Method and apparatus for providing user-personalized smoking cessation digital therapy based on multi ai agents

Abstract

The present invention relates to a digital smoking cessation treatment method and device that analyzes a user's behavioral, psychological, biological, and environmental data in real time through multiple AI agents to predict the likelihood of smoking cessation failure and performs personalized intervention based on the results. A device for providing digital smoking cessation therapy includes: an AI-based conversational interface module that interacts with a user; a data collection module that collects data from said interaction and related data affecting the user's smoking in real time; a multi-agent analysis module that analyzes said collected data through a plurality of AI agents; a risk prediction module that predicts the likelihood of smoking cessation failure based on the analysis results; and an intervention execution module that provides personalized intervention to the user according to the prediction results. According to the present invention, the risk of smoking cessation failure can be predicted early by collecting in real time not only data from interactions with the user but also various relevant data affecting the user's smoking, and analyzing them in parallel through multiple AI agents. Furthermore, since customized interventions are automatically executed based on the prediction results, active digital therapy that responds rapidly to changes in the user's condition is possible. In addition, the prediction accuracy and intervention efficiency are gradually improved by retraining the intervention results using feedback data.

Description

Technology Field

[0001] The present invention relates to a digital smoking cessation treatment method and apparatus. More specifically, it relates to a digital smoking cessation treatment method and apparatus that predicts the likelihood of smoking cessation failure by analyzing a user's behavior, psychological, biological, and environmental data in real time through multiple AI agents, and performs personalized intervention based on the results. Background Technology

[0003] Smoking is known to be a major cause of various diseases, including cardiovascular disease, cancer, and lung disease. Accordingly, the need for digital therapeutics (DTx) to aid in quitting smoking is increasing.

[0004] Existing digital smoking cessation therapeutics have primarily relied on methods such as surveys, educational content, and push notifications to support users in quitting smoking. However, these systems struggled to reflect users' actual life contexts or emotional changes in real time, and their capabilities to detect or respond to moments when the risk of failure increased were limited.

[0005] In particular, existing chatbot-based systems operate according to predefined rules or scenarios, making it difficult to respond immediately to complex factors such as the user's emotional state, environmental changes, and stress levels.

[0006] Therefore, there is a growing need for an intelligent system capable of collecting and analyzing data in real time to predict the likelihood of smoking cessation failure and providing customized interventions based on the results. The inventors have developed a digital smoking cessation treatment device and method equipped with the aforementioned intelligent system to complete the present invention. Prior art literature

[0008] Korean Patent Publication No. 10-2025-0069010, (Publication Date: May 19, 2025) The problem to be solved

[0009] The present invention aims to solve the aforementioned problems by providing a technology that increases the success and retention rates of quitting smoking by collecting in real time not only data from interactions with the user but also various relevant data affecting the user's smoking, having multiple AI agents analyze this data in parallel to accurately predict the likelihood of failure, and immediately providing personalized interventions based on the prediction results.

[0010] In addition, another objective is to provide intelligent digital smoking cessation treatment technology that improves personalization precision with long-term use by implementing a closed-loop structure that continuously improves predictive models and intervention strategies by reflecting user responses after intervention as retraining data.

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

[0013] According to an embodiment of the present invention, a smoking cessation digital treatment providing device comprises: an artificial intelligence-based conversational interface module that interacts with a user; a data collection module that collects data from the interaction and related data affecting the user's smoking in real time; a multi-agent analysis module that analyzes the collected data through a plurality of AI agents; a risk prediction module that predicts the likelihood of smoking cessation failure based on the analysis results; and an intervention execution module that provides personalized intervention to the user according to the prediction results.

[0014] The above-mentioned related data may include at least one of conversation data, app usage data, behavior and location data, biometric data, environmental data, and social and schedule data. The above-mentioned conversation data may include at least one of chatbot conversation content, question and answer, emotional expressions, and language patterns. The above-mentioned app usage data may include at least one of usage history of health-related apps, including meditation, breathing, and exercise apps, content viewing records, survey responses, and app access frequency. The above-mentioned behavior and location data may include at least one of the user's movement path, whether they visited location-based smoking areas, activity level, sleep patterns, and access history by time of day. The above-mentioned biometric data may include at least one of heart rate, respiratory rate, body temperature, skin conductivity, blood oxygen saturation, step count, and sleep quality. The above-mentioned environmental data may include at least one of temperature, humidity, fine dust, illuminance, noise, air quality, and weather. The above-mentioned social and schedule data may include at least one of SNS activity records, call and message frequency, schedule calendar information, and social event participation logs.

[0015] The multi-agent analysis module described above may include at least two of a behavioral analysis agent, a psychological analysis agent, an app history analysis agent, and an environmental perception agent. The behavioral analysis agent may analyze at least one of the user's smoking craving records, location history, activity sensor data, and behavioral frequency by time of day. The psychological analysis agent may detect emotional signals within the interaction to determine a risk emotional state. The environmental perception agent may calculate external factors affecting smoking cessation risk using at least one of weather, air quality, noise, illuminance, schedule, and social activity logs.

