Depression digital therapy dynamic intervention method and system based on robust optimization and causal inference

Abstract

The invention relates to a depression digital therapy dynamic intervention method and system, and the method comprises the steps: obtaining the multi-source original data, related to the psychological state, of a target user in a current decision period, and determining the state data used for representing the psychological state of the target user according to the multi-source original data, determining an uncertainty set of measurement errors for quantifying the state data; based on the group characteristics of the target user, determining an expected causal effect value of each intervention scheme in an intervention scheme library on the target user by using a causal effect analysis model; constructing a robust optimization model based on the state data and the uncertainty set, taking the expected causal effect value of each intervention scheme on the target user as an optimization target, and solving the robust optimization model to determine a target intervention scheme from an intervention scheme library; and pushing the target intervention scheme to the target user. According to the method, the accurate depression personalized intervention strategy can be dynamically pushed to the user under the data uncertainty.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular to a dynamic intervention method and system for digital therapy of depression based on robust optimization and causal inference. Background Technology

[0002] Existing digital therapy software for depression, such as CBT-I, generally employs a fixed intervention process: the system pushes tasks of the same intensity and type to all users according to a preset sequence or simple questionnaire thresholds, failing to dynamically adjust based on real-time changes in mood, behavior, or physiology. When users are in a low mood, they may still be assigned high-load tasks, which may increase their burden, reduce compliance, or even induce counterproductive effects. Furthermore, the aforementioned existing technologies primarily design interventions based on group average effects or correlation analysis, making it difficult to effectively identify and evaluate the heterogeneous treatment effects of the same intervention program on different user subgroups, lacking the technological foundation for truly personalized and precise treatment. Summary of the Invention

[0003] To address the aforementioned shortcomings and deficiencies, this application provides a dynamic intervention method and system for digital therapy of depression based on robust optimization and causal inference. This application achieves dynamic optimization and precise delivery of personalized intervention strategies for depression under data uncertainty through a closed-loop fusion of robust optimization and causal inference.

[0004] This application provides a dynamic intervention method for digital therapy of depression according to a first aspect, the method comprising: Acquire multi-source raw data related to the psychological state of the target user in the current decision-making cycle, determine the state data used to characterize the psychological state of the target user based on the multi-source raw data, and determine the uncertainty set of the measurement error used to quantify the state data; Based on the group characteristics of the target users, a causal effect analysis model is used to determine the expected causal effect value of each intervention program in the intervention program library on the target users; A robust optimization model is constructed based on state data and uncertainty set. The expected causal effect value of each intervention plan on the target user is taken as the optimization objective. The robust optimization model is solved to determine the target intervention plan from the intervention plan library. Push the target intervention plan to the target users.

[0005] In some embodiments, the multi-source raw data includes objective sensor data from the user terminal and subjective evaluation data from user input; determining state data to characterize the psychological state of the target user based on the multi-source raw data includes: The objective data and subjective evaluation data from the sensors are preprocessed and fused to generate a multidimensional state vector; the multidimensional state vector includes quantitative indicators of user emotions, behavioral activity and physiological rhythms. Use multidimensional state vectors as state data.

[0006] In some embodiments, sensor objective data and subjective evaluation data are preprocessed and fused to generate a multidimensional state vector, including: The objective data from the sensors is cleaned to remove noise and outliers, and missing values ​​are imputed. Standardize the subjective assessment data to convert it into a unified dimension and scoring range; Objective feature indicators are extracted from the processed sensor objective data, and subjective feature indicators are extracted from the processed subjective evaluation data. By fusing objective and subjective feature indicators, a multidimensional state vector is obtained.

[0007] In some embodiments, based on the group characteristics of the target users, a causal effect analysis model is used to determine the expected causal effect value of each intervention in the intervention program library on the target users, including: Obtain the group characteristics of the target users; Input group characteristics into a pre-trained causal effect analysis model; The expected intervention effect value of each intervention program on the target user is determined based on the output data of the causal effect analysis model.

[0008] In some embodiments, the training process of the causal effect analysis model includes: A training dataset is constructed based on historical, multi-source raw data from multiple users, actual intervention plans, and intervention effect feedback data. Mobile terminal behavior indicators that are strongly correlated with intervention allocation and are not related to potential confounding factors are selected as instrumental variables. The instrumental variable forest model conditioned on the instrumental variables is trained using the training dataset to obtain a causal effect analysis model for evaluating the heterogeneous causal effects of intervention programs on user groups with different characteristics.

