modeling the latent characteristics of data based on the recommendation model (32), it can partially resolve the cold-start problem by extracting meaningful characteristics even from sparse data.

[0067] Figure 7 is a block diagram of the recommended model of Figure 1.

[0068] Referring to Fig. 7, the recommendation model (32) is an encoder (34) that receives user interaction data and user metadata as input, maps the user interaction data to a latent variable (Zi) of the interaction data, and maps the user metadata to a latent variable (Zm) of the metadata, and uses a mapping matrix (Wmap) to map the latent variable (Zm) of the interaction data through the metadata-based latent variable obtained from the user metadata. ) and a mapping network (36) that predicts the latent variables of the predicted interaction data ( ) may include a decoder (38) that generates reconstructed health information.

[0069] Interaction data records traces of user interactions with content. To make this data easier to handle, preprocessing involves binarization. For example, if a user interacts with content (e.g., clicks, likes, etc.), it is marked as a "1," while if there is no interaction, it is converted to a "0."

[0070] Preprocessed interaction data is input into a model called an encoder (34) to compress the data into a latent space. Latent space refers to a space that summarizes important features from the original data. The encoder (34) learns key patterns in the data through multiple layers of neural networks and compresses them into latent variables. These latent variables contain core information from the interaction data and are structured with dimensions much smaller than the original data.

[0071] Latent variables extracted from interaction data are linked to latent variables in metadata. To achieve this, a mapping network (36) layer is added to the model to learn the relationship between interaction data and metadata. The mapping network is a learning layer necessary for recommending content based solely on metadata. It linearly transforms the latent space of metadata into the interaction latent space, enabling prediction of interaction data using latent variables in metadata. For example, this can be viewed as a process for learning how information such as a user's age, gender, and interests influence their interactions with specific content.

[0072] The latent variables connected through the mapping network (36) are now input to the decoder (38). The decoder (38) plays a role in restoring the original interaction data from the latent variables. The decoder (38) uses a reverse neural network structure to reproduce the original characteristics of the interaction data, thereby restoring the latent variables back into the form of interaction data. Through this process, the decoder (38) predicts, as a probability value, which content, i.e., which health information, the user is likely to be interested in. In other words, the data reconstructed by the decoder (38) provides a prediction value regarding which content, i.e., which health information, the user is likely to be interested in in the future, and makes recommendations based on this.

[0073] Through the above learning, latent variables are generated by utilizing the parameters of the encoder (34) and the parameters of the decoder (38), and the prediction output value of the decoder (38) is performed based on the generated latent variables.

[0074] Figure 8 illustrates an example of the detailed internal configuration involved in the learning process of the recommendation model of Figure 7. Figure 9 illustrates the relationship between the variables of each component in the learning process of the recommendation model of Figure 8. Figure 10 illustrates an example of the detailed internal configuration involved in the recommendation process of the recommendation model of Figure 7.

[0075] Referring to FIGS. 8 and 9, the encoder (34) includes a first encoder (34a) that maps the user's interaction data to a latent variable (Zi) of the interaction data, and a second encoder (34b) that maps the user's metadata to a latent variable (Zm) of the metadata.

[0076] The recommendation model (32) maps the user's interaction data to a latent variable (Zi) of the interaction data and maps the user's metadata to a latent variable (Zm) of the metadata through the first encoder (34a) and the second encoder (34b) during learning.

[0077] Referring to FIG. 10, the recommendation model (32) maps arbitrary user metadata to latent variables (Zm) of metadata through the second encoder (34b) and the second encoder (38b) when recommending user health information for arbitrary user metadata, and uses the mapping matrix (Wmap) of the mapping network to map the latent variables (Zm) based on arbitrary user metadata to latent variables (of interaction data) of interaction data. ) is predicted.

[0078] The decoder (38) is a latent variable of the predicted interaction data ( ) generates health information based on the user's interaction data reconstructed through the system.

[0079] The recommended model (32) may be characterized architecturally by additionally including a mapping network in addition to the model including a general encoder (34) and decoder (38).

[0080] In a model that includes only a general encoder (34) and decoder (38), the encoder (34) encodes the input data into a low-dimensional latent space, and the decoder (38) decodes it back into the original dimension to reconstruct the input data. The encoder (34) encodes the input data into a latent variable , and in this process, the features of the input data are extracted and mapped to a latent space that follows a normal distribution. The decoder (38) extracts latent variables sampled from the latent space. It takes as input and outputs it as restored to the original data space. , and in this process, the input data is reconstructed and trained to have a structure similar to the original input data.

[0081] Encoder (34) input data Latent variables in The mean μ and variance σ of 2 It is estimated that

[0082] [Mathematical Formula 1]

[0083]

[0084] [Equation 2]

[0085]

[0086] In equations 1 and 2 is the activation value of the encoder hidden layer, means weight, means bias.

[0087] The estimated mean and variance are used as latent variables using the reparameterization trick. , which allows the VAE model to learn efficiently through backpropagation.

[0088] [Equation 3]

[0089]

[0090] In mathematical formula 3 means normal distribution.

[0091] Decoder (38) is a latent variable Data reconstructed through Creates.

[0092] [Equation 4]

[0093]

[0094] In mathematical formula 4 is the activation value of the decoder hidden layer, is a sigmoid function that represents the reconstructed data as a probability value between 0 and 1.

[0095] A model that only includes a general encoder (34) and decoder (38) is trained by minimizing a combination of two loss functions. The first is the reconstruction loss, which minimizes the difference between the original input data and the reconstructed data. The second is the regularization loss, which encourages the latent variables to follow a normal distribution. The reconstruction loss is a loss function that minimizes the difference between the input data and the reconstructed data. and reconstructed data Measure the difference between

[0096] [Equation 5]

[0097]

[0098] In mathematical equation 5 is a latent variable is the prior distribution (usually a standard normal distribution), is input for is the posterior distribution. The second term refers to the Kullback-Leibler divergence.

[0099] The first term is the reconstruction loss, which is the input data and reconstructed data minimizes the difference between the two. The second term regularization loss is the latent variable A normal distribution We use the Kullback-Leibler Divergence to induce the latent variable to follow. Distribution of and The distribution of latent variables is fitted to a normal distribution by minimizing the Kullback-Leibler divergence between the two.

[0100] The recommendation model (32) combines metadata with the aforementioned encoder (34) and decoder (38) to obtain latent variables obtained from the metadata. The latent space of interaction data through A mapping network (36) is used to predict .

[0101] In this process, the mapping network (36) is a mapping matrix W max Using the interaction latent space It is estimated that

[0102] [Equation 6]

[0103]

[0104] Additionally, to increase the consistency between the metadata-based latent space and the interaction-based latent space in the loss function, we use a regularization loss (L map ) is added. Through this, a mapping network (36) is trained to transform the metadata-based latent space into an interaction-based latent space.

[0105] [Equation 7]

[0106]

[0107] The final loss function is the reconstruction loss of metadata (L m ), reconstruction loss of interaction data (L i), and the regularization loss function (L) of the mapping network (36) map ) is defined. Optimization is performed through this loss, and the recommendation model (32) is repeatedly updated to perform user-tailored recommendations.

[0108] [Equation 8]

[0109]

[0110] In mathematical equation 8 are the weights of the loss and regularization loss of the interaction data.

[0111] In other words, the recommendation model (32) is trained to minimize the final loss function, which includes the reconstruction loss of user metadata, the reconstruction loss of user interaction data, and the regularization loss of the mapping network.

[0112] The recommendation model (32) exhibits better generalization performance compared to existing deep learning models and maintains high performance even when recommending new users or items. This is because the recommendation model (32) maps user or item characteristics to a latent space, providing an opportunity to explore the possibilities for items not yet rated by users. Furthermore, because the recommendation model (32) is a type of generative model, it also enables predictions for data points that do not actually exist.

[0113] [Experimental Example]

[0114] To evaluate the recommendation accuracy for a personalized health information service, the recommendation system utilized data from a health information needs survey conducted on 2,000 men and women aged 19 to 75. The demographic, disease, and lifestyle information used in the metadata training the recommendation algorithm are presented in Table 1.

[0115] The health information content presented in Figure 6 was analyzed by reviewing and comprehensively analyzing previous studies that analyzed messages from Health Recommender Systems (HRS) and patient portal sites using natural language processing. Health information recommendations were divided into four main categories: disease understanding and treatment, health promotion and prevention, medical services and policies, and hospital information and usage guide. Subcategories were organized for each category. For learning, a health information needs survey was conducted on 2,000 adult men and women aged 19 to 75 years. The preferences for detailed health information items were determined in order of 1st, 2nd, and 3rd place, and data was collected to enable a detailed understanding of the preferences, and the collected data was used as learning data.

[0116] [Table 1]

[0117]

[0118]

[0119] The ranking data for health information preferences was preprocessed to be expressed as a one-hot vector by assigning a value of 1 to the index of ranked data and 0 to the index of unranked data through one-hot encoding.

[0120] To evaluate recommendation accuracy, 80% of the total data was used as a training dataset and 20% as a test dataset. After training with the training dataset, the optimal model was derived by validating it with the test dataset. To evaluate the performance of the recommendation system, we use Precision@K as an evaluation metric. Precision@K is the precision calculated for K recommendation results, and measures the proportion of items among the top K recommended items recommended by the model that are actually preferred by users. This metric does not consider the ranking of the recommended items and is an indicator of the accuracy of the top K recommendations.

