Method, software and system for placebo training for cognitive digital therapy

Abstract

The invention provides a placebo training method, software and system for cognitive digital therapy, and belongs to the technical field of cognitive digital therapy. Comprising the steps of S1, constructing a training task library; s2, marking and classifying the training tasks; s3, acquiring feature information of a user; and S4, based on the feature information, determining a training plan of the user, and training the user according to the training plan. According to the method, the rich training task library can be constructed by utilizing the text understanding and output capability of the large language model, so that a placebo training plan with a good confusion effect and a reasonable placebo training plan can be formed according to different experiment groups and feature information of corresponding users.

Description

Technical Field

[0001] The present application relates to the technical field of cognitive digital therapy, and in particular to a placebo training method, software and system for cognitive digital therapy. Background Art

[0002] Digital therapeutics (DTx) refer to evidence-based, clinically validated software for the treatment, management and prevention of diseases. Cognitive digital therapeutics are software-driven interventions that provide patients with cognitive decline or cognitive impairment with digital cognitive assessment, prevention, treatment and management based on evidence-based medicine. Their efficacy needs to be clinically validated.

[0003] Randomized Controlled Trial (RCT) is a standard method for testing the effectiveness of a drug or therapy, following the three principles of randomization, control, and blinding or masking. The overall design idea is to randomly group the subjects, set up experimental and control groups, and apply different interventions to compare the differences in intervention effects. Generally, the experimental group receives treatment measures (drugs or therapies) and the control group receives placebo interventions. Well-designed RCT studies usually adopt a double-blind design, that is, the subjects do not know whether they are receiving treatment measures or placebos, and the researchers do not know whether a subject is in the experimental or control group. Therefore, the influence of subjective factors of the subjects or researchers on the scientific research results can be eliminated as much as possible, thereby reducing bias. It is generally believed that double-blind placebo randomized controlled trials are the gold standard for evaluating the efficacy of interventions.

[0004] However, compared with traditional clinical trials, the "control" and "blind method" of digital therapy clinical trials are more challenging. In traditional drug experiments, a placebo is a "simulated" drug whose physical properties such as appearance, size, color, dosage form, weight, taste and smell are as similar as possible to the test drug, but it does not contain active ingredients. However, digital therapy products are software products, and there is no "physical drug". The industry currently lacks unified standards to select placebo products that are similar to digital therapies in "physical properties", and it is also difficult to select more suitable digital therapy placebo products.

[0005] On the other hand, due to the limitations of the digital therapy product form, users can easily judge whether they are using real treatment measures or placebo measures based on the audiovisual information. For example, the digital therapy group uses exact software, and the tasks that produce therapeutic effects are diverse and rich in audiovisual information; while the control group uses fake software, single tasks, single audiovisual information, and the tasks are too simple, or even only in the form of popular science messages sent through text messages, WeChat and other media. Therefore, patients with mild cognitive impairment (i.e., clinical trial volunteers) can easily find that the intervention method they are using is a placebo.

[0006] Therefore, there is an urgent need for an effective digital therapeutic placebo product. Summary of the invention

[0007] The purpose of this application is to solve the problem of poor confusion effect of digital therapy placebo products in the prior art. Therefore, this application provides a placebo training method, software and system for cognitive digital therapy, which utilizes the understanding and output capabilities of a large language model for text to build a rich training task library, thereby forming a reasonable placebo training plan with good confusion effect for different experimental groups and corresponding user feature information.

[0008] The present application embodiment provides a placebo training method for cognitive digital therapy, comprising:

[0009] S1. Build a training task library, including:

[0010] S11, constructing an original database; obtaining original data, and converting the original data into an input format that conforms to the large language model for storage after preprocessing the original data;

[0011] S12, creating a vector index of the original database; splitting the data in the original database into small, continuous text segments, converting the split text segments into text vectors, and storing the generated text vectors locally, while creating a vector index;

[0012] S13, constructing a training task generation model; pre-training the large language model so that the large language model generates a variety of standardized task types in the form of dialogue; and then performing intensive training on the large language model based on vector indexes so that the large language model can generate training tasks related to the original data according to the task type;

[0013] S14, using the large language model to generate a variety of training tasks, and storing them in the training task library;

[0014] S2. Label and classify the training tasks, including:

[0015] S21, marking the subject and task type of the training tasks in the training task library;

[0016] S22, calculating the completion time of the training task and marking it on the corresponding training task;

[0017] S3, obtaining characteristic information of the user;

[0018] S4. Determine a training plan for the user based on the characteristic information, and train the user according to the training plan.