[0009] In some embodiments, a robust optimization model is constructed based on state data and an uncertainty set. The expected causal effect value of each intervention scheme on the target user is used as the optimization objective. The robust optimization model is solved to determine the target intervention scheme from the intervention scheme library, including: A robust optimization model is established by using state data as input parameters and defining the parameter fluctuation range using uncertainty sets. The robust optimization model is solved with the objective of minimizing the cost function based on the expected intervention effect. Based on the solution results, the target intervention program is determined from the intervention program library.

[0010] In some embodiments, the method further includes: Acquire behavioral pattern indicators of target users from multiple sources within a preset historical time window; The behavioral pattern indicators are matched with a pre-built group feature template library for similarity. The group feature template library is established by clustering historical user data and labeling the typical behavioral patterns of each user group based on the clustering results. Collect group features from the group feature template library whose similarity to behavioral pattern indicators is greater than a preset similarity threshold as the group features of the target user.

[0011] According to a second aspect, this application provides a dynamic intervention system for digital therapy of depression, the system comprising: The user data determination module is used to acquire multi-source raw data related to the psychological state of the target user in the current decision-making cycle, determine the state data used to characterize the psychological state of the target user based on the multi-source raw data, and determine the uncertainty set of measurement error used to quantify the state data. The expected intervention effect determination module is used to determine the expected causal effect value of each intervention program in the intervention program library on the target users based on the group characteristics of the target users and using a causal effect analysis model. The target intervention program determination module is used to build a robust optimization model based on state data and uncertainty set. It takes the expected causal effect value of each intervention program on the target user as the optimization objective, solves the robust optimization model, and determines the target intervention program from the intervention program library. The intervention plan push module is used to push target intervention plans to target users.

[0012] According to a third aspect, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any of the above-described embodiments of the dynamic intervention method for digital therapy of depression.

[0013] According to a fourth aspect, this application provides a computer device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executed, implements the steps of any of the above-described embodiments of the dynamic intervention method for digital therapy of depression.

[0014] This application achieves precise personalization of intervention plans through causal inference and ensures the effectiveness of intervention plans in real-world uncertain environments through robust optimization. Specifically, this application acquires multi-source raw data related to the target user's psychological state during the current decision-making cycle. Based on this multi-source raw data, it determines the state data used to characterize the target user's psychological state and identifies the uncertainty set used to quantify the measurement error of the state data. This integrates scattered and noisy perceptual information into state data representing the psychological state and explicitly characterizes its fluctuation range. Next, based on the group characteristics of the target user, a causal effect analysis model is used to determine the expected causal effect value of each intervention plan in the intervention plan library on the target user. This allows for the evaluation of the expected causal effect value of different intervention plans on the target user, which helps to accurately identify the most likely effective intervention plan for the target user. Then, a robust optimization model is constructed based on the state data and the uncertainty set. The expected causal effect value of each intervention plan on the target user is used as the optimization objective. The robust optimization model is solved to determine the target intervention plan from the intervention plan library. Even under uncertain conditions such as user state fluctuations and data inconsistencies, a safe, effective, and executable target intervention plan can still be selected, significantly improving the reliability and adaptability of the intervention system. Attached Figure Description

[0015] Figure 1 A flowchart of a dynamic intervention method for digital therapy of depression in one or more embodiments of this application; Figure 2 This is a schematic diagram of the structure of a dynamic intervention system for digital therapy of depression according to one or more embodiments of this application; Figure 3 This is a schematic diagram of the internal structure of a computer device according to one or more embodiments of this application. Detailed Implementation

[0016] To make the objectives, technical solutions, and advantages of this application clearer, the embodiments of this application will be described in further detail below with reference to the accompanying drawings. It should be understood that the described embodiments are merely some embodiments of this application, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0017] In the following description, when referring to the accompanying drawings, the same numbers in different drawings denote the same or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0018] In the description of this application, it should be understood that the terms "first," "second," "third," etc., are used only to distinguish similar objects and are not necessarily used to describe a specific order or sequence, nor should they be construed as indicating or implying relative importance. Those skilled in the art can understand the specific meaning of the above terms in this application according to the specific circumstances. Furthermore, in the description of this application, unless otherwise stated, "multiple" refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. The character " / " generally indicates that the preceding and following related objects have an "or" relationship.

[0019] To address the shortcomings or defects of related technologies, this application provides a dynamic intervention method for digital therapy of depression. This method achieves precise personalization of intervention plans through causal inference and ensures the effectiveness of intervention plans in real uncertain environments through robust optimization.

[0020] In some exemplary embodiments of this application, such as Figure 1 As shown, the method includes steps S110-S140. The following describes each step in detail using the method applied to an intervention system as an example. The intervention system (hereinafter referred to as the system) can be a server, and user terminals (such as smartphones, wearable devices, etc.) can communicate with the system through a wireless network.