[0121] Experimental results showed that the trained recommendation system achieved a Precision@3 performance of 73.6% when recommending the top three contents.

[0122] [Equation 9]

[0123]

[0124] As mentioned above, the amount of information available in today's digital society is incomparably greater than in the past. However, seniors, a vulnerable population, often find it difficult to discern what they need from the vast amount of information. Therefore, in situations where access and judgment are difficult, it's crucial to identify which diseases and categories of health information they need and prefer, and then automatically suggest or provide this information.

[0125] Fig. 11 is a display screen of a user terminal displaying the health information of Fig. 1.

[0126] As illustrated in FIG. 11, a health information recommendation device (10) according to one embodiment collects data through a simple questionnaire via a user terminal (40) and provides disease-specific health information content for major diseases of the elderly, such as hypertension, diabetes, depression, and sleep disorders, rather than general health information, thereby maximizing the efficiency of personal health management by expanding the radius from home to the hospital and supporting the user to actively and independently perform health management, thereby helping to maintain an optimal health state.

[0127] Furthermore, the platform recommends content related to "Disease Understanding and Treatment" and "Health Promotion and Prevention," as well as "Hospital Information and Use" and "Medical Services and Policies," to enhance understanding of the hospital system and provide customized health information content for seniors to utilize hospitals more conveniently. This is expected to facilitate access to health information and enhance understanding of hospital use by providing seniors with the information they need through recommendations, eliminating the need for them to search for it themselves.

[0128] The health information recommendation device (10) described above with reference to FIG. 1 can receive health information of a user selected by an actual user, be retrained, and recommend health information for any user metadata based on the retrained recommendation model. As illustrated in FIG. 1, the learned recommendation model (32) can receive health information of a user selected by an actual user, be retrained, and the processing unit (30) can recommend health information for any user metadata based on the retrained recommendation model.

[0129] According to a health information recommendation device (10) according to one embodiment, based on the user's input data, a pre-learned recommendation model (32) is utilized to convert data into a latent space and learn latent characteristics, thereby recommending user-tailored health information.

[0130] According to a health information recommendation device (10) according to one embodiment, the user's health information and log data are utilized for re-learning to update the recommendation model, thereby continuously recommending optimized health information to the user.

[0131] A health information recommendation method according to another embodiment includes an input step of receiving metadata of an arbitrary user; and a processing step of recommending health information of a user for metadata of an arbitrary user using a recommendation model (32) trained to generate health information of the user for metadata of the user using learning data consisting of input data including metadata of the user and interaction data of the user and output data including health information of the user corresponding to the metadata of the user and the interaction data of the user.

[0132] As described with reference to FIGS. 4 and 5, the user's metadata includes at least one of demographic information including at least one of gender, age, place of residence, marital status, and subjective economic status, and information regarding a disease or lifestyle including at least one of diagnosed disease, smoking, drinking, exercise, snacking, and preference for salty foods. The user metadata may include, but is not limited to, both demographic information and disease and lifestyle information.

[0133] User interaction data includes health information content including preferences.

[0134] A user's health information includes at least one of the following: disease-related information, health promotion information, hospital use information, and medical service policy information. Disease-related information includes at least one of the following: high blood pressure, diabetes, depression, or sleep disorders.

[0135] First, we collect user demographic data, data on diseases and lifestyle habits, and health information content preferences. As described above in the experimental example, the subjects: 2,000 adult men and women aged 19 to 75 years old. The questionnaire includes demographic information (e.g., gender, age group, place of residence, resident, marital status, subjective economic level), disease and lifestyle information (e.g., diagnosed disease, smoking, drinking, exercise, snacking, salty food consumption, eating out), and health information content information (e.g., preferences).

[0136] Users are categorized based on their diagnosed or interested condition. For example, a diagnosed or interested condition may include at least one of the following: hypertension, diabetes, depression, or sleep disorder. For example, an application user can select a diagnosed or interested condition and then categorize users based on that diagnosis into hypertension, diabetes (chronic conditions), depression, and sleep disorders (mental illnesses).

[0137] As described with reference to FIGS. 7 and 8, the recommendation model (32) is an encoder (34) that receives user interaction data and user metadata as input, maps the user interaction data to a latent variable (Zi) of the interaction data, and maps the user metadata to a latent variable (Zm) of the metadata, and uses a mapping matrix (Wmap) to map the latent variable (Zm) of the interaction data through the metadata-based latent variable obtained from the user metadata. ) and a mapping network (36) that predicts the latent variables of the predicted interaction data ( ) and a decoder (38) that generates reconstructed health information.

[0138] The collected metadata, interaction data, and health information are input into the recommendation model (32).

[0139] As described with reference to mathematical expression 8, the recommendation model (32) can be trained to minimize a final loss function including the reconstruction loss of user metadata, the reconstruction loss of user interaction data, and the regularization loss of the mapping network.

[0140] As described in the learning process and inference process with reference to FIGS. 8 to 10, the encoder (34) includes a first encoder (34a) that maps the user's interaction data to a latent variable (Zi) of the interaction data, and a second encoder (34b) that maps the user's metadata to a latent variable (Zm) of the metadata.

[0141] The recommendation model (32) maps the user's interaction data to a latent variable (Zi) of the interaction data and maps the user's metadata to a latent variable (Zm) of the metadata through the first encoder (34a) and the second encoder (34b) during learning.

[0142] Referring to Fig. 10, the recommendation model (32) maps arbitrary user metadata to a latent variable (Zm) of metadata through the second encoder (34b) when recommending user health information for arbitrary user metadata, and uses the mapping matrix (Wmap) of the mapping network to map the latent variable (Zm) based on arbitrary user metadata to a latent variable (Zm) of interaction data. ) is predicted.

[0143] The decoder (38) is a latent variable of the predicted interaction data ( ) generates health information based on the user's interaction data reconstructed through the system.

[0144] The recommendation model (32) provides health information that matches the analyzed user's disease and preferences.

[0145] As described above with reference to FIG. 10, the learned recommendation model (32) is relearned by receiving health information of a user selected by an actual user, and the processing unit (30) can recommend health information for any user metadata based on the relearned recommendation model.

[0146] As previously described with reference to Figure 10, the log data collected through the feedback monitoring unit can be regularly reflected in the recommendation method for retraining. This allows for updated health information tailored to the user's condition, providing recommended health information content.

[0147] According to a health information recommendation method according to another embodiment, a pre-learned recommendation model is utilized based on the user's input data to convert the data into a latent space and learn latent characteristics, thereby recommending customized health information for the user.

[0148] According to a health information recommendation method according to another embodiment, a health information recommendation device that continuously recommends optimized health information to a user can be provided by updating a recommendation model by utilizing the user's health information and log data for re-learning.

[0149] Figure 12 is a block diagram of computing systems according to another embodiment.

[0150] Referring to FIG. 12, a computing system (300) may include a memory (310) and a processor (320).

[0151] The memory (310) can store data (80) and software, etc. required to perform the health information recommendation method (50), but can also be stored separately in a separate large-capacity storage server, etc. The memory (310) can be a volatile memory (e.g., SRAM, DRAM) or a non-volatile memory (e.g., NAND Flash). The memory (310) stores a health information recommendation model (32).

[0152] The processor (320) can perform the health information recommendation method (50) described above.

[0153] When a health information recommendation task is requested, the processor (320) executes the health information recommendation model (32) stored in the memory (310) to provide health information.

[0154] A computing system according to embodiments of the present invention may include a computer device (300) including a memory (310) and a processor (320), and a server (400) including a memory (410) and a processor (420), as illustrated in FIG. 13. The computer device (300) and the server (400) may be connected wired or wirelessly via a network.

[0155] The memory (410) of the server (400) can store a health information recommendation model (32).

[0156] When a health information recommendation task is requested (query), the processor (320) of the computer device (300) can transmit this request (query) to the server (400).

[0157] The processor (420) of the server (400) can execute a health information recommendation model (32) to recommend health information and transmit the result to a computer device (300).

[0158] Various possible examples are described below for the computer system described with reference to FIGS. 12 and 13.

[0159] The health information recommendation device (10) may be configured as a computing system (300) as shown in FIG. 12, or may be configured as a GPU server equipped with a GPU processor and general memory and storage for storing data (80) and a health information recommendation model (32), but the present invention is not limited thereto.

[0160] The aforementioned health information recommendation device (10) may be implemented by a computing device including at least some of a processor, a memory, a user input device, and a presentation device. The memory is a medium that stores computer-readable software, applications, program modules, routines, instructions, and / or data, which are coded to perform a specific task when executed by the processor. The processor can read and execute the computer-readable software, applications, program modules, routines, instructions, and / or data stored in the memory. The user input device may be a means for allowing a user to input a command to cause the processor to perform a specific task or to input data necessary for the execution of a specific task. The user input device may include a physical or virtual keyboard or keypad, key buttons, a mouse, a joystick, a trackball, a touch-sensitive input device, or a microphone. The presentation device may include a display, a printer, a speaker, or a vibration device.

[0161] Memory is a medium that stores computer-readable software, applications, program modules, routines, instructions, and / or data, which are coded to perform a specific task when executed by a processor. The processor can read and execute the computer-readable software, applications, program modules, routines, instructions, and / or data stored in the memory. A user input device may be a means for a user to input a command to cause the processor to perform a specific task or to input data necessary for the execution of a specific task. A user input device may include a physical or virtual keyboard or keypad, key buttons, a mouse, a joystick, a trackball, a touch-sensitive input device, or a microphone. A presentation device may include a display, a printer, speakers, or a vibration device.

