Method, device and recording medium for supporting clinical decisions by using generative artificial intelligence model

Abstract

A method by which an electronic device supports clinical decisions by using a generative artificial intelligence model, according to one embodiment of the present disclosure, may comprise the steps of: inputting, as a prompt, user text input from a user terminal to the generative artificial intelligence model to which a first prompting technique and / or a second prompting technique are / is applied; acquiring the result of the generative artificial intelligence model for the prompt; and providing the result of the generative artificial intelligence model to the user terminal and / or an expert terminal.

Description

Method, device, and recording medium for supporting clinical decisions using a generative artificial intelligence model

[0001] The present disclosure relates to a technology that supports clinical decisions regarding depression using a generative artificial intelligence model.

[0002] Depression is a significant global issue that leads to various social consequences, including reduced job productivity and a high disability burden. As the early detection and intervention of clinically significant depression become increasingly important, efforts are underway to overcome the limitations of objectivity and accuracy in existing depression screening tools.

[0003] Recently, researchers have been striving to identify objective indicators of depression through image analysis, blood biomarkers, and Ecological Moment Assessments (EMA). In particular, among Ecological Moment Assessments, user-generated text data, especially diaries, is garnering attention as a clinically important and analyzable source for detecting or diagnosing depression, alongside the advancement of Large Language Models (LLM).

[0004] However, existing large-scale language model-based depression analysis technologies still have limitations. For example, while large-scale language models can grasp the surface meaning of text, they sometimes fail to fully understand contextual information and the user's hidden intentions. This reduces the accuracy of depression diagnosis. Furthermore, although large-scale language models can extract and analyze depression-related information from text data, they may struggle to deeply infer and diagnose the user's condition based on this data.

[0005] The present disclosure overcomes these limitations and provides a technology that supports the diagnosis of depression by using a generative artificial intelligence model including a large-scale language model to analyze users' diary data more accurately and in depth.

[0006] The present disclosure provides a technology that supports clinical decisions regarding depression using a generative artificial intelligence model.

[0007] The technical problems to be solved in this disclosure are not limited to those mentioned above, and various unmentioned technical problems can be inferred by a person skilled in the art from this disclosure.

[0008] A method for supporting clinical decisions using a generative artificial intelligence model by an electronic device according to one embodiment of the present disclosure may include the steps of inputting user text input from a user terminal as a prompt into a generative artificial intelligence model to which at least one of a first prompting technique or a second prompting technique is applied, obtaining a result of the generative artificial intelligence model for the prompt, and providing the result of the generative artificial intelligence model to at least one of the user terminal or an expert terminal.

[0009] In one embodiment, the generative artificial intelligence model may include a Large Language Model (LLM).

[0010] In one embodiment, the first prompting technique may include at least one of zero-shot prompting, few-shot prompting, the GNU Zip algorithm, or k-nearest clustering.

[0011] In one embodiment, the second prompting technique may include Chain-of-Thought (CoT) prompting.

[0012] In one embodiment, the thought chain prompting may include the operation of extracting a first morpheme having a conditional meaning from the user text, the operation of extracting a second morpheme having a response meaning from the user text, and the operation of determining whether there is depression based on at least one of the first morpheme or the second morpheme.

[0013] In one embodiment, the first morpheme may be a morpheme having the meaning of the condition, including situation, way of thinking, and attitude.

[0014] In one embodiment, the second morpheme may be a morpheme having the meaning of the reaction, including emotion, opinion, and intention regarding the condition.

[0015] In one embodiment, the thought chain prompting may further include the operation of extracting a third morpheme having the meaning of depression intensity from the user text, and the operation of determining the depression intensity based on the third morpheme.

[0016] In one embodiment, the depression intensity is classified into mild, moderate, and severe, and the third morpheme may include a morpheme having the meaning of mild, including fatigue and motivation; a morpheme having the meaning of moderate, including anxiety and nervousness; and a morpheme having the meaning of severe, including suicide, hallucination, and delusion.

[0017] In one embodiment, if the depression intensity obtained as a result of the generative artificial intelligence model is moderate or higher, an alarm may be generated on at least one of the user terminal or the expert terminal.

[0018] In one embodiment, the user text may be text entered as a diary at the user terminal.

[0019] An electronic device according to one embodiment of the present disclosure comprises one or more processors and one or more memories for storing at least one instruction executed by the one or more processors, and the one or more processors may be configured to perform a method of any one of the embodiments described above by executing the at least one instruction.

[0020] A non-transient computer-readable recording medium according to one embodiment of the present disclosure may include at least one instruction that causes one or more processors to perform a method of any one of the embodiments described above when executed by one or more processors.

[0021] According to one embodiment of the present disclosure, by analyzing a user's diary data using a generative artificial intelligence model, particularly a large-scale language model (LLM), the limitations of existing diagnostic methods that rely on subjective judgment are overcome, and an objective and accurate diagnosis of depression is made possible.