[0021] S110: Obtain multi-source raw data related to the target user's psychological state in the current decision-making cycle, determine the state data used to characterize the target user's psychological state based on the multi-source raw data, and determine the uncertainty set used to quantify the measurement error of the state data.

[0022] At each predetermined decision-making cycle (e.g., 8:00 AM daily), the intervention system acquires multi-source raw data related to the psychological state of the target user (the user requiring the intervention plan) during the current decision-making cycle. This multi-source raw data can be collected by the user's terminal with the user's permission and uploaded to the intervention system, where it undergoes further processing.

[0023] Multi-source raw data is a collection of raw data related to the user's psychological state, collected from different channels, types of sensors, or input methods. Multi-source raw data includes objective sensor data from the user's terminal and subjective evaluation data from active user input.

[0024] Objective sensor data refers to quantitative data that reflects the user's objective behavior and environment, automatically collected by built-in sensors in mobile terminals such as smartphones and wearable devices, including accelerometers, gyroscopes, GPS, light sensors, and screen usage sensors. Data collected by accelerometers and gyroscopes can be used to calculate a user's daily steps, body movement amplitude, and static duration; GPS location data can be used to analyze a user's daily activity radius, number of frequently visited locations, and outdoor time spent. Screen usage sensors collect screen usage status and logs, which can be used to statistically analyze the total screen-on time, frequency of use of various applications, and time distribution. Subjective evaluation data is data recorded by users in applications, reflecting their subjective psychological feelings, such as mood self-rating scores, sleep quality scores, and diary text content. User terminals must obtain user permission before collecting objective sensor data and subjective evaluation data.

[0025] The intervention system will determine state data and an uncertainty set based on multi-source raw data. State data refers to the quantitative data obtained by processing multi-source raw data to comprehensively characterize the user's current psychological state. The uncertainty set is a mathematical set used to describe and quantify the uncertainty or possible range of variation in the state data caused by measurement errors (or even noise, individual fluctuations, etc.).

[0026] The operation of determining state data to characterize the psychological state of a target user based on multi-source raw data may include: preprocessing and fusing objective sensor data and subjective evaluation data to generate a multi-dimensional state vector; the multi-dimensional state vector includes quantitative indicators of user emotions, behavioral activity, and physiological rhythms; and using the multi-dimensional state vector as state data.

[0027] The intervention system preprocesses and fuses various sensor data and subjective assessment data to obtain a multidimensional state vector. Preprocessing includes noise filtering, outlier removal, missing value imputation, and standardization. Outliers are data points that deviate significantly from the majority of data and may be due to measurement errors or random events. Missing values ​​are data points that were not successfully recorded at the intended time or location.

[0028] The operation of preprocessing and fusing objective sensor data and subjective evaluation data to generate a multidimensional state vector includes: cleaning the objective sensor data to remove noise and outliers and imputing missing values; standardizing the subjective evaluation data to convert it into a unified dimension and scoring range; extracting objective feature indicators from the processed objective sensor data and subjective feature indicators from the processed subjective evaluation data; and fusing the objective and subjective feature indicators to obtain the multidimensional state vector.

[0029] This involves low-pass filtering of various sensor data to remove noise, identifying and eliminating outliers and filling in missing values ​​based on physiologically reasonable ranges (e.g., non-negative step count and less than 20,000 steps / hour). Specifically, missing data can be estimated and filled using statistical methods (e.g., mean and median imputation) or model prediction.

[0030] Standardizing subjective evaluation data, such as through Z-score standardization to achieve a mean of 0 and a variance of 1, eliminates biases arising from different users' rating habits. Standardization transforms data with different dimensions and value ranges into a unified, comparable numerical range through mathematical transformations (such as normalization and Z-score standardization). Dimensions refer to the units or scales of the data, such as time (minutes), frequency, or ratings (1-5 points).

[0031] Feature indicators refer to specific quantitative parameters that can effectively characterize a user's state in a certain aspect, extracted from multi-source raw data through calculation, statistics, or transformation. Examples include "average daily steps," "nighttime screen usage time," and "variance of emotion rating." Specifically, objective feature indicators can be extracted from processed sensor objective data, and subjective feature indicators can be extracted from processed subjective evaluation data.

[0032] After feature extraction, all extracted objective and subjective feature indicators are concatenated in a preset order to form a fixed multidimensional (e.g., n-dimensional) state vector, which can be represented as S_t=[f_1,f_2,...,f_n]. Specifically, the multidimensional state vector is an array containing multiple numerical elements. Each element represents a psychological state dimension (such as emotion, behavior, or physiology) extracted and quantified from the original multidimensional data. These elements can be used to characterize the user's current psychological state.