[0019] By adopting the above technical solution, a large amount of raw data is obtained and converted into the input format of the large language model so that the large language model can process the raw data; then the raw data is split and vector indexes are created to reduce the storage pressure of the data and the response time of processing the data, which is conducive to the subsequent generation of models for training tasks. By pre-training the large language model first, it is ensured that the large language model outputs accurate and reliable task types in a standardized format; it is avoided that errors in the form of task types cause users to identify placebo training tasks. Then, through intensive training, the content of the training tasks generated by the large language model is improved to ensure that the content of the training tasks is related to the original data and has a certain degree of accuracy and scientificity, and further improve the confusion of the placebo training content; finally, according to the user's characteristic information, a personalized training plan is determined, and the training plan can be matched with the cognitive training plan of the experimental group to ensure the consistency of the placebo training and cognitive training in form.

[0020] In some embodiments, the content of the original data is health knowledge popularization; the types of the original data include text data, graphic data, audio data and video data;

[0021] The original data is obtained from medical databases, literature and network resources, and is stored in the original database after manual review.

[0022] Using the above technical solutions, based on existing academic theories, patients with cognitive impairment can receive comprehensive multi-dimensional intervention measures, including cognitive training (including electronic cognitive training), exercise, lifestyle intervention, chronic disease management, nutritional management, and other multi-faceted intervention measures. Cognitive digital therapy products can provide the above multi-dimensional cognitive interventions and play a role in improving patients' cognitive abilities;

[0023] Health knowledge popularization helps to improve patients' cognitive reserve, subtly regulate and improve patients' lifestyles, nutrition, chronic diseases and other risk factors, and users have a significant sense of gain; moreover, the role of health knowledge popularization is subtle, and the effect on cognitive ability improvement is slow and slight, which is difficult to resist cognitive damage caused by diseases such as Alzheimer's disease. Therefore, determining the original data as health knowledge popularization data is beneficial to the treatment of patients and will not affect the comparison between the placebo training effect and the experimental group; further, health knowledge popularization is highly correlated with the content of patient cognitive digital therapy, and its actual content is extremely confusing, making it difficult for patients (especially those with cognitive impairment) to identify placebo training tasks, further increasing the confusion of placebo training content.

[0024] In some embodiments, the preprocessing of the raw data includes determining the subject words of the data, performing word segmentation, and removing duplicates and noise;

[0025] Convert the text data directly into the input format for storage;

[0026] After extracting text from the graphic data and the audio data, the text is converted into the input format for storage;

[0027] The video data is subjected to image text extraction and sound text extraction respectively, and after being processed, it is converted into the input format for storage.

[0028] By adopting the above technical solution, various types of raw data can be converted into a text input format, which is convenient for subsequent processing using a large language model.

[0029] In some embodiments, the task types include:

[0030] Health management tasks require users to complete reading about health management knowledge;

[0031] The picture and text reading task requires users to read a series of pictures and texts about health knowledge;

[0032] The audio lecture task requires users to listen to an audio lecture on health knowledge in the form of a radio station;

[0033] The knowledge quiz task requires users to complete multiple-choice questions about brain health knowledge and healthy living common sense;

[0034] The step 13 comprises:

[0035] Inputting the vector index of the original data into the large language model for multiple rounds of iterations for pre-training, and testing the large language model in the form of questions to obtain responses;

[0036] Perform supervised fine-tuning on the large language model and issue instructions to the large language model through multiple rounds of dialogue to generate task types in a standardized format;

[0037] Based on the vector index, the large language model is subjected to intensive training, and supervised learning is performed on the accuracy of the training task content generated by the large language model.

[0038] By adopting the above technical solution, the design of task types can help users master relevant health management knowledge, but does not involve the training and improvement of cognitive abilities. It is mainly based on passive acceptance of knowledge popularization. It also includes knowledge question and answer tasks, but does not provide accurate answer feedback. On the one hand, it ensures that the knowledge question and answer tasks have a confusing effect, and on the other hand, the knowledge question and answer tasks have no significant effect on improving cognitive abilities.