[0016] The above prediction of the likelihood of smoking cessation failure may utilize one or more of machine learning, deep learning, reinforcement learning, time series forecasting, graph neural networks (GNN), survival analysis, Bayesian inference, or outlier detection techniques. The above personalized intervention may include one or more of modifying smoking cessation goals, providing reminders, recommending personalized content, sending notifications to experts or family members, providing rewards, and suggesting alternative behaviors. The above intervention may be delivered to the user via mobile app notifications, text messages, messenger notifications, or wearable device notifications. A smoking cessation digital treatment delivery device according to an embodiment of the present invention may further include a retraining module that collects user responses and service usage logs after the intervention to retrain the prediction model and intervention strategy.

[0017] Probability of smoking cessation failure R based on the analysis results of the above multi-agent analysis module t a user-based threshold When more than or equal to margin It is configured to execute intervention when is above a set value, and the above R t is calculated using the following formula,

[0018]

[0019] Here, B t is the normalized score (0~1) of the behavior analysis agent, P t is the normalized score (0~1) of the psychoanalysis agent, A t is the normalization score (0~1) of the app history analysis agent, E t is the normalized score (0~1) of the environment perception agent, Δ t is the exponentially weighted moving average (EWMA) trend of the recent τ interval, σ(ㆍ) is the sigmoid function, and is the weight learned and stored in the model, and the user-based threshold is as, is the system default, H u is an indicator of user-specific habits and dependency, C tis the current motivation and compliance indicator, and λ,κ are correction coefficients.

[0021] According to another embodiment of the present invention, a method for providing digital smoking cessation treatment comprises the steps of: interacting with a user through an artificial intelligence-based conversational interface module; collecting data from the interaction and relevant data affecting the user's smoking in real time; analyzing the collected data through a plurality of AI agents; predicting the likelihood of the user failing to quit smoking based on the analysis results; and providing personalized intervention to the user according to the prediction results. The relevant data may include at least one of app usage data, behavioral and location data, biometric data, environmental data, and social and schedule data. The plurality of AI agents may include at least two of a behavioral analysis agent, a psychological analysis agent, an app history analysis agent, and an environmental perception agent. Effects of the invention

[0023] According to the present invention, by collecting in real-time data from interactions with the user as well as various relevant data affecting the user's smoking, and analyzing them in parallel through multiple AI agents, the risk or likelihood of smoking cessation failure can be predicted early and accurately. Furthermore, since user-customized interventions are automatically executed based on the prediction results, active digital therapy is possible to rapidly respond to changes in the user's condition and increase the success and retention rates of smoking cessation. Additionally, the prediction accuracy and intervention efficiency are gradually improved by retraining the intervention results using feedback data.

[0024] Therefore, the present invention is significantly superior to existing rule-based smoking cessation treatment systems in terms of real-time capability, level of personalization, and prediction accuracy, and has the characteristic that the treatment effect becomes more advanced the longer the user is used. Brief explanation of the drawing

[0026] FIG. 1 is a schematic diagram showing the data processing flow between a user terminal and a server according to an embodiment of the present invention. FIG. 2 is a block diagram showing the configuration of a device that provides a user-customized digital smoking cessation treatment based on a plurality of AI agents according to an embodiment of the present invention. FIG. 3 is a block diagram illustrating the detailed configuration of an agent analysis module and the integrated structure of a supervisor agent according to an embodiment of the present invention. FIG. 4 is a block diagram showing the detailed configuration of a risk prediction module according to an embodiment of the present invention. FIG. 5 is a block diagram showing the configuration of a retraining module and a database for updating a user feedback-based model and strategy according to an embodiment of the present invention. FIG. 6 illustrates a flowchart of a method for providing a user-customized digital smoking cessation treatment based on a plurality of AI agents according to an embodiment of the present invention. Specific details for implementing the invention

[0027] The aforementioned objects, features, and advantages of the present invention will become more apparent from the following detailed description in conjunction with the accompanying drawings. However, as the present invention is subject to various modifications and may have various embodiments, specific embodiments are illustrated in the drawings and described in detail below.

[0028] Throughout the entire invention, identical reference numerals indicate identical components in principle. Additionally, functional components with identical appearance in the drawings of each embodiment are described using the same reference numeral, and redundant descriptions thereof are omitted.

[0029] Unless specifically stated or evident from the context, the term “about” in relation to a number may be understood to mean the mentioned number and up to + / -10% of that number, and the term “about” in relation to a numerical range may be understood to mean a range from a number 10% lower than the lower limit of the numerical range to a number 10% higher than the upper limit of the numerical range.

[0030] In the following embodiments, terms such as "comprising" or "having" mean that the features or components described in the specification are present, and do not preclude the possibility that one or more other features or components may be added.

[0031] In the drawings, the size of components may be exaggerated or reduced for convenience of explanation. For example, the size and thickness of each component shown in the drawings are arbitrarily depicted for convenience of explanation, and the present invention is not necessarily limited to what is illustrated.