[0022] In addition, according to one embodiment, the reasoning ability of a generative artificial intelligence model can be enhanced by utilizing a chain of thought prompting technique, and by extracting and analyzing morphemes having the meaning of conditions and responses from user text, the psychological state of the user can be understood in depth and personalized support can be provided.

[0023] In addition, according to one embodiment, by providing the analysis results of a generative artificial intelligence model to an expert, it is possible to support the expert in establishing a diagnosis and treatment plan.

[0024] The effects according to the technical concept of the present disclosure are not limited to the effects mentioned above, and various unmentioned effects can be clearly understood by a person skilled in the art from the present disclosure.

[0025] FIG. 1 is a drawing illustrating an environment in which an electronic device according to one embodiment of the present disclosure can be applied.

[0026] FIG. 2 is a block diagram of an electronic device according to one embodiment of the present disclosure.

[0027] FIG. 3 is a flowchart illustrating a method for supporting clinical decisions using a generative artificial intelligence model according to one embodiment of the present disclosure.

[0028] FIG. 4 is a table showing the test results of a generative artificial intelligence model according to one embodiment of the present disclosure.

[0029] FIG. 5a is an exemplary drawing in which a generative artificial intelligence model according to one embodiment of the present disclosure is used.

[0030] FIG. 5b is an exemplary drawing in which a generative artificial intelligence model according to one embodiment of the present disclosure is used.

[0031] FIG. 6 is an exemplary drawing in which a generative artificial intelligence model according to one embodiment of the present disclosure is used.

[0032] FIG. 7 is a drawing illustrating an environment in which a generative artificial intelligence model according to one embodiment of the present disclosure is utilized.

[0033] Hereinafter, embodiments are described in detail with reference to the attached drawings. However, various modifications may be made to the embodiments, and thus the scope of the patent application is not limited or restricted by these embodiments. It should be understood that all modifications, equivalents, and substitutions to the embodiments are included within the scope of the rights.

[0034] Specific structural or functional descriptions of the embodiments are disclosed for illustrative purposes only and may be modified and implemented in various forms. Accordingly, the embodiments are not limited to the specific disclosed forms, and the scope of this specification includes modifications, equivalents, or substitutions that fall within the technical concept.

[0035] Terms such as "first" or "second" may be used to describe various components, but these terms should be interpreted solely for the purpose of distinguishing one component from another. For example, the first component may be named the second component, and similarly, the second component may be named the first component.

[0036] When it is stated that a component is "connected" to another component, it should be understood that it may be directly connected to or joined to that other component, or that there may be other components in between.

[0037] The terms used in the embodiments are for illustrative purposes only and should not be interpreted as intended to be limiting. Singular expressions include plural expressions unless the context clearly indicates otherwise. In this specification, terms such as "comprising" or "having" are intended to indicate the existence of the features, numbers, steps, actions, components, parts, or combinations thereof described in the specification, and should be understood as not precluding the existence or addition of one or more other features, numbers, steps, actions, components, parts, or combinations thereof.

[0038] Expressions used in the present disclosure, such as “A, B, and C”, “A, B, or C”, “A, B, and / or C”, “at least one of A, B, and C”, “at least one of A, B, or C”, “at least one of A, B, and / or C”, “at least one selected from A, B, and C”, “at least one selected from A, B, or C”, “at least one selected from A, B, and / or C”, etc., may mean each or all possible combinations thereof. For example, “at least one of A or B” may refer to (1) at least one A, (2) at least one B, and (3) at least one A and at least one B.

[0039] Expressions such as “based on” used in the present disclosure are used to describe one or more elements affecting an act or action of a decision or judgment described in the phrase or sentence containing such expression, and such expressions do not exclude additional elements affecting said act or action of a decision or judgment.

[0040] Expressions such as “configured to” used in this disclosure may, depending on the context, have meanings such as “set to,” “capable of,” “modified to,” “made to,” or “capable of.” These expressions are not limited to the meaning of “specifically designed in hardware,” and, for example, a processor (or electronic device) configured to perform a specific operation may mean a general-purpose processor capable of performing that specific operation by executing software, or a special-purpose computer structured through programming to perform that specific operation.

[0041] Unless otherwise defined, all terms used herein, including technical or scientific terms, have the same meaning as generally understood by those skilled in the art to which the embodiments pertain. Terms such as those defined in commonly used dictionaries should be interpreted as having a meaning consistent with their meaning in the context of the relevant technology, and should not be interpreted in an ideal or overly formal sense unless explicitly defined in this application.

[0042] In addition, when describing with reference to the attached drawings, identical components are assigned the same reference numeral regardless of drawing symbols, and redundant descriptions thereof are omitted. In describing the embodiments, if it is determined that a detailed description of related prior art could unnecessarily obscure the essence of the embodiments, such detailed description is omitted.

[0043] Hereinafter, the contents of the present disclosure will be explained in more detail with reference to the attached drawings.

[0044] FIG. 1 is a drawing illustrating an environment in which an electronic device according to one embodiment of the present disclosure can be applied.