[0162] Computing devices can include a variety of devices, including smartphones, tablets, laptops, desktops, servers, and clients. A computing device may be a single, standalone device, or it may include multiple computing devices operating in a distributed environment, each of which collaborates with another through a communications network.

[0163] Meanwhile, computing devices may be quantum computing devices, rather than classical computing devices. Quantum computing devices perform calculations on qubits, not bits. Qubits can be in a superposition of 0 and 1 simultaneously, and with M qubits, they can express 2^M states simultaneously.

[0164] Quantum computing devices can use various types of quantum gates (e.g., Pauli / Rotation / Hadamard / CNOT / SWAP / Toffoli) that input one or more qubits to perform quantum operations and perform designated operations, and can combine quantum gates to form quantum circuits that perform special functions.

[0165] Quantum computing devices can use quantum artificial neural networks (e.g. QCNN, QGRNN) that can perform functions performed by conventional artificial neural networks (e.g. CNN, RNN) at a faster speed while using fewer parameters.

[0166] In addition, the aforementioned health information recommendation device (10) can be executed by a computing device having a processor and a memory storing computer-readable software, applications, program modules, routines, instructions, and / or data structures, etc., coded to perform a health information recommendation method when executed by the processor.

[0167] The embodiments described above may be implemented through various means. For example, the embodiments may be implemented through hardware, firmware, software, or a combination thereof.

[0168] In the case of hardware implementation, the health information recommendation method (50) according to the present embodiments may be implemented by one or more ASICs (Application Specific Integrated Circuits), DSPs (Digital Signal Processors), DSPDs (Digital Signal Processing Devices), PLDs (Programmable Logic Devices), FPGAs (Field Programmable Gate Arrays), processors, controllers, microcontrollers, or microprocessors.

[0169] For example, the health information recommendation method according to the embodiments can be implemented using an artificial intelligence semiconductor device in which the neurons and synapses of a deep neural network are implemented using semiconductor devices. The semiconductor devices may be currently used semiconductor devices, such as SRAM, DRAM, or NAND, or may be next-generation semiconductor devices, such as RRAM, STT MRAM, or PRAM, or may be a combination thereof.

[0170] When implementing a health information recommendation method according to embodiments using an artificial intelligence semiconductor device, the results (weights) of learning a deep learning model using software may be transferred to synapse-mimicking elements arranged in an array, or learning may be performed in the artificial intelligence semiconductor device.

[0171] When implemented via firmware or software, the health information recommendation method according to the present embodiments may be implemented in the form of a device, procedure, or function that performs the functions or operations described above. The software code may be stored in a memory unit and executed by a processor. The memory unit may be located within or outside the processor and may exchange data with the processor via various known means.

[0172] Additionally, terms such as "system," "processor," "controller," "component," "module," "interface," "model," or "unit" as described above may generally refer to a computer-related entity, such as hardware, a combination of hardware and software, software, or software in execution. For example, the aforementioned components may be, but are not limited to, a process driven by a processor, a processor, a controller, a control processor, an object, a thread of execution, a program, and / or a computer. For example, both an application running on a controller or a processor and the controller or the processor may be components. One or more components may be within a process and / or thread of execution, and the components may be located on a single device (e.g., a system, a computing device, etc.) or distributed across two or more devices.

[0173] Meanwhile, another embodiment provides a computer program stored on a computer storage medium that performs the aforementioned health information recommendation method. Furthermore, another embodiment provides a computer-readable storage medium storing a program for implementing the aforementioned health information recommendation method.

[0174] The program recorded on the recording medium can be read, installed and executed by a computer, thereby executing the steps described above.

[0175] In this way, in order for a computer to read a program recorded on a recording medium and execute functions implemented as a program, the above-mentioned program may include code coded in a computer language such as C, C++, JAVA, or machine language that can be read by the computer's processor (CPU) through the computer's device interface.

[0176] Such code may include functional code related to functions defining the aforementioned functions, and may also include control code related to execution procedures required for the computer's processor to execute the aforementioned functions according to a predetermined procedure.

[0177] Additionally, such code may further include memory reference related code regarding where in the internal or external memory of the computer the additional information or media required for the computer's processor to execute the aforementioned functions should be referenced.

[0178] Additionally, if the computer's processor needs to communicate with another computer or server located remotely in order to execute the functions described above, the code may further include communication-related code regarding how the computer's processor should communicate with another computer or server located remotely using the computer's communication module, and what information or media should be sent and received during the communication.

[0179] The computer-readable recording medium that records the program as described above includes, for example, ROM, RAM, CD-ROM, magnetic tape, floppy disk, optical media storage device, etc., and may also include one implemented in the form of a carrier wave (e.g., transmission via the Internet).

[0180] Additionally, computer-readable recording media can be distributed across network-connected computer systems, allowing computer-readable code to be stored and executed in a distributed manner.

[0181] In addition, the functional program for implementing the present invention and the code and code segments related thereto may be easily inferred or changed by programmers in the technical field to which the present invention belongs, taking into consideration the system environment of the computer that reads the recording medium and executes the program.

[0182] The aforementioned health information recommendation method can also be implemented in the form of a recording medium containing computer-executable instructions, such as an application or program module executed by a computer. The computer-readable medium may be any available medium that can be accessed by a computer, and includes both volatile and nonvolatile media, removable and non-removable media. Furthermore, the computer-readable medium may include all computer storage media. The computer storage media includes both volatile and nonvolatile, removable and non-removable media implemented in any method or technology for storing information, such as computer-readable instructions, data structures, program modules, or other data.

[0183] The aforementioned health information recommendation method can be executed by an application installed by default on the terminal (which may include a program included in a platform or operating system installed by default on the terminal), or by an application (i.e., a program) directly installed on the master terminal by the user through an application providing server such as an application store server, an application, or a web server related to the service. In this sense, the aforementioned health information recommendation method can be implemented as an application (i.e., a program) installed by default on the terminal or directly installed by the user, and recorded on a computer-readable recording medium such as on the terminal.

[0184] Fig. 14 is a block diagram of a digital health literacy level prediction device according to another embodiment.

[0185] Referring to FIG. 14, a digital health literacy level prediction device (600) according to one embodiment may include an input unit (610) that receives arbitrary demographic factors, health-related factors, and digital information-related factors as input values, a processing unit (620) that classifies a digital health literacy level from arbitrary demographic factors, health-related factors, and digital information-related factors using a classification model (670) that has been trained using a learning data set (640) that uses demographic factors, health-related factors, and digital information-related factors as input data (641) and binary-classified digital health literacy levels as output data (642).

[0186] The input unit (610) can input any demographic factors, health-related factors, and digital information-related factors as input values. For example, demographic factors may include, but are not limited to, gender, age group, and highest level of education. Health-related factors may include, but are not limited to, health status, exercise, health management proficiency, and health management coping skills. Digital information-related factors may include, but are not limited to, digital device utilization skills, health information search experience, health information utilization skills, and health information reliability judgment skills.

[0187] The processing unit (620) may include a classification model (670) that has been previously trained using a learning data set (640). The classification model (670) may predict the level of digital health literacy based on input values. The classification model (670) may be, but is not limited to, an XGBoost model. The level of digital health literacy may be binary classified into a group with low digital health literacy and a group with high digital health literacy, but is not limited thereto.

[0188] Figure 15 is a diagram illustrating the learning process of a classification model according to another embodiment.

[0189] Referring to FIG. 15, a classification model (670) may be pre-trained using a learning data set (640). The learning data set (640) may include input data (641) and output data (642). The input data (641) may include demographic factors, health-related factors, and digital information-related factors. For example, demographic factors may include, but are not limited to, gender, age group, and highest level of education. Health-related factors may include, but are not limited to, health status, exercise, health management proficiency, and health management coping skills. Digital information-related factors may include, but are not limited to, digital device utilization ability, health information search experience, health information utilization ability, and health information reliability judgment ability.

[0190] The output data (642) may include binary classified digital health literacy levels. The binary classified digital health literacy levels may include a group with low digital health literacy (650) and a group with high digital health literacy (660). The output data (642) may be generated based on, but is not limited to, the eHEALS, a self-report scale measuring digital health literacy levels.

[0191] eHEALS is a tool developed to measure digital health literacy levels. The eHEALS measurement results can be used as training data to train a classification model (670). According to one embodiment, a digital health literacy level prediction device (600) can reduce reliance on subjective self-assessments by analyzing data using machine learning or deep learning techniques. In particular, by analyzing patterns in large-scale data, it can minimize individual interpretation bias and provide a more standardized and uniform assessment. Furthermore, by integrating various data sources, it can accurately assess digital health literacy levels across diverse populations.

[0192] Figure 16 is a diagram exemplarily representing the learning data set of Figure 15. Figure 17 is another diagram exemplarily representing the learning data set of Figure 15. Figure 18 is another diagram exemplarily representing the learning data set of Figure 15.

[0193] Referring to FIGS. 16 to 18, it can be seen that a learning data set (640) according to one embodiment has segmented input data (641) items and binary classified output data (642). The learning data set (640) according to one embodiment was created based on a survey conducted on 1,000 elderly people aged 55 to 75.