[0032] FIG. 1 is a schematic diagram showing the data processing flow between a user terminal and a server according to an embodiment of the present invention. Referring to FIG. 1, the system of the present invention may be configured so that a server (100) and a user terminal (200) communicate through a network.

[0033] The above network may be one or more of wired or wireless communication networks, and may include, for example, the Internet, intranet, local area network (LAN), wide area network (WAN), mobile communication network (3G, LTE, 5G, etc.), Wi-Fi, Bluetooth, near-field communication (NFC), satellite communication network, etc.

[0034] The above server and user terminal can perform data transmission and reception, command transmission, model updates, service requests and responses, etc. on a cloud infrastructure or edge computing environment.

[0035] This communication is based on the TCP / IP protocol, and various application layer protocols such as HTTP, HTTPS, MQTT, and WebSocket may be used as needed. The present invention is not limited to the form of the network or the communication protocol, and can utilize all electronic communication means including future next-generation communication technologies (e.g., 6G, IoT dedicated networks, etc.).

[0036] The server (100) may include one or more computing resources (e.g., cloud server, on-premises server) and may be responsible for centralized processing such as data storage and execution of high-performance AI models. The user terminal (200) may be any electronic device capable of network access and information processing, such as a smartphone, tablet, or wearable device.

[0037] The smoking cessation digital treatment providing device (110) described with reference to FIG. 2 is implemented on the server (100) in one embodiment and can perform core functions of data processing and AI analysis.

[0038] However, the present invention is not limited to the implementation of the server (100). In other embodiments, all or part of the functional modules of the device (110) may be implemented as a dedicated application (app) or other possible forms of software modules installed on a user terminal (200). This may have the advantage of reducing communication delay with the server and providing an on-device processing environment for processing user personal data within the terminal.

[0039] FIG. 2 is a block diagram illustrating the interoperability structure between hardware and modules of a smoking cessation digital treatment providing device (110) according to an embodiment of the present invention. The device includes a conversation interface module (10), a data collection module (20), a multi-agent analysis module (30), a risk prediction module (40), an intervention execution module (50), a relearning module (60), and a control unit (70). Each module may be implemented as a software configuration executed on a processor and memory within a server (100), and some or all of the configurations may be placed on a user terminal (200).

[0040] The conversation interface module (10) is configured as an AI chatbot or voice assistant for interacting with the user, receives text or voice input, and generates a response in the form of natural language.

[0041] The data collection module (20) collects various relevant data that affect the user's smoking in real time and stores them in a standardized data format.

[0042] The collected data may include data of the following categories and may be used by one or more analysis modules of the multi-agent analysis module (30):

[0043] Conversation data includes chatbot conversation content, questions and answers, emotional expressions, language patterns, etc., and is primarily used by a psychological analysis agent (22) for estimating emotional and emotional states and detecting dangerous emotions, and is also used by a behavioral analysis agent (21) for auxiliary purposes when detecting mentions of craving and clues of dangerous situations during the conversation flow.

[0044] App usage data includes app access frequency, content usage history, survey responses, mission performance history, goal setting and change records, etc., and is primarily used by an app record analysis agent (23) for compliance evaluation and detection of abnormal usage patterns, and is also used by a behavior analysis agent (21) for auxiliary estimation of behavior rhythms based on time of day and usage intensity.

[0045] Behavioral and location data include the time, place, cause, and intensity of smoking cravings, movement paths, access locations, and activity levels, and are primarily used by a behavioral analysis agent (21) to predict smoking possibilities based on time and place, and are used by an environment perception agent (24) as auxiliary inputs for combined analysis with environmental conditions of a specific location and time period.

[0046] Biometric data includes heart rate, respiratory rate, sleep indicators, step count, skin conductivity, body temperature, oxygen saturation, etc., and is used by a behavioral analysis agent (21) primarily for estimating physiological correlations of smoking cessation-related behavioral patterns, and is used by a psychological analysis agent (22) as an auxiliary input for correcting emotional states such as stress and fatigue and improving reliability.

[0047] Environmental data includes external environmental variables such as temperature, humidity, fine dust, noise, illuminance, and weather, and is primarily used by an environmental perception agent (24) to calculate an indicator of the impact of external factors on smoking cessation risk, and is also used by a behavior analysis agent (21) to assist in correcting the correlation between specific environmental changes and smoking-inducing behaviors.

[0048] Social and schedule data include schedule calendars, meeting and work schedules, call and message frequencies, SNS activity logs, etc., and are used primarily by a psychological analysis agent (22) and an environment perception agent (24) to estimate emotional and stress states based on social context and to reflect situational factors, and are used auxiliaryly by an app record analysis agent (23) for correlation analysis of social activity volume and app usage compliance.

[0049] As described above, each data point is primarily processed by the agent corresponding to the primary analysis purpose, while other agents can utilize it as auxiliary input for context correction, cross-validation, and feature enhancement. Through this multi-input, multi-agent structure, integrated features are generated, and the supervisor agent provides the integrated results to the risk prediction module to improve the accuracy of personalized intervention decisions.