[0033] The operation of determining the uncertainty set for quantifying the measurement error of state data can be to construct the uncertainty set by estimating a possible error or fluctuation range for the feature indicators of each psychological state dimension in the multidimensional state vector by the intervention system.

[0034] For each characteristic indicator of a psychological state dimension, its uncertainty range can be determined based on historical fluctuations, measurement accuracy, and / or rating consistency.

[0035] The method based on historical fluctuations can be to calculate the standard deviation σ of relevant characteristic indicators (such as f_i) of relevant users (such as target users) in the same period (such as weekday mornings) in the same period of history, and then set the uncertainty interval as [f_i-2σ, f_i+2σ].

[0036] The measurement accuracy-based method can specifically calculate the interval based on the accuracy parameters provided by the sensor manufacturer and the features derived from the sensor (such as the number of steps) proportionally.

[0037] A specific approach based on rating consistency could be to adjust the interval width by combining subjective rating characteristics with the internal consistency coefficient of users' historical ratings.

[0038] S120: Based on the group characteristics of the target users, use a causal effect analysis model to determine the expected causal effect value of each intervention program in the intervention program library on the target users.

[0039] Group characteristics refer to the common features of user groups with similar behavioral patterns or psychological characteristics, which are derived from the analysis and summarization of historical data.

[0040] Causal effect analysis models are pre-trained machine learning or statistical models used to assess the causal impact of different interventions on specific types of users and output the expected causal effect value. The expected causal effect value is specifically a quantitative indicator used to predict the magnitude of the effect or benefit that a particular intervention may produce on the current target users.

[0041] Based on the group characteristics of the target users, the operation of determining the expected causal effect value of each intervention in the intervention program library on the target users using a causal effect analysis model may include: obtaining the group characteristics of the target users; inputting the group characteristics into a pre-trained causal effect analysis model; and determining the expected intervention effect value of each intervention program on the target users based on the output data of the causal effect analysis model.

[0042] Causal effect analysis models can be pre-trained using an instrumental variable forest. After receiving the group characteristics of the target users, the causal effect analysis model locates several decision tree nodes in its internal forest structure that best match the group characteristics of the target users. These nodes store the conditional average treatment effects of various intervention schemes for user groups related to these group characteristics, obtained by correcting estimates based on historical data and using instrumental variables (such as "the average time point of receiving system push notifications in the past week").

[0043] The output data of the causal effect analysis model is a vector, which can be represented as E_t=[e_1,e_2,...,e_m], where m is the number of interventions included in the intervention program library, and e_j represents the expected causal effect value of the j-th intervention program in the intervention program library on the user group to which the target user belongs (e.g., expected to reduce depression score by 0.5 points).

[0044] The intervention system directly uses each e_j in the output data E_t of the causal effect analysis model as the expected intervention effect value of the corresponding intervention plan. The expected intervention effect value is a real number, and its positive or negative sign indicates an expected positive improvement effect or a negative risk of deterioration, respectively.

[0045] In the above process, the intervention system accurately matches users to user groups with similar behavioral and psychological characteristics, and uses a causal effect analysis model trained with instrumental variable correction based on historical data of the user group to make predictions. This allows each intervention plan to be assigned an expected causal effect value that varies from person to person and from group to group. This enables the subsequent robust optimization model to further combine the uncertainty of the target user's real-time state to make refined trade-offs and decisions, realizing a deep integration of personalized knowledge (causal inference) and real-time robust decision-making (robust optimization), which significantly improves the pertinence and effectiveness of dynamic intervention strategies.

[0046] In some technical solutions, the training process of the causal effect analysis model includes: constructing a training dataset based on historical multi-source raw data of multiple users, actual implemented intervention programs, and intervention effect feedback data; selecting mobile terminal behavior indicators that are strongly correlated with intervention allocation and are not related to potential confounding factors as instrumental variables, and using the training dataset to train the instrumental variable forest model conditioned on the instrumental variables, thereby obtaining a causal effect analysis model for evaluating the heterogeneous causal effects of intervention programs on user groups with different characteristics.

[0047] Causal effect analysis models can be trained offline in advance.