[0039] In some embodiments, step S22 includes:

[0040] Construct training ability grading standards based on cognitive ability, education level and age; and collect corresponding task completion time data;

[0041] Build a duration estimation model. Based on the training capability classification standard and task completion time data, build a training duration estimation model and complete the construction of the duration estimation model.

[0042] A duration estimation model is used to predict the training time required for users to complete each training task and mark it.

[0043] By adopting the above technical solution, since the training duration can have a relatively wide range, the user's training ability is graded based on the main factors (cognitive ability, education level and age), and the training duration for different levels of training ability to complete various training tasks is estimated through a duration estimation model, which is conducive to the subsequent reasonable arrangement of different training tasks and the control of the overall training duration of each time according to the user's actual training ability.

[0044] In some embodiments, the training plan includes task content, task cycle, and duration of each training;

[0045] The task cycle and duration of each training session of the training program were the same as those of the experimental group's cognitive training;

[0046] The task content is generated based on the training task library;

[0047] Step S4 includes:

[0048] Based on the characteristic information of the user, determining the level of the user's training ability and determining the training time for the user to complete each task;

[0049] Based on the characteristic information of the user, determining the task content related to the characteristic information;

[0050] Determine the user's training plan, and arrange the number of tasks and task content for each training session so that the duration for the user to complete each task is consistent with the duration of each training session, and the task content is not repeated.

[0051] In some embodiments, step S4 uses a particle swarm optimization algorithm to generate a training plan, specifically including:

[0052] The particle swarm parameters are formed based on the feature information and training plan. The speed is the direction and distance that the particle moves in the next iteration, and the position is the solution to the problem.

[0053] Set the population size N, dimension D, and maximum number of iterations T;

[0054] Initialize the position and velocity of each particle. The position of the i-th particle is X i , the moving vector is V i , the optimal position searched is P i best , its fitness value is f(X i ), the optimal position searched by the particle group is P group best , whose fitness value is f group (X i );

[0055] Then in the tth iteration, the position of the i-th particle is X i t , the moving vector is V i t , the optimal position searched is P i t ,best , its fitness value is f(X i t ),

[0056] The position vector of the particle's next iteration is: V i t+1 =wV i t +r1*(P i t,best -X i t )+r2*(P group t,best -X i t ), r1 and r2 are random numbers in the interval [0, 1],

[0057] Then the updated particle position is: Xi t+1 =X i t +V i t+1 , whose fitness is f(X i t+1 );

[0058] If f(X i t+1 )>f(X i t ), then accept the new solution and continue to the next iteration, otherwise skip the new solution and continue to the next iteration;

[0059] After reaching the maximum number of iterations, the optimal position generated is the training plan of the user.

[0060] The above technical solution adopts a particle swarm optimization algorithm, which has a simple principle, is easy to implement, and requires few parameters to be adjusted, which can reduce the pressure of the algorithm on the computing power demand of the terminal device; at the same time, the particle swarm optimization algorithm does not rely on problem information, uses real numbers for solving, has strong versatility, fast convergence speed, and does not require much memory from the terminal device.

[0061] Moreover, the rapidity of the particle swarm optimization algorithm makes it easier to find the global optimum without being trapped in the local optimum, ensuring the stable generation of feasible training plans.

[0062] In some embodiments, step S3 includes:

[0063] The characteristic information includes:

[0064] Personal information, including name, gender, date of birth, medical history, education level and nutritional indicators; lifestyle information, including smoking history, drinking history, exercise status, work and rest status and diet; cognitive ability information; mental and psychological scores;

[0065] Among them, nutritional indicators include body mass index and waist-to-hip ratio;

[0066] The method of obtaining the characteristic information includes filling it out on your own and obtaining it through testing with an evaluation tool.

[0067] The present application also provides a placebo software for cognitive digital therapy, which is installed in a device terminal and uses the training method of any of the above embodiments to train users.

[0068] By adopting the above technical solution, the placebo software can utilize the large language model's ability to understand and output text and build a rich library of training tasks, thereby forming a reasonable placebo training plan with good confusion effect for different experimental groups and the characteristic information of corresponding users.