[0050] Although not illustrated in the drawings, the device according to one embodiment of the present invention may further include a data mapping module for automatically classifying and transmitting data between the data collection module (20) and the multi-agent analysis module (30). The data mapping module classifies collected user data according to data type, source, format, and time information, and manages each data so that it is transmitted to an AI agent suitable for the purpose of analysis. For example, conversation data is automatically mapped to a psychological analysis agent (32), app usage data to an app history analysis agent (33), and environmental data to an environment perception agent (34). The module optimizes the data flow using one or more of metadata tagging, keyword-based routing, and machine learning classification models, and may include an error verification function to prevent data omission or duplicate transmission. In addition, the processing efficiency and real-time capability of the system can be improved by dynamically adjusting the mapping policy considering the processing load and data priority of each agent.

[0051] The multi-agent analysis module (30) analyzes collected data sequentially or in parallel through multiple AI agents. The multi-agent analysis module (30) includes a behavior analysis agent (31), a psychological analysis agent (32), an app history analysis agent (33), and an environment perception agent (34), and the results of these are integrated by a supervisor agent (35). Each agent receives different data types as input and produces feature vectors, and the supervisor agent (35) integrates them and transmits them to the risk prediction module (40). Through this multi-analysis structure, contextual interpretation of the user state is possible, and higher prediction accuracy can be secured than with a single model.

[0052] The risk prediction module (40) receives the result of the multi-agent analysis module (30) as input, calculates the probability of smoking cessation failure, and sends a trigger signal to the intervention execution module (50) if the result exceeds a user threshold. The intervention execution module (50) performs interventions such as notifications, reminders, reward provision, and alternative behavior suggestions, and the retraining module (60) collects user response data to update the prediction model.

[0053] The control and communication of all these modules are handled by the control unit (70). The control unit (70) may be implemented as a CPU or a similar device depending on hardware, software, or a combination thereof. In terms of hardware, it may be provided in the form of an electronic circuit that processes electrical signals to perform control functions, and in terms of software, it may be provided in the form of a program or code that drives the hardware circuit.

[0054] This structure efficiently maintains data flow between modules and provides scalability that allows each component to be updated independently. Furthermore, it is designed to enable parallel processing and model training in both cloud and edge environments, making it applicable to various digital therapeutic service environments.

[0055] FIG. 3 is a diagram illustrating the internal configuration and operation process of a multi-agent analysis module (30) according to an embodiment of the present invention. This module is a core component for analyzing user-related data input from a conversational interface module (10) and a data collection module (20) in parallel, and includes a plurality of AI agents having different analysis purposes. The multi-agent analysis module (30) includes a behavioral analysis agent (31), a psychological analysis agent (32), an app history analysis agent (33), an environment perception agent (34), and a supervisor agent (35) that integrates the results thereof.

[0056] Each agent of the multi-agent analysis module (30) operates in parallel via an internal bus or data pipeline, and the analysis results of each agent are normalized into feature vectors of a common format and transmitted to the supervisor agent (35). Conversation data, app usage data, behavior and location data, biometric data, environmental data, social and schedule data, etc., which are streamed in real-time from the data collection module (20) are input, and after undergoing time synchronization, missing value correction, outlier removal, and standardization processes in the preprocessing engine, they are distributed to analysis agents (31-34) according to each analysis purpose. This parallel structure ensures both analysis speed and accuracy simultaneously and provides fault tolerance that maintains the entire analysis function even if some data is missing.

[0057] The behavior analysis agent (31) uses one or more of a time series prediction model, LSTM (Long Short-Term Memory), and Temporal Convolutional Network (TCN) to analyze smoking-related behavior patterns. It calculates the probability of a smoking craving occurring at a specific time and place by using the user's smoking craving history, location history, activity sensor data, and behavior frequency by time period as primary inputs.

[0058] In addition, the behavior analysis agent (31) primarily uses behavior and location data and biometric data, and can use app usage data and conversation data, such as smoking-related speech content, as a supplement.

[0059] That is, the user's movement path, activity level, and time of day when craving occurs are used as the primary inputs of this agent, and are combined with environmental factors provided by the environment perception agent (34) to estimate smoking-inducing patterns under specific conditions. The model can be calibrated using personalization parameters (Hu) to reflect differences in lifestyle rhythms among users, and a behavioral policy that minimizes the possibility of smoking can be learned by applying a reinforcement learning-based policy network.

[0060] The psychological analysis agent (32) detects emotional states from the user's conversation and language patterns. Transformer-family models (BERT, KoBERT, RoBERTa) are used to vectorize tone, word choice, context, and emotional expressions, and to calculate an emotional score (Pt). If necessary, voice signal analysis can be combined to improve the accuracy of emotion detection, and the results are classified into 'stable', 'caution', and 'danger' states.

[0061] The psychological analysis agent (32) uses conversation data as the primary input and uses social and schedule data and biometric data as auxiliary inputs to correct the reliability of the emotional state. For example, if expressions of anxiety or anger are detected in the conversation, the agent can comprehensively determine the dangerous emotional state by considering changes in biometric signals (increase in heart rate, increase in skin conductivity, etc.) and social schedules (heavy workload, meetings, etc.). Additionally, regional language models or user-customized fine-tuning models can be applied.