[0045] In one embodiment, the electronic device (100) can communicate with the user terminal (200) and the expert terminal (300) via a network. Here, the network is for data communication between the electronic device (100), the user terminal (200), and the expert terminal (300), and can be implemented via a wired or wireless connection. The network can be implemented as any type of wired or wireless network, such as, for example, a Local Area Network (LAN), a Wide Area Network, a Mobile Radio Communication Network, or a Wireless Broadcasting Internet (Wibro).

[0046] In one embodiment, the electronic device (100) may be implemented as one or more computing devices. For example, all functions of the electronic device (100) may be implemented in one computing device, and as another example, a first function of the electronic device (100) may be implemented in a first computing device and a second function may be implemented in a second computing device. As yet another example, multiple computing devices may be used to each implement all functions or specific functions of the electronic device (100). Such computing devices may include a desktop computer, a laptop computer, a smartphone, an application server, a proxy server, or a cloud server. The configuration of the computing devices is not limited thereto and may be configured in various ways within the scope of the problem to be solved by the present disclosure.

[0047] In one embodiment, the electronic device (100) communicates with a user terminal (200) and an expert terminal (300) and can support clinical decisions using a generative artificial intelligence model. Specifically, user text entered from the user terminal (200) is transmitted to the electronic device (100), and the electronic device (100) inputs the user text as a prompt to a generative artificial intelligence model to which a predetermined prompting technique is applied, and causes the generative artificial intelligence model to generate a result based on the prompt, thereby providing the result to at least one of the user terminal (200) or the expert terminal (300). The specific operational relationship between the electronic device (100), the user terminal (200), and the expert terminal (300) will be described later.

[0048] In one embodiment, the electronic device (100) can manage information regarding a user using the user terminal (200), for example, a diary writer or a patient. Specifically, the electronic device (100) may include a database that stores and manages user information received from the user terminal (200). This database may include the user's personal information, health information, diary writing history, analysis results of a generative artificial intelligence model, etc., and may be securely managed through security technologies such as encryption and access control. For example, text entered in the form of a diary from the user terminal (200) may be stored in the database of the electronic device (100) along with user information. The electronic device (100) may analyze the stored diary data to track changes in the intensity and pattern of depression and provide personalized health management services based on this.

[0049] In one embodiment, the electronic device (100) may manage information regarding a professional using the professional terminal (300), for example, a psychiatrist. Specifically, the electronic device (100) may include a database that stores and manages a professional profile containing information such as the professional's qualifications, field of expertise, available consultation hours, and contact information. This information may be received directly through the professional terminal (300) or provided by an external agency. Additionally, information such as a diagnosis or opinion by a psychiatrist may be entered into the professional terminal (300), and the electronic device (100) may provide a remote diagnosis service by providing the said information to the user terminal (200).

[0050] In one embodiment, an electronic device (100) can efficiently manage and operate a high-performance generative artificial intelligence model. Specifically, the electronic device (100) may include hardware and software components for storing and executing a generative artificial intelligence model that includes a Large Language Model (LLM). For example, it may include computing devices such as a high-performance GPU, TPU, and NPU, large memory, and an optimized software environment for efficient model execution. Additionally, the electronic device (100) may perform the function of continuously monitoring the performance of the generative artificial intelligence model and updating or retraining the model as needed. For example, it may evaluate the model's accuracy, processing speed, resource usage, etc., fine-tune the model using new data, or optimize performance by replacing it with the latest Large Language Model. Additionally, the electronic device (100) may selectively apply prompting techniques (e.g., zero-shot, few-shot, GNU Zip algorithm, k-closest clustering, chain of thought prompting, etc.) to improve the quality of the results of the generative artificial intelligence model. In this way, the electronic device (100) can provide accurate and reliable support services to experts in making clinical decisions by systematically managing and operating a generative artificial intelligence model.

[0051] In one embodiment, the user terminal (200) and the expert terminal (300) may be implemented as terminals capable of transmitting and receiving various information to and from the electronic device (100) via a network. For example, the user terminal (200) and the expert terminal (300) may be terminals capable of network communication that include an input / output interface, such as a smartphone, a tablet PC, a portable multimedia device, a wearable device, a desktop computer, or a laptop computer. The user terminal (200) and the expert terminal (300) may be identical or different from each other, and may be independently selected as appropriate terminals according to the respective environments of the user and the expert.

[0052] In one embodiment, the user terminal (200) may perform the function of receiving input data from the user in the form of text, voice, images, etc., and transmitting it to the electronic device (100). For example, the user may input a diary in text form in which they freely describe their emotions or thoughts, or input a diary using a voice recording function, or describe their state. In addition, the user terminal (200) may provide the user with the analysis results of a generative artificial intelligence model received from the electronic device (100). The analysis results may be displayed in various forms such as text, graphs, and images, and may include additional explanations or information to help the user understand. Furthermore, the user terminal (200) may provide various additional functions for the user's convenience. For example, it may include a diary writing function, an emotion recording function, a sleep pattern analysis function, a medication reminder function, and a consultation scheduling function with a professional. Through these functions, the user terminal (200) can effectively manage the user's mental health status and support them in receiving appropriate help when necessary.