[0069] The present application also provides a placebo system for cognitive digital therapy, comprising:

[0070] Task generation module, including large language model and training task library;

[0071] An interactive module, used to interact with the user, and the user completes the training task through the interactive module;

[0072] The control module is connected with the task generation module and the interaction module, and adopts the training method of any one of the above embodiments to train the user.

[0073] By adopting the above technical solution, it is possible to utilize the large language model's ability to understand and output text and build a rich training task library, thereby forming a reasonable placebo training plan with good confusion effect for different experimental groups and corresponding user feature information.

[0074] Other features and corresponding beneficial effects of the present invention are described in the latter part of the specification, and it should be understood that at least part of the beneficial effects become obvious from the description in the specification of the present invention. DETAILED DESCRIPTION

[0075] The following specific embodiments illustrate the implementation of the present invention, and those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. Although the description of the present invention will be introduced in conjunction with the preferred embodiment, this does not mean that the features of this invention are limited to this implementation. On the contrary, the purpose of introducing the invention in conjunction with the implementation is to cover other options or modifications that may extend based on the claims of the present invention. In order to provide a deep understanding of the present invention, the following description will include many specific details. The present invention can also be implemented without using these details. In addition, in order to avoid confusion or blurring the focus of the present invention, some specific details will be omitted in the description. It should be noted that the embodiments of the present invention and the features in the embodiments can be combined with each other without conflict.

[0076] Obviously, the described embodiments are only some embodiments of the present invention, but not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0077] In the description of the present invention, it should be noted that the terms "first", "second", and "third" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, a feature defined as "first", "second", etc. may explicitly or implicitly include one or more of the features. In the description of the invention of this application, unless otherwise specified, "plurality" means two or more.

[0078] In the description of the present invention, it should be noted that, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0079] In order to make the objectives, technical solutions and advantages of the present invention more clear, the embodiments of the present invention are described in further detail below.

[0080] The present application embodiment provides a placebo training method for cognitive digital therapy, comprising:

[0081] S1. Build a training task library, including:

[0082] S11, constructing an original database; obtaining original data, and converting the original data into an input format that conforms to the large language model for storage after preprocessing the original data;

[0083] S12, creating a vector index of the original database; splitting the data in the original database into small, continuous text segments, converting the split text segments into text vectors, and storing the generated text vectors locally, while creating a vector index;

[0084] S13, constructing a training task generation model; pre-training the large language model so that the large language model can generate a variety of standardized task types in the form of dialogue; then based on the vector index, intensive training is performed on the large language model so that the large language model can generate training tasks related to the original data according to the task type;

[0085] S14, using the large language model to generate a variety of training tasks, and storing them in the training task library;

[0086] S2. Label and classify the training tasks, including:

[0087] S21, marking the subject and task type of the training task in the training task library;

[0088] S22, calculating the completion time of the training task and marking it on the corresponding training task;

[0089] S3, obtaining characteristic information of the user;

[0090] S4. Determine a training plan for the user based on the feature information, and train the user according to the training plan.

[0091] By adopting the above technical solution, a large amount of raw data is obtained and converted into the input format of the large language model so that the large language model can process the raw data; then the raw data is split and vector indexes are created to reduce the storage pressure of the data and the response time of processing the data, which is conducive to the subsequent generation of models for training tasks. By pre-training the large language model first, it is ensured that the large language model outputs accurate and reliable task types in a standardized format; it is avoided that errors in the form of task types cause users to identify placebo training tasks. Then, through intensive training, the content of the training tasks generated by the large language model is improved to ensure that the content of the training tasks is related to the original data and has a certain degree of accuracy and scientificity, and further improve the confusion of the placebo training content; finally, according to the user's characteristic information, a personalized training plan is determined, and the training plan can be matched with the cognitive training plan of the experimental group to ensure the consistency of the placebo training and cognitive training in form.

[0092] In order to better illustrate the above training method, the following is a detailed description taking the placebo software installed in the device terminal and using the above training method as an example.

[0093] In one embodiment, the content of the original data is health knowledge popularization; the types of the original data include text data, graphic data, audio data and video data;

[0094] The original data were obtained from medical databases, literature and online resources, and were stored in the original database after manual review.

[0095] In one embodiment, the content of the original data may include brain health, diet and nutrition, sleep, chronic diseases, disease prevention, emotions, healthy lifestyle, etc. After being reviewed by medical professionals, it is included in the original database, and the inclusion criteria may be correct, medically meaningful, and brain health-related content.