[0194] Referring to Figure 16, the input data (641) may include demographic factors, including gender, age group, and highest level of education. In the gender question, the result value (Low) indicating low digital health literacy is higher for women than the result value (High) indicating high digital health literacy. Examining the gender question as a whole, it can be seen that women tend to have lower digital health literacy levels than men.

[0195] In the age group question, for those aged 55-59, the low digital health literacy result (Low) was smaller than the high digital health literacy result (High). Older age groups may be less familiar with digital devices, so this tendency may be reflected in the classification model's training.

[0196] In the question about highest level of education, for those with a college degree or higher, the result value (Low) indicating a low level of digital health literacy is smaller than the result value (High) indicating a high level of digital health literacy. Since a higher level of education may lead to greater familiarity with digital devices, this tendency may be reflected in the learning of the classification model. Referring to Figure 17, the input data (641) may include health-related factors including health status, exercise status, health management proficiency, and health management coping ability. In the question about health status, when the health status is very poor or poor, the result value (Low) indicating a low level of digital health literacy is larger than the result value (High) indicating a high level of digital health literacy. However, when the health status is healthy or very healthy, the result value (Low) indicating a low level of digital health literacy and the result value (High) indicating a high level of digital health literacy are similar values. The better the health status, the more likely one is to be interested in health, and the more likely one is to have access to digital health-related devices. This tendency can be reflected in the learning of classification models.

[0197] In the health care proficiency questionnaire, for the "very much" and "very much" responses, the results indicating low digital health literacy (Low) were smaller than those indicating high digital health literacy (High). Since individuals who are more skilled in health care are more likely to be interested in health-related services, this trend can be reflected in the classification model's training.

[0198] In the health management coping ability question, for the case of very much so, the result value (Low) indicating a low level of digital health literacy is smaller than the result value (High) indicating a high level of digital health literacy. High health management coping ability can be seen as having a lot of health-related experience, and this tendency can be reflected in the learning of the classification model. In the exercise question, for the cases of often and daily, the result value (Low) indicating a low level of digital health literacy is smaller than the result value (High) indicating a high level of digital health literacy. The more one exercises, the more interested one is in exercise and health, and this tendency can be reflected in the learning of the classification model. Referring to Figure 18, the input data (641) can include digital information-related factors including digital device utilization ability, health information search experience, health information utilization ability, and health information reliability judgment ability. In the digital device literacy questionnaire, for the "very much so" and "very much so" responses, the "low" (Low) score, indicating low digital health literacy, was lower than the "high" (High) score. Since digital device literacy is directly related to digital health literacy, this trend can be reflected in the classification model's training.

[0199] In terms of health information seeking experience, the results indicating low digital health literacy (Low) were smaller than the results indicating high digital health literacy (High). Since those with health information seeking experience are more interested in health information, this tendency can be reflected in the classification model's training.

[0200] In the health information literacy questionnaire, for "very much so" and "very much so," the result value (Low) indicating low digital health literacy was smaller than the result value (High) indicating high digital health literacy. Since higher health information literacy can be judged to indicate higher digital health literacy, this trend can be reflected in the classification model's training.

[0201] In the health information credibility assessment question, for "very much so" and "very much so," the results for low digital health literacy (Low) were smaller than those for high digital health literacy (High). Since the ability to assess health information credibility presupposes a high level of digital health literacy, this tendency can be reflected in the classification model's training.

[0202] To test whether the groups with low and high digital health literacy are independent of each other, a chi-square test was used. ) was conducted. As a result, it was confirmed that there was a statistically significant difference between the two groups according to each variable and digital health literacy level. In the chi-square test, the p value can indicate whether the result data according to each item of the input data (641) is statistically significant. The smaller the p value, the more statistically significant the relationship between the input data (641) and the result data.

[0203] Figure 19 is a diagram illustrating the processing unit of Figure 14.

[0204] Referring to FIG. 19, the processing unit (620) may include a pre-trained classification model (670). The classification model (670) may be an XGBoost model. The classification model (670) may include a first tree (671) and a second tree (672). For example, the first tree (671) and the second tree (672) may include a root node at the top, and the root node may branch into two child nodes, and each child node may branch into a leaf node again, but is not limited thereto. The root node may branch into two child nodes based on a specific value. For example, based on a preset specific value, if the input data (641) is less than the specific value, it may be classified as a left child node, and if it is greater than the specific value, it may be classified as a right child node. The child nodes may branch into two leaf nodes based on a specific value different from that of the aforementioned root node, but are not limited thereto. That is, a node can be a reference point for branching whether a data value is greater than or less than a specific value, and a result value can be assigned to the branched node. The sum of the result values ​​of each tree can mean a predicted label for the result variable y described later. The XGBoost model can classify by creating a continuous tree by reflecting the result of the first tree (671) to the second tree (672). The XGBoost model can apply weights to a part that the first tree (671) did not predict well and reflect it to the second tree (672).

[0205] An XGBoost model can include an objective function. The objective function can include a loss function and a regularization term. When training an XGBoost model, the objective function can be defined to find optimal parameters. In other words, the objective function can define what the model seeks to optimize. The objective function can be used to evaluate the model's predictive performance and update the model's parameters to improve it. The objective function can be expressed as shown in Equation 10.

[0206] [Equation 10]

[0207]

[0208] In Equation 10, obj(θ) is the objective function, L(θ) is the loss function, Ω(θ) is the regularization term, and θ is the model parameter.

[0209] A loss function can measure the difference between a model's predictions and the actual values. It can assess how well a model predicts. For example, the loss function can be, but is not limited to, a logistic loss function. The loss function can be as shown in Equation 11.

[0210] [Equation 11]

[0211]

[0212] y in equation 11 i is the actual observed label for the outcome variable y, is the predicted label for the outcome variable y.

[0213] Loss functions, for example, the logistic loss function, are a loss function mainly adopted in binary classification. They show good performance even in imbalanced data sets, and are highly compatible with regularization conditions (L1 and L2 penalties), which can be advantageous in preventing overfitting.

[0214] The output data (642) or output value may include a binary classified digital health literacy level. Accordingly, a logistic loss function may be used to binary classify the data into groups with high and low digital health literacy levels.

[0215] The regularization term can prevent overfitting to the training data set (640) during the model training process. The regularization term can prevent overfitting by limiting the growth of specific model variables. This can contribute to the creation of a generalized model. For example, the regularization term may be, but is not limited to, an L2 regularization term. The regularization term can be expressed as in Equation 12.

[0216] [Equation 12]

[0217]

[0218] In equation 12 is the loss weight for adding a node, T is the total number of nodes in the tree, λ is the L2 regularization weight, and ω is the set of node values.

[0219] Regularization terms, such as L2 regularization, can reduce the undue influence of a single feature or noise in the model's weights. Regularization terms, along with the loss function, can influence the optimal parameter estimation of the objective function.

[0220] The XGBoost model uses the variables adopted by each decision tree as criteria for splitting nodes. The tree then divides the data into more uniform subgroups based on the values ​​of these variables. Each split within the tree is optimized based on a gain score calculated from the reduction in the loss function before and after the split. The gain score can be expressed as shown in Equation 13.

[0221] [Equation 13]

[0222]

[0223] G in Equation 13 L is the gradient sum at the left child node, and G R is the gradient sum at the right child node, and H L is the Hessian sum at the left child node, and H Ris the Hessian sum at the right child node, λ is the L2 regularization weight, and is a regularization parameter.

[0224] The gain score measures the effect of splitting a specific node, and by calculating the information gain before and after splitting a node, we can quantitatively evaluate how much each split contributed to improving the performance of the overall model. In Equation 4, can provide an additional penalty for node splitting as a regularization parameter to prevent overfitting.

[0225] When splitting a tree node, the minimum instance weight required for the child node can be set. For example, the minimum instance weight can be 2, but is not limited to this. The minimum instance weight determines the splitting of nodes during each tree generation process, thereby adjusting the relationship between data characteristics and target values. This can help prevent model overfitting by ensuring the model learns conservatively and avoiding meaningless splits in nodes with limited data.

[0226] XGBoost (eXtreme Gradient Boosting) is a supervised learning algorithm that adds a technique to prevent overfitting to the gradient boosting algorithm. Gradient boosting can minimize the difference between the predicted probability for each data point and the actual label. Gradient boosting is used to train the t-th model F. t (X) can be updated. The t-th model can be as shown in Equation 14.

[0227] [Equation 14]

[0228]

[0229] F in Equation 14 t-1 (X) is the sum of gradients at the left and right child nodes, η is the learning rate, and h t (X) is the prediction function of the new tree learned at the current stage.

[0230] The t-th model F in gradient boosting t In the process of updating (X), the learning rate η controls the model update size at each boosting step.

[0231] In the process of minimizing the difference between the predicted probability for each data point and the actual label, the gradient of the loss function at each step, g i can be calculated. The gradient can be as shown in Equation 15.

[0232] [Equation 15]

[0233]

[0234] y in equation 15 i is the actual label value of the ith data point, and F t-1 is the predicted value of the ith data point predicted by the previous model, and g i is the gradient value of the ith data point.

[0235] The XGBoost model can sequentially update the model by adding new trees based on the gradient of the loss function for each data point.

[0236] In summary, the XGBoost model according to one embodiment can be trained through the following steps: 1) setting an initial estimate and building a first tree, 2) calculating residuals to add a second tree to the model, 3) calculating the gradient of a loss function to determine an optimal tree split, and 4) defining the minimum instance weight that a child node in each split should have as 2 to prevent overfitting and ensure accuracy.