[0053] In one embodiment, the expert terminal (300) may provide various functions that support experts, including psychiatrists, in verifying the analysis results of a generative artificial intelligence model, communicating with users, including diary writers or patients, and taking necessary measures. Specifically, the expert terminal (300) may perform the function of providing the analysis results of the generative artificial intelligence model received from the electronic device (100) to the expert. The analysis results may include whether the user has depression and the severity of depression, and further include trends in emotional state changes, warning signs, etc., and these analysis results may be displayed in various forms such as graphs, charts, and text. This may contribute to enabling the expert to accurately identify the user's condition based on the analysis results and establish an appropriate treatment plan. Additionally, the expert terminal (300) may provide a function that allows the expert to communicate directly with the user through the user terminal (200). For example, by enabling chatting, video calls, text messages, and alarm generation between the user terminal (200) and the expert terminal (300), the user may be able to proceed with consultation with the expert or check additional information.

[0054] FIG. 2 is a block diagram of an electronic device according to one embodiment of the present disclosure.

[0055] In one embodiment, the electronic device (100) may be a server device or an operating entity for operating a service that supports clinical decisions using a generative artificial intelligence model. Specifically, the electronic device (100) may include one or more processors (110), one or more memories (120), and a communication interface (130) as components. In one embodiment, at least one of the components of the electronic device (100) may be omitted, or another component may be added to the electronic device (100). For example, the communication interface (130) may be included in the electronic device (100), or may be mounted on the outside of the electronic device (100) or connected in a detachable manner. In one embodiment, additionally or substantially, some of the components may be implemented as an integrated unit or as a single or multiple entity. In one embodiment, at least some of the components inside and outside the electronic device (100) may be connected to each other via a bus, GPIO (General Purpose Input / Output), SPI (Serial Peripheral Interface), or MIPI (Mobile Industry Processor Interface), etc., to exchange information including data, signals, etc.

[0056] In one embodiment, the processor (110) may include one or more processors. The processor (110) may control at least one component of an electronic device (100) connected to the processor (110) by running software (e.g., instructions, programs, etc.). Additionally, the processor (110) may perform various operations related to the present disclosure, such as computation, processing, data generation, and processing. Additionally, the processor (110) may retrieve data, etc. from memory (120) or store it in memory (120). Specifically, the processor (110) may execute instructions stored in memory (120) to process data related to at least one of a user terminal (200), an expert terminal (300), or a generative artificial intelligence, and control the operation thereof, and perform all computation and control operations necessary to perform the method according to one embodiment of the present disclosure.

[0057] Additionally, the processor (110) may include a CPU (Central Processing Unit), GPU (Graphics Processing Unit), AP (Application Processor), mobile AP, DSP (Digital Signal Processor), NPU (Neural Processing Unit), MCU (Microcontroller Unit), FPGA (Field-Programmable Gate Array), etc. The configuration of the processor (110) is not limited thereto and may be configured in various ways within the scope of the problem to be solved by the present disclosure.

[0058] In one embodiment, the memory (120) may include one or more memories. The memory (120) may store various information (data) in response to requests from the processor (110), etc. The information stored in the memory (120) may include software, which is information acquired, processed, or used by at least one component of the electronic device (100). In the present disclosure, instructions or programs may be software stored in the memory (120), which may include an operating system for controlling the resources of the electronic device (100), an application, and / or middleware that provides various functions to the application so that the application can utilize the resources of the electronic device (100). Additionally, the memory (120) may store instructions that cause the processor (110) to perform calculations when executed by the processor (110). The memory (120) may store at least a portion of information received from a database through a communication interface (130) and / or information transmitted to a database through a communication interface (130). Specifically, the memory (120) can store control information, etc., of at least one of a user terminal (200), an expert terminal (300), or a generative artificial intelligence, and can store instructions for executing this information by a processor (110).

[0059] Additionally, the memory (120) may include volatile and / or non-volatile memory. The memory (120) may include, for example, DRAM (Dynamic random access memory), SRAM (Static random access memory), TTRAM (Twin transistor RAM), MRAM, TRAM (Thyristor RAM), Z-RAM (Zero capacitor RAM), EEPROM (Electrically Erasable Programmable Read-Only Memory), MRAM (Magnetic RAM), Spin-Transfer Torque MRAM (Spin-Transfer Torque MRAM), Conductive Bridging RAM (CBRAM), FeRAM (Ferroelectric RAM), PRAM (Phase change RAM), cache memory, PROM (Programmable ROM), EPROM (Erasable Programmable ROM), EEPROM (Electrically Erasable Programmable ROM), NAND flash memory, NOR flash memory, HDD (Hard Disk Drive), SSD (Solid State Drive), etc. The configuration of the memory (120) is not limited thereto and can be configured in various ways within the scope of the problem to be solved by the present disclosure.

[0060] In one embodiment, the communication interface (130) may include one or more communication interfaces. The communication interface (130) may perform wireless or wired communication between the electronic device (100) and an external device (e.g., a user terminal (200), an expert terminal (300), or an electronic device not shown). Specifically, the communication interface (130) may be used to transmit and receive control signals between the electronic device (100) and at least one of the user terminal (200) or the expert terminal (300), and to transmit and receive related information or data.