[0096] It should be noted that based on existing academic theories (including the WHO Guidelines for Reducing the Risk of Cognitive Decline and Dementia (2019), the China Guidelines for Primary Prevention of Alzheimer's Disease (2020), the 2018 China Guidelines for the Diagnosis and Treatment of Dementia and Cognitive Impairment, the China Guidelines for Cognitive Training (2022 Edition), the China Expert Consensus on the Diagnosis and Treatment of Mild Cognitive Impairment of Alzheimer's Disease 2021, and the 2018 Practice guideline update summary: Mild cognitive impairment), patients with cognitive impairment can undergo comprehensive multi-dimensional intervention measures, including cognitive training (including electronic cognitive training), physical exercise, lifestyle intervention, chronic disease management, nutritional management and other intervention measures. Cognitive digital therapy products can provide the above-mentioned multi-dimensional cognitive interventions and play a role in improving patients' cognitive abilities.

[0097] Health knowledge popularization helps to improve patients' cognitive reserve, subtly regulate and improve patients' lifestyles, nutrition, chronic diseases and other risk factors, and users have a significant sense of gain; moreover, the role of health knowledge popularization is subtle, and the effect on cognitive ability improvement is slow and slight, which is difficult to resist cognitive damage caused by diseases such as Alzheimer's disease. Therefore, determining the original data as health knowledge popularization data is beneficial to the treatment of patients and will not affect the comparison between the placebo training effect and the experimental group; further, health knowledge popularization is highly correlated with the content of patient cognitive digital therapy, and its actual content is extremely confusing, making it difficult for patients (especially those with cognitive impairment) to identify placebo training tasks, further increasing the confusion of placebo training content.

[0098] In one embodiment, the preprocessing of the raw data includes determining the subject words of the data, performing word segmentation, and removing duplicates and noise;

[0099] Convert text data directly into input format for storage;

[0100] After text extraction of graphic data and audio data, the data is converted into input format for storage;

[0101] The video data is extracted for image text and sound text respectively, and then processed and converted into input format for storage. By converting various types of raw data into text input format, it is convenient for subsequent processing using a large language model.

[0102] In one embodiment, the task types include:

[0103] Health management tasks require users to complete reading about health management knowledge;

[0104] The picture and text reading task requires users to read a series of pictures and texts about health knowledge;

[0105] The audio lecture task requires users to listen to an audio lecture on health knowledge in the form of a radio station;

[0106] The knowledge quiz task requires users to complete multiple-choice questions about brain health knowledge and healthy living common sense;

[0107] Step 13 includes:

[0108] The vector index of the original data is input into the large language model for multiple rounds of iterations for pre-training, and the large language model is tested in the form of questions to obtain responses;

[0109] Perform supervised fine-tuning on the large language model and issue instructions to the large language model through multiple rounds of dialogue to generate task types in a standardized format;

[0110] Based on vector indexing, the large language model is trained intensively, and supervised learning is performed on the accuracy of the training task content generated by the large language model.

[0111] The design of the above task types can help users master relevant health management knowledge, but does not involve the training and improvement of cognitive abilities. It is mainly based on passive acceptance of knowledge popularization. It also includes knowledge question and answer tasks, but does not provide accurate answer feedback. On the one hand, it ensures that the knowledge question and answer tasks have a confusing effect, and on the other hand, the knowledge question and answer tasks have no significant effect on improving cognitive abilities.

[0112] It should be noted that the above task type is an implementation case cited, representing the main task types; in other alternative implementations, the task type may also include other unmentioned task types related to health management knowledge but not involving cognitive abilities.

[0113] In one embodiment, the large language model may use ChatGPT, and the input format may be set to conform to the input format of ChatGPT, so that a vector index may be created using GPT Index and stored locally (eg, in csv format).

[0114] It should be noted that ChatGPT is an existing mature large language model with powerful language generation and understanding capabilities. It can be used to answer various questions, provide information and suggestions, and even participate in conversations and creations. At the same time, the source of the data in the original database of the embodiment of the present application is within the data range used to train ChatGPT; therefore, ChatGPT can be directly applied to the embodiment of the present application, reducing the cost of retraining the model. Pre-training and reinforcement training of ChatGPT can be completed with less computing power requirements.