[0048] Specifically, the intervention system can extract user data from a backend database for all specific users within a historical period (e.g., the past 6 months). Specific users refer to those who have granted the intervention system permission to use their personal data for training causal effect analysis models. User data for specific users includes historical multi-source raw data, actually implemented intervention protocols, and intervention effect feedback data. Historical multi-source raw data refers to the multi-source raw data for specific users at each decision-making cycle (details are explained above and will not be repeated here). Actually implemented intervention protocols refer to the specific intervention protocols and their push timestamps that the intervention system has historically pushed to specific users. Intervention effect feedback data refers to the effect indicators collected by specific users over several subsequent decision-making cycles after implementing the intervention protocols pushed by the intervention system, such as changes in mood scores, task completion scores, and / or symptom change scores derived from a simplified version of a clinical scale (e.g., PHQ-2).

[0049] The intervention system can preprocess and align user data for specific users. For example, it can align the data by user ID and time series, clean up invalid records, and standardize feature indicators. The training sample set specifically includes multiple training samples, each corresponding to a four-tuple of "user-decision cycle-intervention plan-subsequent effect," thus forming a structured dataset (i.e., the training dataset). Each training sample can contain: a user feature vector (composed of the user's state data and historical behavioral pattern indicators during the decision cycle), the assigned intervention plan, instrumental variables, and the observed intervention effect.

[0050] The operation of selecting mobile terminal behavior indicators that are strongly correlated with intervention allocation and are not related to potential confounding factors as instrumental variables can be carried out by screening a set of mobile terminal behavior indicators that may be related to the system intervention allocation logic but are not related to potential confounding variables from the objective sensor data part of the training sample set. For example, "mobile phone usage time", "location movement frequency", "number of times the mobile phone screen is unlocked in the hour before the system's preset push time on the same day", "frequency of users being outdoors during the system's usual push time in the past 3 days", "delay time of user's click response to yesterday's push notification", etc.

[0051] Potential confounding variables refer to other unobserved or difficult-to-measure variables (such as genetic factors, family environment, and personality traits) that simultaneously affect both the likelihood (or degree) of a user receiving an intervention and the final therapeutic outcome. These variables may cause spurious correlations. Instrumental variables selected in the above manner are highly correlated with the intervention allocation (cause) under study, but have no other direct or indirect causal relationship with the studied outcome (effect) except through the intervention pathway, and are unrelated to potential confounding factors affecting the outcome. Using these instrumental variables to estimate causal effects in observational data helps to isolate the pure causal effect of the intervention, rather than the superficial correlation influenced by other confounding factors.

[0052] Next, an instrumental variable forest model is constructed and trained. The instrumental variable forest model is a variant of the generalized random forest. Its special characteristic lies in the fact that the splitting rule depends not only on the relationship between features and outcomes, but also on the relationship between features, instrumental variables, and treatments, in order to consistently estimate the conditionally averaged treatment effect. This embodiment uses the instrumental variable forest model to estimate the heterogeneous causal effects of an intervention on individuals with different characteristics, given the presence of instrumental variables. Heterogeneous causal effects refer to the phenomenon where the magnitude or direction of the causal effect of the same intervention differs among users with different characteristics and types (i.e., different group characteristics).

[0053] When training the instrumental variable forest model, the training dataset is used as input data. Each training sample includes a user feature vector (a multidimensional state vector of a specific user), a processing variable (representing the actual intervention plan), an outcome variable (representing the intervention effect), and an instrumental variable.

[0054] The instrumental variable forest model learns the local causal effects of different interventions on outcomes given user characteristics and instrumental variables by constructing a large number of decision trees. The training process aims to find a data partitioning method that maximizes the accuracy of heterogeneous causal effect estimation.

[0055] After training is complete, the forest model with fixed parameters is used as the causal effect analysis model. This model can receive new user characteristics (and group characteristics) as input and output the expected causal effect value of each intervention in the intervention program library for that user.

[0056] In some technical solutions, the group characteristics of target users can be obtained through the following process: acquiring behavioral pattern indicators of target users from multiple sources within a preset historical time window; matching the behavioral pattern indicators with a pre-built group characteristic template library, which is established by pre-clustering historical user data and labeling the typical behavioral patterns of each user group based on the clustering results; and acquiring group characteristics from the group characteristic template library whose similarity to the behavioral pattern indicators is greater than a preset similarity threshold as the group characteristics of the target users.

[0057] Among them, behavioral pattern indicators refer to characteristic indicators extracted from the target user's historical multi-source raw data (i.e., the target user's multi-source raw data within a preset historical time window) that can summarize their long-term or periodic behavioral habits (such as "average wake-up time on weekdays" and "social application usage frequency pattern").

[0058] The group feature template library includes group feature templates corresponding to various user groups. Each template represents the feature information of a typical user group discovered through cluster analysis. Cluster analysis is an unsupervised machine learning method. This technical solution uses this method to automatically group historical user data with similar features into different categories or groups, thereby constructing the group feature template library. Similarity matching refers to calculating the degree of similarity between the behavioral pattern indicators of the target user and each group feature template in the group feature template library. The degree of similarity can be represented using metrics such as cosine similarity or the reciprocal of Euclidean distance.