[0237] Figure 20 is a drawing for explaining the importance of each characteristic according to another embodiment.

[0238] Referring to Fig. 20, the learning results of the classification model (670) can be confirmed. In particular, the importance of each feature can be confirmed. Here, the feature refers to each item of the input data (641) described above. For example, the feature may be, but is not limited to, gender, exercise status, and health information utilization ability. It can be confirmed that the importance of health information reliability judgment ability is 0.2520, the importance of health information utilization ability is 0.1715, and the importance of digital device utilization ability is 0.1077, which are higher in importance than other features. In other words, it can be confirmed that the importance of health information reliability judgment ability and health information utilization ability among the input data (641) play the most important role in determining the level of digital health literacy.

[0239] Figure 21 is a diagram illustrating a customized program according to the level of digital health literacy according to another embodiment.

[0240] Referring to Figure 21, the digital health literacy level can be divided into a group with low digital health literacy (650) and a group with high digital health literacy (660), but is not limited thereto. Customized capacity building programs can be provided according to the classified digital health literacy level. A basic capacity building program (651) can be provided to the group with low digital health literacy (650). For example, in the case of the group with low digital health literacy (650), it is important to improve basic digital skills and understanding of health information, so education using easily accessible and simplified explanations and visual tools, such as basic smartphone usage and internet search, may be necessary, but is not limited thereto.

[0241] A capacity-building application program (661) can be provided to a group with high digital health literacy (660). For example, for this group (660), since they possess basic digital skills and an understanding of health information, providing education on personal information security using online platforms and understanding specialized health information can be helpful, but is not limited to this. To achieve this, it is necessary to predict the level of digital health literacy and develop a customized capacity-building program strategy based on this level.

[0242] In one embodiment, an automatic recommendation algorithm may be used to recommend appropriate capacity building programs to a group with low digital health literacy (650) and a group with high digital health literacy (660), respectively. The automatic recommendation algorithm may include, but is not limited to, algorithms such as machine learning algorithms, artificial neural network-based learning algorithms, or quantum machine learning algorithms. For example, the automatic recommendation algorithm may include, but is not limited to, regression algorithms and classification algorithms based on supervised learning, clustering algorithms and dimensionality reduction algorithms based on unsupervised learning, or convolutional neural networks (CNNs) and recurrent neural networks (RNNs) based on artificial neural networks.

[0243] Classifying digital health literacy levels using a digital health literacy prediction algorithm allows for more accurate and comprehensive predictions of digital health literacy levels by analyzing large-scale data from diverse populations, compared to traditional self-reporting tests. Furthermore, the digital health literacy prediction algorithm is designed to minimize bias and take into account the versatility of digital technology, which can help provide customized digital health literacy capacity building programs tailored to individual needs and characteristics.

[0244] Figure 22 is a flowchart of a method for predicting a digital health literacy level according to another embodiment.

[0245] Referring to FIG. 22, a method for predicting a digital health literacy level according to another embodiment may include a collection step (S110) of generating a learning data set (640) that uses demographic factors, health-related factors, and digital information-related factors as input data (641) and uses a binary-classified digital health literacy level as output data (642), a learning step (S120) of learning a classification model (670) that predicts a digital health literacy level using the learning data set (640), an input step (S130) of inputting any demographic factors, health-related factors, and digital information-related factors as input values ​​to the classification model (670), and an output step (S140) of outputting a binary-classified digital health literacy level corresponding to the input value as a result value.

[0246] The collection step (S110) may generate a learning data set (640). The learning data set (640) may include input data (641) and output data (642). The input data (641) may include demographic factors, health-related factors, and digital information-related factors. For example, demographic factors may include, but are not limited to, gender, age group, and highest level of education. Health-related factors may include, but are not limited to, health status, exercise, health management proficiency, and health management coping skills. Digital information-related factors may include, but are not limited to, digital device utilization skills, health information search experience, health information utilization skills, and health information reliability judgment skills.

[0247] The output data (642) may include binary classified digital health literacy levels. The binary classified digital health literacy levels may include a group with low digital health literacy (650) and a group with high digital health literacy (660). The output data (642) may be generated based on, but is not limited to, the eHEALS, a self-report scale measuring digital health literacy levels.

[0248] eHEALS is a tool developed to measure digital health literacy levels. The eHEALS measurement results can be used as training data to train a classification model (670). According to one embodiment, a method for predicting digital health literacy levels can reduce reliance on subjective self-assessments by analyzing data using machine learning or deep learning techniques. In particular, analyzing patterns in large-scale data can minimize individual interpretation bias, providing a more standardized and uniform assessment. Furthermore, the integration of various data sources can accurately assess digital health literacy levels across diverse populations.

[0249] The learning step (S120) can train a classification model (670) using a learning data set (640). The classification model (670) can predict the level of digital health literacy based on input values. The classification model (670) may be, but is not limited to, an XGBoost model. The XGBoost model may include an objective function. The objective function may include a loss function and a regularization term.

[0250] When training an XGBoost model, an objective function can be defined to find optimal parameters. The objective function can be expressed as in Equation 10.

[0251] [Equation 10]

[0252]

[0253] In Equation 10, obj(θ) is the objective function, L(θ) is the loss function, Ω(θ) is the regularization term, and θ is the model parameter.

[0254] A loss function can measure the predictive power of a trained model. For example, the loss function may be, but is not limited to, a logistic loss function. The loss function can be expressed as in Equation 11.

[0255] [Equation 11]

[0256]

[0257] y in equation 11 i is the actual observed label for the outcome variable y, is the predicted label for the outcome variable y.

[0258] Loss functions, for example, the logistic loss function, are a loss function mainly adopted in binary classification. They show good performance even in imbalanced data sets, and are highly compatible with regularization conditions (L1 and L2 penalties), which can be advantageous in preventing overfitting.

[0259] The output data (642) or output value may include a binary classified digital health literacy level. Accordingly, a logistic loss function may be used to binary classify the data into groups with high and low digital health literacy levels.

[0260] The regularization term is a regularization term that prevents overfitting to the training data set (640) during the model training process, thereby improving the model's predictive power and generalizability. For example, the regularization term may be an L2 regularization term, but is not limited thereto. The regularization term may be expressed as in Equation 12.

[0261] [Equation 12]

[0262]

[0263] In equation 12 is the loss weight for adding a node, T is the total number of nodes in the tree, λ is the L2 regularization weight, and ω is the set of node values.

[0264] Regularization terms, such as L2 regularization, can reduce the undue influence of a single feature or noise in the model's weights. Regularization terms, along with the loss function, can influence the optimal parameter estimation of the objective function.

[0265] The XGBoost model uses the variables adopted by each decision tree as criteria for splitting nodes. The tree then divides the data into more uniform subgroups based on the values ​​of these variables. Each split within the tree is optimized based on a gain score calculated from the reduction in the loss function before and after the split. The gain score can be expressed as shown in Equation 13.

[0266] [Equation 13]

[0267]

[0268] G in Equation 13 Lis the gradient sum at the left child node, and G R is the gradient sum at the right child node, and H L is the Hessian sum at the left child node, and H R is the Hessian sum at the right child node, λ is the L2 regularization weight, and is a regularization parameter.

[0269] The gain score measures the effect of splitting a specific node, and by calculating the information gain before and after splitting a node, we can quantitatively evaluate how much each split contributed to improving the performance of the overall model. In Equation 4, can provide an additional penalty for node splitting as a regularization parameter to prevent overfitting.

[0270] When splitting a tree node, the minimum instance weight required for the child node can be set. For example, the minimum instance weight can be 2, but is not limited to this. The minimum instance weight determines the splitting of nodes during each tree generation process, thereby adjusting the relationship between data characteristics and target values. This can help prevent model overfitting by ensuring the model learns conservatively and avoiding meaningless splits in nodes with limited data.

[0271] XGBoost (eXtreme Gradient Boosting) is a supervised learning algorithm that adds a technique to prevent overfitting to the gradient boosting algorithm. Gradient boosting can minimize the difference between the predicted probability for each data point and the actual label. Gradient boosting is used to train the t-th model F. t (X) can be updated. The t-th model can be as shown in Equation 14.

[0272] [Equation 14]

[0273]

[0274] F in Equation 14t-1 (X) is the sum of gradients at the left and right child nodes, η is the learning rate, and h t (X) is the prediction function of the new tree learned at the current stage.

[0275] The t-th model F in gradient boosting t In the process of updating (X), the learning rate η controls the model update size at each boosting step.

[0276] In the process of minimizing the difference between the predicted probability for each data point and the actual label, the gradient of the loss function at each step, g i can be calculated. The gradient can be as shown in Equation 15.

[0277] [Equation 15]

[0278]

[0279] y in equation 15 i is the actual label value of the ith data point, and F t-1 is the predicted value of the ith data point predicted by the previous model, and g i is the gradient value of the ith data point.

[0280] Based on the gradient of the loss function for each data point, the model can be sequentially updated by adding new trees.

[0281] In summary, the XGBoost model according to one embodiment can be trained through the following steps: 1) setting an initial estimate and building a first tree, 2) calculating residuals to add a second tree to the model, 3) calculating the gradient of a loss function to determine an optimal tree split, and 4) defining the minimum instance weight that a child node in each split should have as 2 to prevent overfitting and ensure accuracy.