[0062] The app history analysis agent (33) detects abnormal behavior patterns by analyzing various activity data, such as sleep, step count, meditation content usage history, and survey responses, in addition to the smoking cessation app usage logs. Since this data reflects the user's lifestyle rhythm and level of will to quit smoking, it is used as a key indicator for evaluating compliance.

[0063] The app history analysis agent (33) primarily uses app usage data and can use behavior and location data and social and schedule data as auxiliary data. For example, if the frequency of app access decreases and the amount of social activity decreases sharply at the same time, or if the time spent near a specific location (a place where smoking is allowed) increases, it can be determined as a sign of decreased compliance.

[0064] This agent analyzes the transition probabilities of user behavior sequences and calculates the compliance metric (Ct) using Sequence-to-Sequence models, Hidden Markov Models (HMMs), or Transformer-based Self-Attention networks. It detects abnormal patterns such as discontinuing app usage, failing to respond to notifications, drastically reducing sleep duration, decreasing step counts, and ceasing to use meditation content to adjust risk scores, and can evaluate changes in individual users' stress and motivation by combining this with survey response results.

[0065] The environment perception agent (34) evaluates the impact of external factors (weather, air quality, noise, light intensity, schedule, social activity logs, etc.) on the will to quit smoking. By normalizing external API and sensor data, the environmental impact indicator (E t It produces ), which can be expressed by the following formula.

[0066]

[0067] (However, W t : Weather, A t : Air quality, S t : Noise, L t : Illuminance, C t : Social activity intensity)

[0069] The environment perception agent (34) primarily uses environmental data and also uses social and schedule data to calculate the actual impact of external factors on the risk of quitting smoking. In addition, behavioral and location data can be used secondarily to analyze the correlation between specific environmental conditions (e.g., nighttime, drinking location, proximity to smoking area, etc.) and smoking-inducing behaviors.

[0070] The environmental impact indicators calculated in this way are combined with the results of behavioral analysis and psychological analysis and used as input for integrated risk prediction. The supervisor agent (35) is the output value (B) of each agent. t , P t , A t , E t Collecting ) and applying the weighted sum and interaction term to the integrated feature vector (F t It generates a vector. This vector is normalized through a sigmoid function σ(·) and passed to a risk prediction module (40). Additionally, it includes an outlier detection function to correct abnormal output values ​​of specific agents.

[0071] The multi-agent analysis module (30) may additionally include new agents such as social relationship analysis and biological stress prediction. In addition, communication between agents may use various protocols such as REST API, gRPC, and MQTT, and is applicable to cloud, edge, and on-device environments. Through this structure, the present invention performs multidimensional analysis of the user state in real time and can calculate the probability of smoking cessation failure with high accuracy.

[0072] FIG. 4 is a block diagram showing the configuration and processing procedure of a risk prediction module (40) according to an embodiment of the present invention. The module receives an integrated feature vector transmitted from a multi-agent analysis module (30) as input, calculates the probability of a user failing to quit smoking, and performs the role of transmitting a signal to an intervention execution module (50) according to the prediction result. The risk prediction module (40) includes a risk level calculation unit (41), a risk determination unit (42), and an intervention trigger generation unit (43).

[0073] The risk calculation unit (41) receives the integrated feature vector (F) transmitted from the supervisor agent (35). t The risk of smoking cessation failure is calculated based on ). This calculation unit may include multiple artificial intelligence models, specifically applying one or more of machine learning, deep learning, reinforcement learning, time series forecasting, graph neural networks (GNN), survival analysis, Bayesian inference, and outlier detection techniques. This structure enables risk estimation from multiple perspectives by simultaneously considering complex user data. For example, a time series-based model reflects the temporal pattern of smoking cravings, and a graph-based model learns the interrelationships between behavior, psychology, and environment to estimate complex causes of smoking cessation failure.

[0074] The risk calculation unit (41) calculates the risk of smoking cessation failure R at a specific time point t using learned model parameters. t Calculate R t is the normalized score B obtained from the results of behavioral analysis, psychological analysis, app history analysis, and environmental cognitive analysis. t , P t , A t , E t and the exponentially weighted moving average (EWMA) trend term Δ of the recent τ interval t It can be expressed by the following formula, including .

[0075]

[0076] Here, σ(·) is the sigmoid function, and w0, wb , w p , w a , w e , w bp , w be , w pa γ is the learned weight. The result is expressed in the form of a probability value between 0 and 1, and the closer it is to 1, the higher the probability of failing to quit smoking.

[0077] The risk determination unit (42) determines the calculated risk level R t user-based threshold Determine whether intervention is necessary by comparing with. is the system default value ,User unique habits and dependency indicators H u , Current motivation and compliance indicator C t It is calculated by reflecting [this]. It can be expressed as a formula as follows, where λ and κ are correction factors.