[0115] Furthermore, the accuracy, coherence, readability, feasibility, diversity and emotion in the audio of the training task content generated by ChatGPT can be continuously and long-term improved through supervised fine-tuning methods, thereby continuously optimizing the training task library and improving the quality of the generated training task content.

[0116] In one embodiment, step S22 includes:

[0117] Construct training ability grading standards based on cognitive ability, education level and age; and collect corresponding task completion time data;

[0118] Build a duration estimation model. Based on the training capability classification standard and task completion time data, build a training duration estimation model and complete the construction of the duration estimation model.

[0119] A duration estimation model is used to predict the training time required for users to complete each training task and mark it.

[0120] By adopting the above technical solution, since the training duration can have a relatively wide range, the user's training ability is graded based on the main factors (cognitive ability, education level and age), and the training duration for different levels of training ability to complete various training tasks is estimated through a duration estimation model, which is conducive to the subsequent reasonable arrangement of different training tasks and the control of the overall training duration of each time according to the user's actual training ability.

[0121] In one embodiment, cognitive ability can be divided into four levels from mild to severe, including completely normal, suspected decline (or subjective cognitive decline), mild decline (mild cognitive impairment) and dementia.

[0122] In one embodiment, the training plan includes task content, task cycle, and duration of each training;

[0123] The task cycle and duration of each training session in the training plan were the same as those of the experimental group’s cognitive training;

[0124] The task content is generated based on the training task library;

[0125] Step S4 includes:

[0126] Based on the user's characteristic information, determine the user's training ability level and determine the training time for the user to complete each task;

[0127] Based on the characteristic information of the user, determine the task content related to the characteristic information;

[0128] Determine the user's training plan, arrange the number and content of tasks for each training session, so that the time it takes for the user to complete each task is consistent with the duration of each training session, and the task content is not repeated.

[0129] It should be noted that the training plan needs to be the same as the cognitive training of the experimental group. For example, according to the "Chinese Guidelines for Cognitive Training (2022 Edition)", cognitive digital therapy for Alzheimer's patients should provide cognitive training at least 3 times a week, each time for 15-24 minutes.

[0130] Therefore, in one embodiment, the task cycle and duration of each training session of the training plan can be set to be the same as that of the experimental group, at least 3 times per week and 15-24 minutes per training session.

[0131] In other alternative embodiments, adjustments can also be made based on clinical experiments, and the task cycle of the training plan and the duration of each training session can also be set to 5 training sessions per week and 15-20 minutes per training session, but all need to be the same as the experimental group.

[0132] In one embodiment, step S4 uses a particle swarm optimization algorithm to generate a training plan, specifically including:

[0133] The particle swarm parameters are formed based on the feature information and training plan. The speed is the direction and distance that the particle moves in the next iteration, and the position is the solution to the problem.

[0134] Set the population size N, dimension D, and maximum number of iterations T;

[0135] Initialize the position and velocity of each particle. The position of the i-th particle is X i , the moving vector is V i , the optimal position searched is P i best , its fitness value is f(X i ), the optimal position searched by the particle group is P group best , whose fitness value is f group (X i );

[0136] Then in the tth iteration, the position of the i-th particle is X i t , the moving vector is V i t , the optimal position searched is P i t ,best , its fitness value is f(X i t ),

[0137] The position vector of the particle's next iteration is: V i t+1 =wV i t +r1*(P i t,best -X i t )+r2*(P group t,best -X i t ), r1 and r2 are random numbers in the interval [0, 1],

[0138] Then the updated particle position is: X i t+1 =X i t +V i t+1 , whose fitness is f(X i t+1 );

[0139] If f(X i t+1 )>f(X i t ), then accept the new solution and continue to the next iteration, otherwise skip the new solution and continue to the next iteration;

[0140] After reaching the maximum number of iterations, the optimal position generated is the user's training plan.

[0141] The above technical solution adopts a particle swarm optimization algorithm, which has a simple principle, is easy to implement, and requires few parameters to be adjusted, which can reduce the pressure of the algorithm on the computing power demand of the terminal device; at the same time, the particle swarm optimization algorithm does not rely on problem information, uses real numbers for solving, has strong versatility, fast convergence speed, and does not require much memory from the terminal device.

[0142] Moreover, the rapidity of the particle swarm optimization algorithm makes it easier to find the global optimum without being trapped in the local optimum, ensuring the stable generation of feasible training plans.