[0059] S130: Construct a robust optimization model based on state data and uncertainty set, take the expected causal effect value of each intervention scheme on the target user as the optimization objective, solve the robust optimization model, and determine the target intervention scheme from the intervention scheme library.

[0060] Robust optimization models are a type of operations research optimization model. Their characteristic is that, when there is uncertainty in the parameters (in this case, state data), they seek an optimal decision scheme that performs well (or avoids the worst case) under all possible uncertain scenarios (defined by the uncertainty set).

[0061] The intervention program library is a predefined collection of various digital therapy intervention programs for depression (such as mindfulness exercises, behavioral activation tasks, cognitive restructuring exercises, etc.). The target intervention program is the optimal intervention program for the current decision-making cycle selected from the intervention program library by solving a robust optimization model, and is to be pushed to the target user.

[0062] A robust optimization model is constructed based on state data and uncertainty set. The expected causal effect value of each intervention scheme on the target user is used as the optimization objective. The robust optimization model is solved to determine the target intervention scheme from the intervention scheme library. The process may include: establishing a robust optimization model by using state data as input parameters and defining the parameter fluctuation range with uncertainty set; solving the robust optimization model with the objective of minimizing the cost function based on the expected intervention effect value; and determining the target intervention scheme from the intervention scheme library based on the solution results.

[0063] A cost function is a mathematical function used to measure the costs or negative benefits of an intervention under given conditions and uncertainty. Its optimization objective is to minimize the function value. Solving for the cost function involves using mathematical optimization algorithms (such as linear programming, convex optimization algorithms, etc.) to calculate the decision variable values ​​(i.e., the target intervention) that minimize the cost function under constraints.

[0064] The robust optimization model can be shown below: ; in, It is a multi-dimensional state vector based on the target user's current state. and intervention programs The cost function is described. The core idea of ​​this model is to optimize for the worst-case scenario. This means that when making intervention decisions, the intervention system considers all uncertainties that may exist in the target user's current state and selects an intervention plan that guarantees good results even in the worst-case scenario. This approach can significantly improve the robustness of the intervention system and avoid intervention failures caused by data noise or biases in user state prediction. For example, when a user's physiological data and emotional logs are inconsistent, the robust optimization model will not easily make aggressive interventions, but will instead choose a conservative but effective plan that can adapt to both potential states.

[0065] The following constraints are used when solving the robust optimization model: (1) Resource constraints: ; in, Indicating intervention plan The required user attention, time, and effort, and resource constraints can ensure Not exceeding the maximum threshold .

[0066] (2) Security constraints: ; in, Represents the state transition function. This represents the set of safe states; safety constraints ensure the user's state after intervention, i.e., the multidimensional state vector for the next decision cycle. It is still within a safe range.

[0067] S140: Push the target intervention plan to the target users.

[0068] The intervention system sends the optimal intervention plan (i.e., the target intervention plan) obtained from the robust optimization model to the target user via push notifications or in-app reminders. For example, "Your current mood score is low, but your activity level is good. We suggest trying a 15-minute walk outdoors, and practicing mindfulness breathing while walking." The intervention system records the target user's adherence to the target intervention plan (such as whether they clicked the "complete" button) and uses it as an important input for the next optimization.

[0069] This application acquires multi-source raw data related to the psychological state of target users during the current decision-making cycle. Based on this data, it determines state data to characterize the psychological state of the target users and identifies an uncertainty set to quantify the measurement error of the state data. This integrates scattered and noisy perceptual information into state data that characterizes the psychological state and explicitly describes its fluctuation range. Next, based on the group characteristics of the target users, a causal effect analysis model is used to determine the expected causal effect value of each intervention in the intervention plan library on the target users. This allows for the evaluation of the expected causal effect value of different intervention plans on the target users, which helps to accurately identify the most likely effective intervention plan for the target users. Then, a robust optimization model is constructed based on the state data and the uncertainty set. The expected causal effect value of each intervention plan on the target users is used as the optimization objective. The robust optimization model is solved to determine the target intervention plan from the intervention plan library. Even under uncertainties such as user state fluctuations and data inconsistencies, a safe, effective, and executable target intervention plan can still be selected, significantly improving the reliability and adaptability of the intervention system.

[0070] It should be noted that, unless otherwise explicitly stated herein, the execution order of the steps included in the dynamic intervention method for digital therapy of depression provided in any of the above embodiments is not strictly limited, and these steps can be executed in other orders. Moreover, at least some of these steps may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least a portion of the sub-steps or stages of other steps.