[0282] The input step (S130) can input any demographic, health-related, and digital information-related factors as input values. For example, demographic factors may include, but are not limited to, gender, age group, and highest level of education. Health-related factors may include, but are not limited to, health status, exercise, health management proficiency, and health management coping skills. Digital information-related factors may include, but are not limited to, digital device utilization skills, health information search experience, health information utilization skills, and health information reliability judgment skills.

[0283] The output step (S140) can output a result value corresponding to the input value input to the classification model (670). The result value can include a binary classified digital health literacy level.

[0284] The method for predicting a digital health literacy level according to another embodiment described with reference to FIG. 22 can apply the same or similar contents as those described in the digital health literacy level predicting device (600) according to one embodiment described with reference to FIGS. 14 to 21.

[0285] Figure 23 is a configuration diagram of a computing system according to another embodiment.

[0286] Referring to FIG. 23, a computing system may include an input unit, a memory, and a processor. The memory (710) may store any demographic factors, health-related factors, and digital information-related factors (611) input from the input unit (610), but may also be stored separately in a separate large-capacity storage server, etc. The memory (710) may be a volatile memory (e.g., SRAM, DRAM) or a non-volatile memory (e.g., NAND Flash).

[0287] The processor (720) may input any demographic factors, health-related factors, and some digital information-related factors (611) as input values. For example, demographic factors may include, but are not limited to, gender, age group, and highest level of education. Health-related factors may include, but are not limited to, health status, exercise, health management proficiency, and health management coping skills. Digital information-related factors may include, but are not limited to, digital device utilization skills, health information search experience, health information utilization skills, and health information reliability judgment skills.

[0288] The processor (720) can output binary classified digital health literacy levels as output data. The binary classified digital health literacy levels can include a group with low digital health literacy (650) and a group with high digital health literacy (660).

[0289] The processor (720) can output an output value according to an input value using a model that has performed XGBoost-based learning on a learning data set (640) that uses demographic factors, health-related factors, and digital information-related factors (611) as input data (641) and a binary-classified digital health literacy level as output data (642).

[0290] The memory (710) stores a classification model (670) that has performed XGBoost-based learning. When a task of outputting a result value from any demographic factor, health-related factor, and digital information-related factor (611) is requested, the classification model (670) that has performed XGBoost-based learning stored in the memory (710) is executed to output a result value from the input value.

[0291] A computing system according to embodiments of the present invention may include a computer device (700) including an input unit (610), a memory (710), and a processor (720), and a server (800) including a memory (810) and a processor (820). The computer device (700) and the server (800) may be connected wired or wirelessly via a network (900).

[0292] The memory (810) of the server (800) can store an optimization model (670) that has undergone the aforementioned artificial neural network-based learning.

[0293] When a task of predicting a level of digital health literacy from arbitrary demographic factors, health-related factors, and digital information-related factors (611) is requested (queried), the processor (720) of the computer device (700) retrieves arbitrary demographic factors, health-related factors, and digital information-related factors (611) stored in the memory (710). The memory (710) of the computer device (700) can store any of the aforementioned demographic factors, health-related factors, and digital information-related factors (611).

[0294] The processor (720) of the computer device (700) can transmit any demographic factors, health-related factors and digital information-related factors (611) stored in the memory (710) and this request (query) to the server (800).

[0295] The processor (820) of the server (800) can execute a classification model (670) that has performed XGBoost-based learning on any demographic factor, health-related factor, and digital information-related factor (611) received, stored in memory, to predict a digital health literacy level from any demographic factor, health-related factor, and digital information-related factor (611), and transmit the result to a computer device (700).

[0296] Figure 24 is a configuration diagram of a client-server computer system according to another embodiment.

[0297] Various possible examples are described below for the computer system described with reference to FIGS. 23 and 24.

[0298] The digital health literacy level prediction device (600) may be configured as a computing system (800) as shown in FIG. 24, or may be configured as a GPU server equipped with a GPU processor and general memory, but the present invention is not limited thereto.

[0299] The aforementioned digital health literacy level prediction device (600) may be implemented by a computing device including at least some of a processor, a memory, a user input device, and a presentation device. The memory is a medium that stores computer-readable software, applications, program modules, routines, instructions, and / or data, etc., which are coded to perform a specific task when executed by the processor. The processor can read and execute the computer-readable software, applications, program modules, routines, instructions, and / or data stored in the memory. The user input device may be a means for allowing a user to input a command to cause the processor to perform a specific task or to input data necessary for the execution of a specific task. The user input device may include a physical or virtual keyboard or keypad, key buttons, a mouse, a joystick, a trackball, a touch-sensitive input device, or a microphone. The presentation device may include a display, a printer, a speaker, or a vibration device.

[0300] Computing devices can include a variety of devices, including smartphones, tablets, laptops, desktops, servers, and clients. A computing device may be a single, standalone device, or it may include multiple computing devices operating in a distributed environment, each of which collaborates with another through a communications network.

[0301] Meanwhile, computing devices may be quantum computing devices, rather than classical computing devices. Quantum computing devices perform calculations on qubits, not bits. Qubits can be in a superposition of 0 and 1 simultaneously, and with M qubits, they can express 2^M states simultaneously.

[0302] Quantum computing devices can use various types of quantum gates (e.g., Pauli / Rotation / Hadamard / CNOT / SWAP / Toffoli) that input one or more qubits to perform quantum operations and perform designated operations, and can combine quantum gates to form quantum circuits that perform special functions.

[0303] Quantum computing devices can use quantum artificial neural networks (e.g. QCNN, QGRNN) that can perform functions performed by conventional artificial neural networks (e.g. CNN, RNN) at a faster speed while using fewer parameters.

[0304] In addition, the aforementioned digital health literacy level prediction device (600) can be executed by a computing device having a processor and a memory storing computer-readable software, applications, program modules, routines, instructions, and / or data structures coded to perform a digital health literacy level prediction method utilizing a deep learning model when executed by the processor.

[0305] The artificial intelligence model described in these embodiments may be a current or future machine learning model, such as a model that performs algorithm-based machine learning operating on the aforementioned computing device or a model that performs artificial neural network-based learning.

[0306] Models that perform algorithm-based machine learning can be classical machine learning models such as tree-based models, k-Nearest Neighbors, k-Means Clustering, Principal Component Analysis (PCA), and support vector machines (SVM).

[0307] A tree-based model can be, for example, a decision tree model, a regression model, or a random tree model.

[0308] Meanwhile, an artificial intelligence model can be an ensemble model that solves problems by training and combining multiple models rather than using just one trained model.

[0309] Ensemble models combine multiple individually trained models to prevent overfitting and improve generalization performance. Ensemble models can be helpful in improving performance when the performance of individual models is not sufficient.

[0310] Ensemble models can be broadly divided into voting and boosting methods.

[0311] Voting methods derive a final result through voting on the results generated by multiple models. Examples include bagging, which combines algorithms of the same type but trains them on different data sets, and voting, which combines different types of algorithms.

[0312] Boosting is a method of combining weak machine learning models to create a more accurate and powerful model. Boosting involves sequentially performing tasks on each weak machine learning model, with subsequent models exploring additional areas missed by the previous models. Examples of boosting methods include random forests, gradient boosting, and XGBoost (eXtra Gradient Boost).

[0313] An artificial neural network (ANN) is a machine learning algorithm that analyzes and learns complex data based on a large number of interconnected artificial neurons, mimicking the operating principles of the human brain. An ANN can be any type of ANN, including the multilayer perceptron (MLP), the most basic ANN structure consisting of an input layer, a hidden layer, and an output layer; a convolutional neural network (CNN), which performs convolution operations to extract image features and reduces dimensionality through pooling operations; and a recurrent neural network (RNN), an ANN structure used to process ordered data. These ANNs can be modified in various ways depending on the complexity and diversity of the data.

[0314] A model that has undergone learning based on an artificial neural network can also be an ensemble model that solves problems by learning multiple models and combining them rather than learning just one model.

[0315] Meanwhile, algorithm-based machine learning models and models trained using artificial neural networks can be used complementarily. For example, an algorithm-based machine learning model can use the results of an artificial neural network-based model, and vice versa. An ensemble model combining algorithm-based machine learning models and artificial neural network-based models can also be used.

[0316] The embodiments described above may be implemented through various means. For example, the embodiments may be implemented through hardware, firmware, software, or a combination thereof.

[0317] In the case of hardware implementation, the method for predicting digital health literacy level can be implemented by one or more ASICs (Application Specific Integrated Circuits), DSPs (Digital Signal Processors), DSPDs (Digital Signal Processing Devices), PLDs (Programmable Logic Devices), FPGAs (Field Programmable Gate Arrays), processors, controllers, microcontrollers or microprocessors.

[0318] For example, a method for predicting digital health literacy levels can be implemented using an AI semiconductor device, where the neurons and synapses of a deep neural network are implemented using semiconductor devices. These semiconductor devices can be current semiconductor devices such as SRAM, DRAM, or NAND, next-generation semiconductor devices such as RRAM, STT MRAM, or PRAM, or a combination of these.

[0319] When implementing a method for predicting the level of digital health literacy according to embodiments using an artificial intelligence semiconductor device, the results (weights) of learning a deep learning model using software may be transferred to synapse-mimicking elements arranged in an array, or learning may be performed in the artificial intelligence semiconductor device.