[0143] In one embodiment, step S3 includes:

[0144] Feature information includes:

[0145] Personal information, including name, gender, date of birth, medical history, education level and nutritional indicators; lifestyle information, including smoking history, drinking history, exercise status, work and rest status and diet; cognitive ability information; mental and psychological scores;

[0146] Among them, nutritional indicators include body mass index and waist-to-hip ratio;

[0147] The methods for obtaining characteristic information include filling it out on your own and obtaining it through testing with assessment tools.

[0148] In one embodiment, the characteristic information can be obtained through evaluation and information collection when the user registers the software. Specifically, the medical history may include the presence or absence of hypertension, diabetes, cardiovascular and cerebrovascular diseases, neuropsychiatric diseases, hyperlipidemia and other diseases, as well as the corresponding medication history;

[0149] Body Mass Index (BMI) in nutritional indicators = weight ÷ height 2 BMI < 18.5 is malnutrition, 18.5 ≤ BMI < 25 is basically normal, 25 ≤ BMI < 30 is overweight, and BMI ≥ 30 is obese. Waist-hip ratio (WHR) is waist circumference divided by hip circumference, male ≥ 0.95, female ≥ 0.80, above this standard is considered obesity risk.

[0150] Exercise status can include the number of exercises per week, exercise type and exercise duration.

[0151] Dietary conditions may include appetite, number of meals, and food types, and the information source can be collected through a dietary diversity assessment scale.

[0152] Cognitive ability information can be assessed through various cognitive function assessment tools such as MoCA and MMSE.

[0153] Mental health scores can be collected through the Self-Rating Depression Scale, the Self-Rating Anxiety Scale, the Pittsburgh Sleep Quality Questionnaire, etc.

[0154] It should be noted that the characteristic information is related to the content and form of the training tasks. Since the total time expected to complete the training on the day is relatively fixed, it is necessary to allocate reasonable training tasks based on the characteristic information. For example, in clinical trials, patients with low education levels and low cognitive abilities may take a long time to complete a single training task. Therefore, it is necessary to arrange different types of tasks with appropriate training durations based on the prescribed training duration (such as 30 minutes per day); at the same time, corresponding disease or healthy lifestyle tasks will be allocated based on the patient's medical history, lifestyle, diet, nutrition and other characteristics.

[0155] The present application also provides a placebo system for cognitive digital therapy, including:

[0156] Task generation module, including large language model and training task library;

[0157] An interactive module, used to interact with the user, and the user completes the training task through the interactive module;

[0158] The control module is connected with the task generation module and the interaction module, and adopts the training method of any one of the above embodiments to train the user.

[0159] The placebo system in the above embodiment can utilize the large language model's ability to understand and output text to build a rich training task library, so as to form a reasonable placebo training plan with good confusion effect for different experimental groups and corresponding user feature information.

[0160] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A placebo training method for cognitive digital therapy, characterized in that: include: S1. Build a training task library, including: S11, constructing an original database; obtaining original data, and converting the original data into an input format that conforms to the large language model for storage after preprocessing the original data; S12, creating a vector index of the original database; splitting the data in the original database into small, continuous text segments, converting the split text segments into text vectors, and storing the generated text vectors locally, while creating a vector index; S13, constructing a training task generation model; pre-training the large language model so that the large language model generates a variety of standardized task types in the form of dialogue; and then performing intensive training on the large language model based on vector indexes so that the large language model can generate training tasks related to the original data according to the task type; S14, using the large language model to generate a variety of training tasks, and storing them in the training task library; S2. Label and classify the training tasks, including: S21, marking the subject and task type of the training tasks in the training task library; S22, calculating the completion time of the training task and marking it on the corresponding training task; S3, obtaining characteristic information of the user; S4. Determine a training plan for the user based on the characteristic information, and train the user according to the training plan.

2. A placebo training method for cognitive digital therapy according to claim 1, characterized in that: The content of the original data is health knowledge popularization; the types of the original data include text data, graphic data, audio data and video data; The original data is obtained from medical databases, literature and network resources, and is stored in the original database after manual review.