[0071] Based on the same inventive concept, this application also provides a dynamic intervention system for digital therapy of depression. In some embodiments, such as Figure 2 As shown, the dynamic intervention system for digital therapy of depression includes the following modules: The user data determination module 110 is used to acquire multi-source raw data related to the psychological state of the target user in the current decision-making cycle, determine the state data used to characterize the psychological state of the target user based on the multi-source raw data, and determine the uncertainty set of the measurement error used to quantify the state data. The expected intervention effect determination module 120 is used to determine the expected causal effect value of each intervention program in the intervention program library on the target users based on the group characteristics of the target users and using a causal effect analysis model. The target intervention program determination module 130 is used to construct a robust optimization model based on state data and uncertainty set, take the expected causal effect value of each intervention program on the target user as the optimization objective, solve the robust optimization model, and determine the target intervention program from the intervention program library. The intervention plan push module 140 is used to push the target intervention plan to the target user.

[0072] In some embodiments, the multi-source raw data includes objective sensor data from the user terminal and subjective evaluation data actively input by the user; the user data determination module 110 is specifically used to preprocess and fuse the objective sensor data and subjective evaluation data to generate a multi-dimensional state vector; the multi-dimensional state vector includes quantitative indicators of user emotions, behavioral activity and physiological rhythms; and the multi-dimensional state vector is used as state data.

[0073] In some embodiments, the user data determination module 110 is specifically used to clean the sensor objective data to remove noise and outliers, and to impute missing values; to standardize the subjective evaluation data to convert it into a unified dimension and scoring range; to extract objective feature indicators from the processed sensor objective data and subjective feature indicators from the processed subjective evaluation data; and to fuse the objective feature indicators and subjective feature indicators to obtain a multidimensional state vector.

[0074] In some embodiments, the expected intervention effect determination module 120 is specifically used to obtain the group characteristics of the target users; input the group characteristics into a pre-trained causal effect analysis model; and determine the expected intervention effect value of each intervention scheme on the target users based on the output data of the causal effect analysis model.

[0075] In some embodiments, the system further includes a training module. The training module is used to train a causal effect analysis model. The specific training process includes: constructing a training dataset based on historical multi-source raw data from multiple users, actual implemented intervention programs, and intervention effect feedback data; selecting mobile terminal behavior indicators that are strongly correlated with intervention allocation and unrelated to potential confounding factors as instrumental variables; and using the training dataset to train a forest model of instrumental variables conditioned on these instrumental variables, thereby obtaining a causal effect analysis model for evaluating the heterogeneous causal effects of intervention programs on user groups with different characteristics.

[0076] In some embodiments, the target intervention scheme determination module 130 is specifically used to establish a robust optimization model by using state data as input parameters and defining the parameter fluctuation range by an uncertainty set; to solve the robust optimization model with the objective of minimizing the cost function based on the expected intervention effect value; and to determine the target intervention scheme from the intervention scheme library based on the solution results.

[0077] In some embodiments, the system further includes a group feature acquisition module. The group feature acquisition module is used to acquire behavioral pattern indicators of target users from multi-source raw data within a preset historical time window; perform similarity matching between the behavioral pattern indicators and a pre-built group feature template library, which is established by pre-clustering historical user data and labeling typical behavioral patterns of each user group based on the clustering results; and acquire group features from the group feature template library whose similarity to the behavioral pattern indicators is greater than a preset similarity threshold as the group features of the target users.

[0078] Specific limitations regarding the dynamic intervention system for digital therapy for depression can be found in the limitations of the dynamic intervention method for digital therapy for depression mentioned above, and will not be repeated here. Each module in the aforementioned dynamic intervention system for digital therapy for depression can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the corresponding operations of each module.

[0079] This application also provides a computer device. In some embodiments, the computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it can implement the steps of the dynamic intervention method for digital therapy of depression provided in any of the above embodiments.

[0080] In some embodiments, the internal structure diagram of a computer device may be as follows: Figure 3 As shown, the computer device includes a processor, memory, and a network interface connected via a system bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database stores multi-source raw data, robust optimization models, and other data; the specific data stored can be further defined in the above method embodiments. The network interface communicates with external terminals via a network connection. When the computer program is executed by the processor, it implements a dynamic intervention method for digital therapy of depression.

[0081] Those skilled in the art will understand that Figure 3 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0082] This application also provides a computer-readable storage medium, in some embodiments of which a computer program is stored on the computer-readable storage medium, and when the computer program is executed by a processor, it implements the steps of the dynamic intervention method for digital therapy of depression provided in any of the above embodiments.