[0078]

[0079] This configuration allows for different adjustments to intervention sensitivity based on each user's level of dependence and compliance. For example, for users with high nicotine dependence or low compliance, a low threshold is set, enabling early intervention.

[0080] The risk assessment unit (42) is In this case, it is determined to be a dangerous state and a signal is transmitted to the intervention trigger generation unit (43). Also, a margin value within a certain range It can be configured to enable proactive intervention even when the threshold is exceeded. This margin-based structure is useful for detecting rapid increases in risk early and performing preventive interventions.

[0081] The intervention trigger generation unit (43) receives the judgment result of the risk determination unit (42) and transmits a trigger signal to the intervention execution module (50). In addition to simple threshold comparison, the trigger generation unit determines the intervention time by analyzing the change trend of the prediction result. For example, R during a continuous time interval t Cases where is continuously rising or Δ t If the value remains positive for a certain period, it is determined that the risk level is rapidly increasing, and a signal is transmitted to execute an intervention immediately. In this process, the trigger generator may apply hysteresis logic to prevent unnecessary over-intervention, and may also be configured so that the intervention is executed only if the same condition is maintained for a certain period.

[0082] The risk prediction module (40) also collects user response data after the intervention and transmits it to the retraining module (60) to continuously improve the performance of the prediction model. If the user suppresses smoking or maintains a smoke-free state after the intervention, the data is used as positive feedback to update the model's weights, whereas if relapse occurs after the intervention, the model's sensitivity is strengthened as negative feedback.

[0083] The risk prediction module according to the present embodiment enables adaptive smoking cessation treatment optimized for individual users. For example, if it is determined that emotional changes have a significant impact on smoking relapse for a specific user, the model can be automatically adjusted to assign higher weight to the results of the psychological analysis. On the other hand, for users where environmental factors are the primary factor, the output of the environmental perception agent is reflected more significantly in the prediction.

[0084] Additionally, the risk prediction module (40) can be executed in either a cloud server or an edge device environment. In the case of a cloud-based system, it is advantageous for large-scale training and model management, while in the case of an edge-based system, user data privacy protection and real-time responsiveness are improved. The two methods can be combined in a hybrid form, and processing efficiency can be maximized through a distributed computing structure between the server and the terminal.

[0085] In one embodiment, the risk calculation unit (41) may be configured with multiple models in an ensemble form instead of a single model. For example, a short-term prediction model and a long-term prediction model may be operated in parallel to evaluate the possibility of short-term smoking cessation failure and the possibility of long-term maintenance, respectively, and then the two results may be integrated to perform a more precise prediction. In addition, the risk judgment unit (42) may perform a judgment that reflects uncertainty by applying Bayesian inference or fuzzy logic in addition to a simple threshold comparison method.

[0086] As explained above, the risk prediction module (40) according to FIG. 4 is a key component that integrates the results of various analysis agents to quantitatively evaluate the likelihood of a user failing to quit smoking and determines the timing of intervention based on the results. Through this module, the system can predict the user's condition in real time and improve the success rate of quitting smoking by performing timely customized intervention.

[0087] The intervention execution module (50) is configured to perform appropriate smoking cessation intervention for the user based on the results transmitted from the risk prediction module (40). This module includes an intervention strategy determination unit, a content creation unit, and a notification transmission unit, and executes customized intervention in the form of adjusting smoking cessation goals, providing reminders, providing reward guidance, and suggesting alternative behaviors. It automatically selects and delivers various intervention scenarios according to changes in the user's state, and subsequent responses are provided as feedback to the relearning module (60).

[0088] FIG. 5 is a block diagram showing the configuration and operation procedure of a retraining module (60) according to an embodiment of the present invention. The module collects user response data from the execution result of the intervention execution module (50) and continuously updates the prediction model and intervention strategy based on this. The retraining module (60) includes a feedback collector (61), a retraining orchestrator (62), and a model and strategy update unit (63), and is linked with a database (80) to improve system performance.

[0089] The feedback collector (61) collects changes in user behavior after intervention. The data includes whether the smoking cessation goal was achieved, the time of relapse, frequency of app usage, notification response rate, and changes in sleep patterns. This data is transmitted in real time through the user terminal (200) and the server (100) and is safely stored after undergoing an anonymization process.

[0090] The retraining orchestrator (62) converts the collected feedback data into a training dataset and periodically updates the prediction model and intervention strategy. Retraining can be performed through either online training or batch training, and the system automatically calibrates the model whenever new data is received, or performs retraining using accumulated data at regular intervals.

[0091] The model and strategy update unit (63) plays the role of reflecting the learning results throughout the system. The updated model is version-controlled in the model registry (82) and is deployed to the operating environment after performance verification. If necessary, it is also possible to restore to the previous model. In addition, long-term performance can be tracked and managed by combining it with user-specific data stored in the smoking cessation DTx DB (81).

[0092] The retraining module (60) can be extended to a reinforcement learning or federated learning structure. In the reinforcement learning method, policy optimization is performed by giving a higher reward the longer the smoking cessation period, and in the federated learning method, personal information protection and learning efficiency are simultaneously secured by sending only the statistical information of the model learned in the local environment to the server.