[0061] Additionally, the communication interface (130) may, for example, use a wireless network interface such as Wi-Fi (Wireless Fidelity), Bluetooth, NFC (Near Field Communication), LTE, LTE-A, 5G, etc., a wired network interface such as Ethernet, USB (Universal Serial Bus), etc., a virtual network interface such as VLAN (Virtual LAN), VPN (Virtual Private Network), etc., and a storage interface such as SATA (Serial ATA), SAS (Serial Attached SCSI), NVMe (Non-Volatile Memory Express), etc. The configuration of the communication interface (130) is not limited thereto and can be configured in various ways within the scope of the problem to be solved by the present disclosure.

[0062] Things described in the present disclosure as being performed by the electronic device (100) may be understood as being performed by at least one of one or more processors (110), one or more memories (120), or one or more communication interfaces (130) of the electronic device (100).

[0063] FIG. 3 is a flowchart illustrating a method for supporting clinical decisions using a generative artificial intelligence model according to one embodiment of the present disclosure.

[0064] Referring to FIG. 3, a method for supporting clinical decisions using a generative artificial intelligence model by an electronic device (100) according to one embodiment of the present disclosure may include the step of inputting user text input from a user terminal (200) as a prompt into a generative artificial intelligence model to which at least one of a first prompting technique or a second prompting technique is applied (S100), the step of obtaining a result of the generative artificial intelligence model for the prompt (S200), and the step of providing the result of the generative artificial intelligence model to at least one of a user terminal or an expert terminal (S300). The specific configuration of each step is described in detail below.

[0065] In step S100 of FIG. 3, the electronic device (100) is configured to input user text input from a user terminal (200) as a prompt into a generative artificial intelligence model to which at least one of a first prompting technique or a second prompting technique is applied. Specifically, the electronic device (100) may receive user input in the form of text from the user terminal (200). This user text may be text input from the user terminal (200) in the form of a diary, counseling content, or other format in which the user freely expresses their emotions, thoughts, experiences, etc. A diary is text data in which the user freely records their emotions, thoughts, experiences, etc., and is used as an Ecological Moment Assessment (EMA) and can provide useful information for understanding the user's inner state.

[0066] In one embodiment, the generative artificial intelligence model may include a large-scale language model (LLM). A large-scale language model is an artificial intelligence model capable of understanding human language, generating new text, and answering various questions by learning a vast amount of text data. In the present disclosure, the large-scale language model may perform roles such as analyzing text input from a user terminal (200), determining the presence and severity of depression based on the content, or providing necessary information to experts. The large-scale language model may contribute to accurately determining the user's mental health status through the ability to grasp the context of the text, understand the relationships between sentences, and infer hidden meanings. The large-scale language model may include various models such as, for example, ChatGPT, Gemini, LLAMA, etc. The configuration of the generative artificial intelligence model is not limited thereto and may be configured in various ways within the scope of the problem to be solved by the present disclosure.

[0067] In one embodiment, at least one of a first prompting technique or a second prompting technique may be applied to the generative artificial intelligence model described above. Here, the first prompting technique may include at least one of zero-shot prompting, few-shot prompting, the GNU Zip algorithm, or k-nearest clustering.

[0068] Specifically, zero-shot prompting is a technique that provides only user text as input without separate examples or additional information. For example, a generative AI model with zero-shot prompting applied can be configured to utilize existing knowledge without learning specific examples of depression diaries to perform a depression diagnosis task. In one embodiment, it may be required to use zero-shot prompting among the first prompting techniques.

[0069] In addition, few-shot prompting is a technique that induces a desired response format by providing a small number of examples to a generative AI model. For example, in the case of a generative AI model to which few-shot prompting is applied, if a few example responses such as "I feel good" or "I feel a little sad" are provided along with the question "How do you feel today?", the generative AI model can learn these patterns and generate answers in a similar format. In the case of a diary, by training the diary content and setting the label to "depression diagnosis," the task of diagnosing depression from the diary can be performed more effectively.

[0070] Furthermore, the GNU Zip algorithm is a text data compression algorithm that can be used to improve processing speed by reducing the size of text input to generative AI models. Specifically, the GNU Zip algorithm is a non-neural network method that compresses text data to measure similarity, and it can be used to analyze the similarities and differences in diary texts. It is also useful when processing vast amounts of text data, such as long diaries or counseling records.

[0071] Furthermore, k-nearest clustering is an algorithm that groups texts with similar characteristics and can be used to provide additional information to generative AI models by identifying existing data that exhibits data patterns similar to the user's text. For example, by finding diaries from other users that display emotional expressions similar to the user's content and providing them to the generative AI model, it enables more accurate analysis.