[0320] When implemented using firmware or software, the digital health literacy level prediction method according to the present embodiments may be implemented in the form of a device, procedure, or function that performs the functions or operations described above. The software code may be stored in a memory unit and executed by a processor. The memory unit may be located within or outside the processor and may exchange data with the processor via various known means.

[0321] Additionally, terms such as "system," "processor," "controller," "component," "module," "interface," "model," or "unit" as described above may generally refer to a computer-related entity, such as hardware, a combination of hardware and software, software, or software in execution. For example, the aforementioned components may be, but are not limited to, a process driven by a processor, a processor, a controller, a control processor, an object, a thread of execution, a program, and / or a computer. For example, both an application running on a controller or a processor and the controller or the processor may be components. One or more components may be within a process and / or thread of execution, and the components may be located on a single device (e.g., a system, a computing device, etc.) or distributed across two or more devices.

[0322] Meanwhile, another embodiment provides a computer program stored on a computer storage medium that performs a method for predicting digital health literacy levels. Furthermore, another embodiment provides a computer-readable storage medium storing a program for implementing the method for predicting digital health literacy levels.

[0323] The program recorded on the recording medium can be read, installed and executed by a computer, thereby executing the steps described above.

[0324] In this way, in order for a computer to read a program recorded on a recording medium and execute functions implemented as a program, the above-mentioned program may include code coded in a computer language such as C, C++, JAVA, or machine language that can be read by the computer's processor (CPU) through the computer's device interface.

[0325] Such code may include functional code related to functions defining the aforementioned functions, and may also include control code related to execution procedures required for the computer's processor to execute the aforementioned functions according to a predetermined procedure.

[0326] Additionally, such code may further include memory reference related code regarding where in the internal or external memory of the computer the additional information or media required for the computer's processor to execute the aforementioned functions should be referenced.

[0327] Additionally, if the computer's processor needs to communicate with another computer or server located remotely in order to execute the functions described above, the code may further include communication-related code regarding how the computer's processor should communicate with another computer or server located remotely using the computer's communication module, and what information or media should be sent and received during the communication.

[0328] The computer-readable recording medium that records the program as described above includes, for example, ROM, RAM, CD-ROM, magnetic tape, floppy disk, optical media storage device, etc., and may also include one implemented in the form of a carrier wave (e.g., transmission via the Internet).

[0329] Additionally, computer-readable recording media can be distributed across network-connected computer systems, allowing computer-readable code to be stored and executed in a distributed manner.

[0330] In addition, the functional program for implementing the present invention and the code and code segments related thereto may be easily inferred or changed by programmers in the technical field to which the present invention belongs, taking into consideration the system environment of the computer that reads the recording medium and executes the program.

[0331] The method for predicting the level of digital health literacy described through FIG. 22 may also be implemented in the form of a recording medium containing computer-executable instructions, such as an application or program module executed by a computer. The computer-readable medium may be any available medium that can be accessed by a computer, and includes both volatile and nonvolatile media, removable and non-removable media. Furthermore, the computer-readable medium may include all computer storage media. The computer storage media includes both volatile and nonvolatile, removable and non-removable media implemented in any method or technology for storing information, such as computer-readable instructions, data structures, program modules, or other data.

[0332] Figure 25 is a block diagram of a digital health literacy-based content service device according to another embodiment.

[0333] Referring to FIG. 25, a digital health literacy-based content service device (1000) according to another embodiment includes an input unit (1100), a user evaluation unit (1200), a first content provision unit (1300), a content extraction unit (1400), and a second content provision unit (1500).

[0334] The input unit (1100) can receive user information and interaction information from the user terminal.

[0335] The user evaluation unit (1200) can evaluate the user's digital health literacy level using the user's information received according to the first learning model that has been learned in advance.

[0336] The first content provider (1300) can select digital health literacy support content corresponding to the evaluation results of the user evaluation provider (1200) and provide it to the user.

[0337] The content extraction unit (1400) can extract health information content to be recommended to the user by using the user information and interaction information received according to the second learning model learned in advance.

[0338] Here, the first and second learning models may use different learning algorithms. Specifically, the first learning model may use the user's digital health literacy assessment results corresponding to the user's demographic data, health status data, and digital literacy data as a learning data set, while the second learning model may use health information content results corresponding to the user's demographic data, health status data, and interaction data as a learning data set.

[0339] The second content provider (1500) can provide extracted health information content to the user.

[0340] FIG. 26 is a flowchart of a digital health literacy-based content service method according to another embodiment, and FIG. 27 illustrates the digital health literacy-based content service of FIG. 26.

[0341] Referring to FIGS. 26 and 27, a digital health literacy-based content service method according to another embodiment includes a receiving step (S210), an evaluation step (S220), a first content provision step (S230), an extraction step (S240), and a second content provision step (S250).

[0342] In the receiving step (S210), user information and interaction information can be received from the user terminal. For example, after a user installs the app and completes membership registration, user inflow is achieved. Then, information such as demographic information, health status, diagnosed medical conditions, drinking and smoking habits, exercise habits, and interaction information can be received from the user terminal.

[0343] In the evaluation step (S220), the user's digital health literacy level can be assessed using the received user information based on the pre-trained first learning model. For example, the received user information is input into the first learning model, which can classify and evaluate the user's digital health literacy level based on factors such as device utilization ability, information search ability, reliability assessment ability, and information communication ability. At this time, the user's digital device usage level and information utilization tendencies may also be considered.

[0344] In the first content provision stage (S230), digital health literacy support content corresponding to the assessment results can be selected and provided to users. For example, users with low literacy levels can be provided with content on basic device usage and information retrieval methods, while users with high literacy levels can be provided with more advanced health information analysis and utilization content.

[0345] In the extraction step (S240), health information content to be recommended to the user can be extracted using the received user information and interaction data based on a pre-trained second learning model. For example, based on the pre-trained second learning model, the received user information and interaction data can be analyzed to extract health information content suitable for the user (e.g., disease understanding and treatment, health promotion and prevention, medical services and policies, hospital information, etc.). Disease characteristics such as hypertension, diabetes, depression, and sleep disorders can also be considered.

[0346] In the second content provision stage (S250), extracted health information content can be provided to users, and users can manage their health and utilize information more efficiently through customized health information content.

[0347] The aforementioned method for predicting a digital health literacy level can be executed by an application that is installed by default on a terminal (which may include a program included in a platform or operating system installed by default on the terminal), or by an application (i.e., a program) that a user directly installs on a master terminal via an application providing server, such as an application store server, an application, or a web server related to the service. In this sense, the aforementioned method for predicting a digital health literacy level can be implemented as an application (i.e., a program) that is installed by default on a terminal or directly installed by a user, and can be recorded on a computer-readable recording medium such as on the terminal.

[0348] The foregoing description of the present invention is for illustrative purposes only, and those skilled in the art will readily appreciate that the present invention can be readily modified into other specific forms without altering the technical spirit or essential characteristics of the present invention. Therefore, the embodiments described above should be understood as illustrative in all respects and not restrictive. For example, each component described as a single entity may be implemented in a distributed manner, and similarly, components described as distributed may be implemented in a combined manner.

[0349] The scope of the present invention is indicated by the claims described below rather than the detailed description above, and all changes or modifications derived from the meaning and scope of the claims and their equivalent concepts should be interpreted as being included in the scope of the present invention.

[0350] The above description is merely an illustrative example of the technical idea of ​​the present disclosure, and those skilled in the art to which the present disclosure pertains will appreciate that various modifications and variations can be made without departing from the essential characteristics of the technical idea of ​​the present disclosure. In addition, the present embodiments are not intended to limit the technical idea of ​​the present disclosure but rather to explain it, and therefore the scope of the technical idea of ​​the present disclosure is not limited by these embodiments. The scope of protection of the present disclosure should be interpreted by the claims below, and all technical ideas within a scope equivalent thereto should be interpreted as being included within the scope of the rights of the present disclosure.

[0351]

[0352] CROSS-REFERENCE TO RELATED APPLICATION

[0353] This patent application claims priority under 35 USC §119(a) to Korean Patent Application No. 10-2024-0088889, filed in Korea on July 5, 2024, and Korean Patent Application No. 10-2024-0186476, filed in Korea on December 13, 2024, the entire contents of which are incorporated herein by reference. In addition, this patent application claims priority in countries other than the United States for the same reasons, the entire contents of which are incorporated herein by reference.

Claims

1. An input unit that receives user information and interaction information from a user terminal; A user evaluation unit that evaluates the user's digital health literacy level using the received user information according to a pre-learned first learning model; A first content provision unit that selects digital health literacy support content corresponding to the evaluation results of the user evaluation unit and provides the content to the user; A content extraction unit that extracts health information content to be recommended to the user by using the received user information and interaction information according to the pre-learned second learning model; and A digital health literacy-based content service device including a second content provision unit that provides the extracted health information content to a user.

2. In paragraph 1, A digital health literacy-based content service device in which the first learning model and the second learning model use different learning algorithms.

3. In paragraph 2, The above first learning model uses the user's digital health literacy assessment result data corresponding to the user's demographic data, health status data, and digital literacy data as a learning data set, The second learning model is a digital health literacy-based content service device that uses health information content result data corresponding to the user's demographic data, health status data, and user interaction data as a learning data set.