3. A placebo training method for cognitive digital therapy according to claim 2, characterized in that: The preprocessing of the raw data includes determining the subject words of the data, performing word segmentation, and removing duplicates and noise; Convert the text data directly into the input format for storage; After extracting text from the graphic data and the audio data, the text is converted into the input format for storage; The video data is subjected to image text extraction and sound text extraction respectively, and after being processed, it is converted into the input format for storage.

4. A placebo training method for cognitive digital therapy according to claim 1, characterized in that: The task types include: Health management tasks require users to complete reading about health management knowledge; The picture and text reading task requires users to read a series of pictures and texts about health knowledge; The audio lecture task requires users to listen to an audio lecture on health knowledge in the form of a radio station; The knowledge quiz task requires users to complete multiple-choice questions about brain health knowledge and healthy living common sense; The step 13 comprises: Inputting the vector index of the original data into the large language model for multiple rounds of iterations for pre-training, and testing the large language model in the form of questions to obtain responses; Perform supervised fine-tuning on the large language model and issue instructions to the large language model through multiple rounds of dialogue to generate task types in a standardized format; Based on the vector index, the large language model is subjected to intensive training, and supervised learning is performed on the accuracy of the training task content generated by the large language model.

5. A placebo training method for cognitive digital therapy according to claim 1, characterized in that: Step S22 includes: Construct training ability grading standards based on cognitive ability, education level and age; and collect corresponding task completion time data; Build a duration estimation model. Based on the training capability classification standard and task completion time data, build a training duration estimation model and complete the construction of the duration estimation model. A duration estimation model is used to predict the training time required for users to complete each training task and mark it.

6. A placebo training method for cognitive digital therapy according to claim 5, characterized in that: The training plan includes task content, task cycle and duration of each training; The task cycle and duration of each training session of the training program were the same as those of the experimental group's cognitive training; The task content is generated based on the training task library; Step S4 includes: Based on the characteristic information of the user, determining the level of the user's training ability and determining the training time for the user to complete each task; Based on the characteristic information of the user, determining the task content related to the characteristic information; Determine the user's training plan, and arrange the number of tasks and task content for each training session so that the duration for the user to complete each task is consistent with the duration of each training session, and the task content is not repeated.

7. A placebo training method for cognitive digital therapy according to claim 6, characterized in that: Step S4 uses a particle swarm optimization algorithm to generate a training plan, which specifically includes: The particle swarm parameters are constructed based on the characteristic information. The speed is the direction and distance that the particle moves in the next iteration, and the position is the solution to the problem. Set the population size N, dimension D, and maximum number of iterations T; Initialize the position and velocity of each particle. The position of the i-th particle is X i , the moving vector is V i , the optimal position searched is P i best , its fitness value is f(X i ), the optimal position searched by the particle group is P group best , whose fitness value is f group (X i ); Then in the tth iteration, the position of the i-th particle is X i t , the moving vector is V i t , the optimal position searched is P i t,best , its fitness value is f(X i t ), The position vector of the particle's next iteration is: V i t+1 =wV i t +r1*(P i t,best -X i t )+r2*(P group t,best -X i t ), r1 and r2 are random numbers in the interval [0, 1], Then the updated particle position is: X i t+1 =X i t +V i t+1 , whose fitness is f(X i t+1 ); If f(X i t+1 )>f(X i t ), then accept the new solution and continue to the next iteration, otherwise skip the new solution and continue to the next iteration; After reaching the maximum number of iterations, the optimal position generated is the training plan for the user.

8. A placebo training method for cognitive digital therapy according to claim 1, characterized in that: Step S3 includes: The characteristic information includes: Personal information, including name, gender, date of birth, medical history, education level and nutritional indicators; lifestyle information, including smoking history, drinking history, exercise status, work and rest status and diet; cognitive ability information; mental and psychological scores; Among them, nutritional indicators include body mass index and waist-to-hip ratio; The method of obtaining the characteristic information includes filling it out on your own and obtaining it through testing with an evaluation tool.

9. A placebo software for cognitive digital therapy, characterized in that The device is installed on a device terminal and uses the training method described in any one of claims 1 to 8 to train the user.

10. A placebo system for cognitive digital therapy, characterized in that include: Task generation module, including large language model and training task library; An interactive module, used to interact with the user, and the user completes the training task through the interactive module; The control module is connected with the task generation module and the interaction module, and uses the training method as described in any one of claims 1 to 8 to train the user.