[0072] In one embodiment, the second prompting technique may include Chain-of-Thought (CoT) prompting. Chain-of-Thought prompting is a technique that guides a generative AI model to undergo a human-like reasoning process, helping it go beyond simply understanding text to grasp the meaning and context contained within it and reach logical conclusions. Chain-of-Thought prompting can improve complex problem-solving capabilities by presenting the generative AI model with a step-by-step method of thinking.

[0073] For example, assuming a user writes a diary entry stating, "I messed up my presentation at work today. I feel so embarrassed and have lost my confidence," the electronic device (100) can induce a generative AI model to perform the following step-by-step reasoning through chain-of-thought prompting: (1) recognize the event that the user failed the presentation at work (event identification), (2) identify that the user is feeling negative emotions through expressions such as "embarrassed" and "have lost my confidence" (emotion analysis), (3) infer that the failure of the presentation may cause a decline in self-esteem and feelings of depression (causation inference), and (4) determine that the user is currently suffering from depression and that the severity is mild (conclusion drawing). In this way, by utilizing chain-of-thought prompting, the generative AI model can go beyond simply understanding the text content to analyze the user's emotional state and situation step-by-step, and as a result, provide practical assistance to the clinical decision-making of experts.

[0074] In one embodiment, the electronic device (100) may use not only a method of inputting user text from a user terminal (200) as a prompt to a generative artificial intelligence model to which at least one of a first prompting technique or a second prompting technique is applied, but also a method of inputting user text from a user terminal (200) as a prompt to a generative artificial intelligence model and applying at least one of the first prompting technique or the second prompting technique to the input prompt as an additional prompt. The configuration of the first prompting technique and the second prompting technique is not limited to the configuration described above and may be configured in various ways within the scope of the problem to be solved by the present disclosure.

[0075] FIG. 4 is a table showing the test results of a generative artificial intelligence model according to one embodiment of the present disclosure.

[0076] Figure 4 shows the results of a test conducted on 91 participants. Specifically, depression was assessed on 91 participants using the PHQ-9 questionnaire before and after writing diaries, and suicide risk was assessed using Beck's Suicide Ideation Scale. Additionally, 428 diaries were collected from 91 participants, and the performance of the models was evaluated by varying at least one of the following: a version of a generative AI model such as ChatGPT, the application of the first prompting technique, or the second prompting technique.

[0077] In FIG. 4, models A, B, C, D, E, F, and G represent models configured by varying at least one of the following: a version of a generative artificial intelligence model such as ChatGPT, whether a first prompting technique is applied, or a second prompting technique. For example, model A represents a model in which the first prompting technique is applied to ChatGPT 3.5, model B represents a model in which the first prompting technique and the second prompting technique are applied to ChatGPT 3.5, model C represents a ChatGPT 3.5 model, model D represents a model in which the second prompting technique is applied to ChatGPT 3.5, model E represents a ChatGPT 4 model, model F represents a model in which the second prompting technique is applied to ChatGPT 4, and model G represents a model in which the GNU Zip algorithm and k-closest clustering are applied among the first prompting techniques.

[0078] In addition, as performance evaluation indicators of the model in Figure 4, Recall is the value obtained by dividing True Positives by the sum of True Positives and False Negatives, Precision is the value obtained by dividing True Positives by the sum of True Positives and False Positives, Specificity is the value obtained by dividing True Negatives by the sum of True Negatives and False Positives, Accuracy is the value of the ratio including True Positives and True Negatives, and Balanced Accuracy is the value obtained by dividing the sum of Recall and Specificity by 2.

[0079] As a result of reviewing each model based on the above performance evaluation indicators, Model B showed the best performance with an accuracy of 0.902, precision of 0.759, and specificity of 0.955, but had an average recall of 0.643, and Model E showed the best performance with a recall of 1.000, but had an average accuracy of 0.713. Model F showed balanced accuracy similar to Model E, but had a slightly lower recall compared to Model E. Any one of Models A to G described in FIG. 4 can be used as a generative artificial intelligence model according to one embodiment, and preferably, Model B, having an accuracy of 0.902 and a specificity of 0.955, can be used as a generative artificial intelligence model.

[0080] FIG. 5a is an exemplary drawing in which a generative artificial intelligence model according to one embodiment of the present disclosure is used, and FIG. 5b is an exemplary drawing in which a generative artificial intelligence model according to one embodiment of the present disclosure is used.

[0081] The chain of thought prompting as a second prompting technique is explained in more detail through Figures 5a and 5b. Referring to Figure 5a, for a generative AI model, after inserting a diary as user text, if the prompt "Please read the diary and classify it as normal or depression. Please respond only as normal or depression" is entered, "Corresponds to normal" or "Corresponds to depression" may be output as a result. Below, the case of applying chain of thought prompting to a generative AI model is explained in detail.

[0082] In one embodiment, the chain of thought prompting may include the operation of extracting a first morpheme having a conditional meaning from user text, the operation of extracting a second morpheme having a response meaning from user text, and the operation of determining whether there is depression based on at least one of the first morpheme or the second morpheme. This is a technology that extracts morphemes having a specific meaning from text containing the user's emotions, thoughts, experiences, etc., and analyzes them to identify signs of depression.