[0093] Ultimately, the relearning module (60) is configured to continuously improve the prediction accuracy and intervention effect of the system based on user feedback, and plays a role in increasing the precision of personalized smoking cessation treatment over time.

[0094] The function of each module described above can be implemented by the following digital smoking cessation treatment method (600).

[0095] FIG. 6 is a block diagram showing the procedure flow of a digital smoking cessation treatment method (600) according to an embodiment of the present invention. The method (600) is performed while a user terminal (200) and a server (100) are connected via a network, and includes a series of steps to analyze user data and provide customized intervention using an artificial intelligence-based conversational interface.

[0096] The method (600) first includes a step (s605) of interacting with a user through an artificial intelligence-based conversational interface module. In this step, the chatbot or voice recognition interface receives the user's utterance or input to identify the smoking cessation goal, smoking habit, emotional state, lifestyle pattern, etc. The conversational interface analyzes the user's tone, word choice, response speed, etc., through a natural language processing engine to estimate the emotional state and converts it into a structured data form so that it can be used in the next step.

[0097] The method (600) includes the step (s610) of collecting data from the interaction and related data affecting the user's smoking in real time. The data collection module integrates various data, such as conversation data, app usage data, behavior and location data, biometric data, environmental data, and social and schedule data, and transmits them to a server. The collected data is stored after undergoing standardization and encryption processes, and time synchronization is performed to maintain the correlation between the data.

[0098] The method (600) includes the step (s615) of analyzing the collected data using a plurality of AI agents. The agent analysis module operates multiple agents for behavioral analysis, psychological analysis, app history analysis, and environmental perception analysis in parallel to comprehensively evaluate the user's behavioral, psychological, and environmental factors. The analysis results of each agent are integrated by a supervisor agent and normalized into an integrated feature vector, which is used as a prediction input for the next step.

[0099] The method (600) includes a step (s620) of predicting the probability of a user failing to quit smoking based on the analysis results. The risk prediction module receives an integrated feature vector as input and calculates the probability of smoking cessation failure R through a specific formula, machine learning, or deep learning-based prediction model. t Calculates. The calculated R t is expressed as a probability value between 0 and 1, and a higher value indicates a higher probability of failure. In addition, user-specific personalization parameters (H u ) and compliance indicators (C t ) is reflected, and user-customized predictions are performed.

[0100] The method (600) includes a step (s625) of providing personalized intervention to the user based on the prediction results. The intervention execution module performs interventions such as resetting smoking cessation goals, providing motivational messages, offering reward notifications, and suggesting alternative behaviors to the user when the predicted risk level exceeds a threshold. The intervention strategy decision unit selects the most effective intervention method by synthesizing the results of psychological analysis and environmental perception, and the content creation unit automatically generates personalized messages or visual materials. The notification transmission unit delivers the intervention in real time through mobile applications, text messages, messengers, wearable devices, etc.

[0101] The method (600) includes a step (s630) of collecting user response and service usage logs after intervention to retrain a prediction model and an intervention strategy. The retraining module analyzes user response data through a feedback collector, a retraining orchestrator, and a model and strategy update unit, and periodically updates the model's weights. If smoking cessation is successfully maintained after intervention, the data is learned as positive feedback, and if relapse occurs, it is reflected as negative feedback to strengthen prediction sensitivity. Additionally, a reinforcement learning-based policy update is performed to maximize the probability of maintaining long-term smoking cessation.

[0102] In this way, the method (600) cyclically performs a series of procedures from step (s605) to step (s630) through network communication between the server (100) and the user terminal (200). The results of each step are reflected in real time, so that the prediction and intervention strategies are automatically adjusted according to changes in the user's state, and the precision of personalized smoking cessation treatment is continuously improved through repeated execution.

[0103] The features, structures, effects, etc. described in the embodiments above are included in at least one embodiment of the present invention and are not necessarily limited to only one embodiment. Furthermore, the features, structures, effects, etc. exemplified in each embodiment may be combined or modified and implemented in other embodiments by a person skilled in the art to which the embodiments belong. Accordingly, details regarding such combinations and modifications should be interpreted as being included within the scope of the present invention. Explanation of the symbols

[0105] 100 servers 200 user terminals 110 Smoking Cessation Digital Therapy Provider 10 Conversation Interface Modules 20 Data Collection Module 30 Multi-Agent Analysis Modules 40 Risk Prediction Module 50 Intervention Execution Modules 60 Relearning Modules 70 Control Unit