4. A receiving step for receiving user information and interaction information from a user terminal; An evaluation step of evaluating the user's digital health literacy level using the received user information according to the first learning model that has been learned in advance; A first content provision step of selecting digital health literacy support content corresponding to the above evaluation results and providing it to the user; An extraction step for extracting health information content to be recommended to a user by using the received user information and interaction information according to a pre-learned second learning model; and A digital health literacy-based content service method including a second content provision step of providing the extracted health information content to a user.

5. In paragraph 4, A digital health literacy-based content service method wherein the first learning model and the second learning model use different learning algorithms.

6. In paragraph 5, The above first learning model uses the user's digital health literacy assessment result data corresponding to the user's demographic data, health status data, and digital literacy data as a learning data set, The second learning model is a digital health literacy-based content service method that uses health information content result data corresponding to the user's demographic data, health status data, and user interaction data as a learning data set.

7. Input section for receiving metadata from any user; and A digital health literacy-based content service device including a processing unit that recommends user health information for any user's metadata using a recommendation model learned to generate user health information for the user's metadata and the user's interaction data using learning data consisting of input data including the user's metadata and the user's interaction data and output data including the user's health information corresponding to the user's metadata and the user's interaction data.

8. In paragraph 7, The metadata of the user includes at least one of demographic information including at least one of gender, age, place of residence, marital status, and subjective economic status, and information on diseases or lifestyle habits including at least one of diagnosed diseases, smoking, drinking, exercise, snacking, and preference for salty foods. The above user interaction data includes health information content including preferences, The health information of the user includes at least one of disease-related information, health promotion information, hospital use information, and medical service policy information. A digital health literacy-based content service device wherein the disease-related information includes at least one of hypertension, diabetes, depression, or sleep disorder.

9. In paragraph 7, The above recommended model is, The user's interaction data and the user's metadata are inputted, and the user's interaction data is a latent variable (Z) of the interaction data. i ) and map the user's metadata to the latent variable (Z) of the metadata. m ) encoder mapping; Mapping matrix (W map ) using the metadata-based latent variable (Z) obtained from the above user metadata. m ) through the latent variable of interaction data ( ) mapping network to predict; and Latent variables in the predicted interaction data ( ) A digital health literacy-based content service device including a decoder that generates reconstructed health information.

10. In paragraph 9, A digital health literacy-based content service device in which the above recommendation model is trained to minimize a final loss function including reconstruction loss of the user metadata, reconstruction loss of the user interaction data, and regularization loss of the mapping network.

11. In paragraph 10, The above encoder converts the user's interaction data into a latent variable (Z) of the interaction data. i ) and the first encoder that maps the user's metadata to the latent variable (Z) of the metadata. m ) and includes a second encoder that maps to The above recommendation model converts the user's interaction data into a latent variable (Z) of the interaction data through the first encoder and the second encoder during learning. i ) and map the user's metadata to the latent variable (Z) of the metadata. m ) and map it to When recommending health information for a user based on the metadata of the arbitrary user, the arbitrary user metadata is converted into a latent variable (Z) of the metadata through the second encoder. m ) and the mapping matrix (W) of the mapping network map ) using the above arbitrary user metadata-based latent variable (Z m ) through the latent variable of interaction data ( ) A digital health literacy-based content service device that predicts.

12. In paragraph 11, The above learned recommendation model is retrained by receiving health information of users selected by actual users. The above processing unit is a digital health literacy-based content service device that recommends health information for the arbitrary user metadata based on the relearned recommendation model.

13. Input step for receiving metadata of any user; and A digital health literacy-based content service method comprising a processing step of recommending user health information for any user's metadata using a recommendation model learned to generate user health information for the user's metadata and the user's interaction data using learning data consisting of input data including the user's metadata and the user's interaction data and output data including the user's health information corresponding to the user's metadata and the user's interaction data.

14. In paragraph 13, The metadata of the user includes at least one of demographic information including at least one of gender, age, place of residence, marital status, and subjective economic status, and information on diseases or lifestyle habits including at least one of diagnosed diseases, smoking, drinking, exercise, snacking, and preference for salty foods. The above user interaction data includes health information content including preferences, The health information of the user includes at least one of disease-related information, health promotion information, hospital use information, and medical service policy information. A method for providing digital health literacy-based content, wherein the disease-related information includes at least one of hypertension, diabetes, depression, or sleep disorder.

15. In paragraph 13, The above recommended model is, The user's interaction data and the user's metadata are inputted, and the user's interaction data is a latent variable (Z) of the interaction data. i ) and map the user's metadata to the latent variable (Z) of the metadata. m ) encoder mapping; Mapping matrix (W map ) using the metadata-based latent variable (Z) obtained from the above user metadata. m ) through the latent variable of interaction data ( ) mapping network to predict; and Latent variables in the predicted interaction data ( ) A method for providing content services based on digital health literacy, including a decoder that generates reconstructed health information.

16. In paragraph 15, A method for providing content services based on digital health literacy, wherein the recommendation model is trained to minimize a final loss function including reconstruction loss of the user metadata, reconstruction loss of the user interaction data, and regularization loss of the mapping network.

17. In paragraph 16, The above encoder converts the user's interaction data into a latent variable (Z) of the interaction data. i ) and the first encoder that maps the user's metadata to the latent variable (Z) of the metadata. m ) and includes a second encoder that maps to The above recommendation model converts the user's interaction data into a latent variable (Z) of the interaction data through the first encoder and the second encoder during learning. i ) and map the user's metadata to the latent variable (Z) of the metadata. m ) and map it to When recommending health information for a user based on the metadata of the arbitrary user, the arbitrary user metadata is converted into a latent variable (Z) of the metadata through the second encoder. m ) and the mapping matrix (W) of the mapping network map ) using the above arbitrary user metadata-based latent variable (Z m ) through the latent variable of interaction data ( ) A method for providing content services based on digital health literacy.

18. In paragraph 13, The above learned recommendation model is retrained by receiving health information of users selected by actual users. The above processing step is a digital health literacy-based content service method for recommending health information for the arbitrary user metadata based on the relearned recommendation model.

19. An input section that receives arbitrary demographic factors, health-related factors, and digital information-related factors as input values; and A digital health literacy-based content service device including a processing unit that classifies a digital health literacy level from any of the demographic factors, health-related factors, and digital information-related factors using a classification model trained using a learning data set that uses demographic factors, health-related factors, and digital information-related factors as input data and a binary-classified digital health literacy level as output data.

20. In paragraph 19, The above demographic factors include gender, age group, and highest level of education. The above health-related factors include health status, exercise, health management skills, and health management coping ability. The above digital information-related factors include digital health literacy-based content service devices including digital device utilization ability, health information search experience, health information utilization ability, and health information reliability judgment ability.

21. In paragraph 19, The above digital health literacy level is a digital health literacy-based content service device that binary classifies groups into low and high digital health literacy groups.

22. In paragraph 19, The above classification model is a digital health literacy-based content service device that is an XGBoost model.

23. In paragraph 22, The above XGBoost model includes an objective function, The above objective function is a digital health literacy-based content service device including a loss function and a regularization term.

24. In paragraph 23, The above loss function is a digital health literacy-based content service device that is a logistic loss function.

25. In paragraph 23, The above normalization term is a digital health literacy-based content service device that applies L2 normalization.

26. In paragraph 19, The above output data is a digital health literacy-based content service device generated based on eHEALS, a self-report scale measuring digital health literacy.

27. A collection step for creating a learning data set that uses demographic factors, health-related factors, and digital information-related factors as input data and binary classified digital health literacy levels as output data; A learning step for learning a classification model that predicts the level of digital health literacy using the above learning data set; An input step of inputting arbitrary demographic factors, health-related factors, and digital information-related factors as input values ​​into the classification model; and A digital health literacy-based content service method including an output step for outputting a binary classified digital health literacy level corresponding to the above input value as a result value.

28. In paragraph 27, The above demographic factors include gender, age group, and highest level of education. The above health-related factors include health status, exercise, health management skills, and health management coping ability. The above digital information-related factors include digital device utilization ability, health information search experience, health information utilization ability, and health information reliability judgment ability, and are a content service method based on digital health literacy.

29. In paragraph 27, The above binary classified digital health literacy level is a digital health literacy-based content service method in which groups with low digital health literacy and groups with high digital health literacy are classified.

30. In paragraph 27, The above classification model is a digital health literacy-based content service method using an XGBoost model.

31. In paragraph 30, The above XGBoost model includes an objective function, A method for providing content services based on digital health literacy, wherein the objective function includes a loss function and a regularization term.

32. In paragraph 31, The above loss function is a digital health literacy-based content service method that is a logistic loss function.

33. In paragraph 31, The above normalization term is a digital health literacy-based content service method that applies L2 normalization.

34. In paragraph 27, The above output data is a digital health literacy-based content service method generated based on eHEALS, a self-report scale measuring digital health literacy.

35. A memory storing arbitrary demographic factors, health-related factors, and digital information-related factors, and a classification model, wherein the classification model is a model that performs machine learning by using arbitrary demographic factors, health-related factors, and digital information-related factors as input data and using the level of digital health literacy according to the arbitrary demographic factors, health-related factors, and digital information-related factors as output data; and A computer device including a processor that, when a task of predicting a level of digital health literacy is requested, executes the classification model that has performed the machine learning stored in the memory and outputs a level of digital health literacy from arbitrary demographic factors, health-related factors, and digital information-related factors.

Images