[0083] Specifically, the first morpheme may be a morpheme having the meaning of a condition, including situation, way of thinking, and attitude, and the second morpheme may be a morpheme having the meaning of a response, including emotion, opinion, and intention regarding the condition. Here, a morpheme is the smallest unit of language that has meaning, and refers to a word or word component that expresses only grammatical or relational meaning. However, in the present disclosure, a morpheme may include not only words or word components, but also phrases, sentences, and paragraphs.

[0084] Referring to FIG. 5b, in addition to the prompt of FIG. 5a, a prompt may be used that classifies normal or depressed, for example, "according to the [procedure] below," and [procedure] as "1. Extract morphemes representing conditions including situations, ways of thinking, and attitudes from the diary; 2. Extract morphemes representing reactions including emotions, opinions, and intentions from the diary; 3. Classify as normal or depressed based on the extracted information." Then, as a result, "corresponds to normal" or "corresponds to depression" may be output. By doing so, the user's psychological state can be analyzed in depth, thereby improving the accuracy of the depression diagnosis.

[0085] For example, when user text such as “I was scolded by my boss at work today. I was so angry and felt so wronged that I cried” is input into a generative artificial intelligence model as a diary, the electronic device (100) can extract “scolding” which has the meaning of a condition (e.g., situation) as a first morpheme, and extract at least one of “anger,” “feeling wronged,” or “tears” which has the meaning of a reaction (e.g., emotion) as a second morpheme, and output “corresponds to depression” based on the first morpheme and the second morpheme. Since chain of thought prompting grasps the meaning and context from keywords within the text and reaches a logical conclusion, it can increase accuracy compared to cases where chain of thought prompting is not used.

[0086] FIG. 6 is an exemplary drawing in which a generative artificial intelligence model according to one embodiment of the present disclosure is used.

[0087] In one embodiment, the thought chain prompting may further include the operation of extracting a third morpheme having the meaning of depression intensity from user text, and the operation of determining the depression intensity based on the third morpheme. This goes beyond simply determining whether or not there is depression and allows for supporting customized responses according to the intensity by determining the intensity of depression.

[0088] Specifically, when the intensity of depression is classified into mild, moderate, and severe, the third morpheme may include a morpheme having a mild meaning including fatigue and motivation, a morpheme having a moderate meaning including anxiety and agitation, and a morpheme having a severe meaning including suicide, hallucination, and delusion.

[0089] Specifically, in the case of mild depression, morphemes related to fatigue and lack of motivation may appear primarily. For example, expressions such as "tired," "lethargic," "lack of motivation," and "difficult to concentrate" may be included. At this stage, there may be some disruption to daily life, but generally, daily life can be maintained.

[0090] Furthermore, in cases of moderate depression, morphemes related to anxiety and restlessness may appear. For example, expressions such as "anxious," "restless," "worried," "cannot sleep well," and "lack of appetite" may be included. At this stage, significant disruption to daily life may occur, and difficulties may be experienced in interpersonal relationships and work performance.

[0091] Finally, in cases of severe depression, morphemes related to suicide, hallucinations, and delusions may appear. For example, expressions such as "I want to die," "I want to commit suicide," "I hear hallucinations," or "I feel like someone is watching me" may be included. At this stage, normal life is nearly impossible, and urgent intervention may be required due to the high risk of self-harm or harm to others.

[0092] Referring to FIG. 6, in addition to the prompt of FIG. 5b, a prompt may be used that includes, for example, “determine mild, moderate, or severe as the intensity of depression if classified as depression” and “respond as mild, moderate, or severe as the intensity of depression if classified as depression” and as [procedure] “4. extract morphemes indicating mild, including fatigue and motivation from the diary; 5. extract morphemes indicating moderate, including anxiety and agitation from the diary; 6. extract morphemes indicating severe, including suicide, hallucinations, and delusions from the diary; 7. classify as mild, moderate, or severe based on the extracted information if classified as depression”. Then, as a result, for example, “corresponds to depression, and the intensity of depression is moderate.” may be output.

[0093] FIG. 7 is a drawing illustrating an environment in which a generative artificial intelligence model according to one embodiment of the present disclosure is utilized.

[0094] Referring to FIG. 7, first, as described above, a user writes a diary in which they freely describe their emotions, thoughts, experiences, etc. through a user terminal (200) such as a smartphone or tablet PC, and the diary can be transmitted to an electronic device (100) as text data. The electronic device (100) can analyze the received diary data, i.e., the text data, by inputting it as a prompt into a generative artificial intelligence model (in particular, a large-scale language model).

[0095] In one embodiment, the electronic device (100) is configured to obtain the result of a generative artificial intelligence model for the aforementioned prompt and to provide the result of the generative artificial intelligence model to at least one of a user terminal (200) or an expert terminal (300). Specifically, the generative artificial intelligence model may analyze text data to generate an initial response including at least one of whether the user has depression or the intensity of depression. In addition, the initial response may further include information about the user's current state, self-care advice, or a recommendation for expert counseling. This response may be provided to at least one of the user terminal (200) or the expert terminal (300).