[0083] In the above embodiments of this application, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0084] Those skilled in the art will understand that implementing all or part of the processes in the above method embodiments can be accomplished by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), Synchlink, DRAM (SLDRAM), memory bus, direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0085] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0086] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A dynamic intervention method for digital therapy of depression, characterized in that, The method includes: Acquire multi-source raw data related to the psychological state of the target user in the current decision-making cycle, determine state data to characterize the psychological state of the target user based on the multi-source raw data, and determine an uncertainty set to quantify the measurement error of the state data; Based on the group characteristics of the target users, the expected causal effect value of each intervention program in the intervention program library on the target users is determined using a causal effect analysis model. A robust optimization model is constructed based on the state data and the uncertainty set. The expected causal effect value of each intervention scheme on the target user is taken as the optimization objective. The robust optimization model is solved to determine the target intervention scheme from the intervention scheme library. The target intervention plan is pushed to the target user.

2. The method according to claim 1, characterized in that, The multi-source raw data includes objective sensor data from user terminals and subjective evaluation data actively input by users. The step of determining the state data used to characterize the psychological state of the target user based on the multi-source raw data includes: The objective data and subjective evaluation data from the sensors are preprocessed and fused to generate a multidimensional state vector; the multidimensional state vector includes quantitative indicators of user emotions, behavioral activity, and physiological rhythms. The multidimensional state vector is used as the state data.

3. The method according to claim 2, characterized in that, The preprocessing and fusion of the objective sensor data and subjective evaluation data to generate a multi-dimensional state vector includes: The objective data from the sensor is cleaned to remove noise and outliers, and missing values ​​are imputed. The subjective evaluation data is standardized to convert it into a unified dimension and scoring range. Objective feature indicators are extracted from the processed sensor objective data, and subjective feature indicators are extracted from the processed subjective evaluation data. The objective feature index and the subjective feature index are fused to obtain the multidimensional state vector.

4. The method according to claim 1, characterized in that, The step of determining the expected causal effect value of each intervention program in the intervention program library on the target users based on the group characteristics of the target users and using a causal effect analysis model includes: Obtain the group characteristics of the target users; The group characteristics are input into a pre-trained causal effect analysis model; The expected intervention effect value of each intervention program on the target user is determined based on the output data of the causal effect analysis model.

5. The method according to claim 4, characterized in that, The training process of the causal effect analysis model includes: A training dataset is constructed based on historical, multi-source raw data from multiple users, actual intervention plans, and intervention effect feedback data. Mobile terminal behavior indicators that are strongly correlated with intervention allocation and are not related to potential confounding factors are selected as instrumental variables. The instrumental variable forest model conditioned on the instrumental variables is trained using the training dataset to obtain a causal effect analysis model for evaluating the heterogeneous causal effects of intervention programs on user groups with different characteristics.

6. The method according to claim 1, characterized in that, The process of constructing a robust optimization model based on the state data and the uncertainty set, using the expected causal effect value of each intervention scheme on the target user as the optimization objective, and solving the robust optimization model to determine the target intervention scheme from the intervention scheme library includes: A robust optimization model is established by using the state data as input parameters and defining the parameter fluctuation range using the uncertainty set. The robust optimization model is solved with the objective of minimizing the cost function based on the expected intervention effect value. Based on the solution results, the target intervention plan is determined from the intervention plan library.

7. The method according to claim 1, characterized in that, The method further includes: Obtain behavioral pattern indicators of the target user from multiple sources within a preset historical time window; The behavioral pattern indicators are matched with a pre-built group feature template library for similarity. The group feature template library is established by clustering historical user data and labeling the typical behavioral patterns of each user group based on the clustering results. The group features in the group feature template library that have a similarity greater than a preset similarity threshold with the behavior pattern index are obtained as the group features of the target user.

8. A dynamic intervention system for digital therapy of depression, characterized in that, The system includes: The user data determination module is used to acquire multi-source raw data related to the psychological state of the target user in the current decision-making cycle, determine state data to characterize the psychological state of the target user based on the multi-source raw data, and determine an uncertainty set to quantify the measurement error of the state data. The expected intervention effect determination module is used to determine the expected causal effect value of each intervention plan in the intervention plan library on the target users based on the group characteristics of the target users and using a causal effect analysis model. The target intervention scheme determination module is used to construct a robust optimization model based on the state data and the uncertainty set, take the expected causal effect value of each intervention scheme on the target user as the optimization objective, solve the robust optimization model, and determine the target intervention scheme from the intervention scheme library. The intervention plan push module is used to push the target intervention plan to the target user.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

Images