Claims

Claim 1 A device for providing digital treatment for smoking cessation comprises: an AI-based conversational interface module that interacts with a user; a data collection module that collects in real time data from said interaction and related data affecting the user's smoking; a multi-agent analysis module that analyzes said collected data through a plurality of AI agents; a risk prediction module that predicts the probability of smoking cessation failure based on the analysis results; and an intervention execution module that provides personalized intervention to the user according to the prediction results, wherein said related data includes at least one of conversational data, app usage data, behavioral and location data, biometric data, environmental data, and social and schedule data, said multi-agent analysis module includes at least two of a behavioral analysis agent, a psychological analysis agent, an app history analysis agent, and an environmental perception agent, and said risk prediction module generates an integrated feature vector by applying an interaction term based on the correlation between agents to the analysis results of each of said plurality of AI agents to determine the probability of smoking cessation failure (R t A digital smoking cessation therapy providing device that produces ). Claim 2 delete Claim 3 A digital treatment for smoking cessation providing device according to claim 1, wherein the conversation data includes at least one of chatbot conversation content, question and answer, emotional expression and language pattern. Claim 4 A smoking cessation digital treatment providing device according to claim 1, wherein the app usage data includes at least one of usage history of health-related apps including meditation, breathing, and exercise apps; content viewing history; and survey responses and app access frequency. Claim 5 A digital treatment device for smoking cessation according to claim 1, wherein the behavior and location data include at least one of the user's movement path, whether the user visited a location-based smoking area, activity level, sleep pattern, and access history by time of day. Claim 6 A digital treatment device for smoking cessation according to claim 1, wherein the biometric data includes at least one of heart rate, respiratory rate, body temperature, skin conductivity, blood oxygen saturation, step count, and sleep quality. Claim 7 A digital treatment device for smoking cessation according to claim 1, wherein the environmental data includes at least one of temperature, humidity, fine dust, illuminance, noise, air quality, and weather. Claim 8 A digital treatment device for smoking cessation according to claim 1, wherein the social and schedule data includes at least one of SNS activity records, call and message frequency, schedule calendar information, and social event participation logs. Claim 9 delete Claim 10 A digital smoking cessation treatment providing device according to claim 1, wherein the behavior analysis agent analyzes at least one of the user's smoking craving records, location history, activity sensor data, and behavior frequency by time period. Claim 11 A digital treatment device for smoking cessation, wherein, in claim 1, the psychological analysis agent detects emotional signals within the interaction and determines a dangerous emotional state. Claim 12 A digital smoking cessation treatment providing device according to claim 1, wherein the environment perception agent calculates external factors affecting smoking cessation risk using at least one of weather, air quality, noise, illuminance, schedule, and social activity log. Claim 13 A digital smoking cessation treatment providing device according to claim 1, wherein the prediction of the likelihood of smoking cessation failure utilizes one or more of machine learning, deep learning, reinforcement learning, time series forecasting, graph neural networks (GNN), survival analysis, Bayesian inference, or outlier detection techniques. Claim 14 A digital smoking cessation therapy providing device according to claim 1, wherein the personalized intervention comprises one or more of modifying smoking cessation goals, providing reminders, recommending personalized content, sending notifications to experts or family, providing rewards, and suggesting alternative behaviors. Claim 15 A smoking cessation digital treatment providing device, wherein the intervention is delivered to the user via mobile app notifications, text messages, messenger or wearable device notifications. Claim 16 A digital smoking cessation treatment providing device according to claim 1, further comprising a retraining module that collects user responses and service usage logs after intervention to retrain a prediction model and intervention strategy. Claim 17 In claim 1, the probability of smoking cessation failure R based on the analysis result of the multi-agent analysis module. t a user-based threshold When more than or equal to margin It is configured to execute intervention when is above a set value, and the above R t is calculated using the following formula, Here, B t is the normalized score (0~1) of the behavior analysis agent, P t is the normalized score (0~1) of the psychoanalysis agent, A t is the normalization score (0~1) of the app history analysis agent, E t is the normalized score (0~1) of the environment perception agent, Δ t is the exponentially weighted moving average (EWMA) trend over the recent τ interval, σ(ㆍ) is the sigmoid function, and w0, w b , w p , w a , w e , w bp , w be , w pa , γ are weights learned and stored in the model, and the user threshold is as, is the system default, H u is an indicator of user-specific habits and dependency, C t A digital smoking cessation therapy providing device, where is a current motivation and compliance indicator, and λ and κ are correction coefficients. Claim 18 A method for providing digital treatment for smoking cessation comprises: a step of interacting with a user through an AI-based conversational interface module; a step of collecting data from the interaction and related data affecting the user's smoking in real time; a step of analyzing the collected data through a plurality of AI agents; a step of predicting the user's likelihood of smoking cessation failure based on the analysis results; and a step of providing personalized intervention to the user according to the prediction results, wherein the related data includes at least one of conversational data, app usage data, behavioral and location data, biometric data, environmental data, and social and schedule data, and the multi-agent analysis module includes at least two of a behavioral analysis agent, a psychological analysis agent, an app history analysis agent, and an environmental perception agent, and the risk prediction module generates an integrated feature vector by applying an interaction term based on the correlation between agents to the analysis results of each of the plurality of AI agents to determine the likelihood of smoking cessation failure (R t A method for providing digital smoking cessation therapy that produces ). Claim 19 delete Claim 20 delete Claim 21 A computer-readable storage medium storing a program for performing the method of claim 18, wherein the computer causes the computer to perform the step of claim 18 when the program is executed.

Images