[0096] When the results of the generative artificial intelligence model are provided to the expert terminal (300), the expert (e.g., a psychiatrist) can verify the initial response, diagnose whether depression is present, and, if diagnosed with depression, determine the severity of the depression. Subsequently, when the expert inputs the depression diagnosis results, including whether depression is present and the severity of depression, into the expert terminal (300), the results can be provided to the user terminal (200). This depression diagnosis result is a diagnosis result based on the expert's professional knowledge and experience in addition to the analysis results of the generative artificial intelligence model, and thus can be a more objective diagnosis result that reduces the possibility of misdiagnosis. If necessary, the expert can intervene more actively and select customized responses according to the user's condition and situation, such as face-to-face counseling with the user, establishing a treatment plan, and prescribing medication.

[0097] In one embodiment, the results of the expert's diagnosis and intervention are transmitted to an electronic device (100) and can be used as training data for a generative artificial intelligence model. Through this, the generative artificial intelligence model can generate more accurate and improved responses, and the improved responses provide more objective and reliable diagnosis results to the user and the expert.

[0098] In one embodiment, if the depression intensity obtained as a result of the generative artificial intelligence model is moderate or higher, an alarm may be generated on at least one of the user terminal (200) or the expert terminal (300). As described above, the electronic device (100) may classify the user's depression intensity as mild, moderate, or severe based on the analysis results of the generative artificial intelligence model. If the depression intensity is determined to be moderate or higher, that is, if the user's condition is judged to be severe enough to interfere with daily life or to pose a risk of self-harm or harm to others, the electronic device (100) may immediately generate an alarm on at least one of the user terminal (200) or the expert terminal (300). In one embodiment, it is also possible to generate an alarm on at least one of the user terminal (200) or the expert terminal (300) only when the depression intensity obtained as a result of the generative artificial intelligence model is severe.

[0099] Specifically, the alarm provided to the user terminal (200) may include a popup of a first message, occurring in the form of at least one of sound or vibration. For example, the first message may include information such as whether there is depression, the severity of depression, the need for face-to-face consultation with a specialist (e.g., a psychiatrist), and the contact information and location of a nearby specialist or a dedicated specialist. This can induce the user to take active and prompt action.

[0100] Additionally, the alarm provided to the expert terminal (300) may be generated in the form of at least one of sound or vibration and may include a popup of a second message. For example, the second message may include information such as the user's information, whether or not they have depression, the severity of depression, and related text data (e.g., morphemes extracted from a diary). This enables the expert to objectively assess the situation and support the establishment of an accurate diagnosis and appropriate intervention strategy.

[0101] The method according to the embodiments described above may be implemented in the form of program instructions that can be executed through various computer means and recorded on a computer-readable medium. The computer-readable medium may include program instructions, data files, data structures, etc., either individually or in combination. The program instructions recorded on the medium may be those specifically designed and configured for the embodiments, or they may be those known and available to those skilled in the art of computer software technology. Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks, and magnetic tapes; optical recording media such as CD-ROMs and DVDs; magneto-optical media such as floptical disks; and hardware devices specifically configured to store and execute program instructions, such as ROM, RAM, and flash memory. Examples of program instructions include machine code, such as that generated by a compiler, as well as high-level language code that can be executed by a computer using an interpreter, etc. The hardware devices described above may be configured to operate as one or more software modules to perform the operation of the embodiments, and vice versa.

[0102] Software may include computer programs, code, instructions, or a combination of one or more of these, and may configure a processing unit to operate as desired or command the processing unit independently or collectively. Software and / or data may be permanently or temporarily embodied in any type of machine, component, physical device, virtual equipment, computer storage medium or device, or transmitted signal wave so as to be interpreted by the processing unit or to provide instructions or data to the processing unit. Software may be distributed over networked computer systems and may be stored or executed in a distributed manner. Software and data may be stored on one or more non-transitory computer-readable recording media.

[0103] Although the embodiments have been described above with reference to the limited drawings, those skilled in the art can apply various technical modifications and variations based on the above. For example, suitable results may be achieved even if the described techniques are performed in a different order than described, and / or if the components of the described system, structure, device, circuit, etc. are combined or assembled in a form different from described, or replaced or substituted by other components or equivalents.

[0104] Therefore, other implementations, other embodiments, and equivalents to the claims also fall within the scope of the claims set forth below.

Claims

1. A method for supporting clinical decisions using a generative artificial intelligence model by an electronic device, A step of inputting user text entered from a user terminal as a prompt into a generative artificial intelligence model to which at least one of a first prompting technique or a second prompting technique is applied; A step of obtaining the result of the generative artificial intelligence model for the above prompt; and A method comprising the step of providing the result of the generative artificial intelligence model to at least one of the user terminal or expert terminal.

2. In Paragraph 1, A method in which the above-mentioned generative artificial intelligence model includes a Large Language Model (LLM).

3. In Paragraph 2, The above-described first prompting technique comprises at least one of zero-shot prompting, few-shot prompting, the GNU Zip algorithm, or k-nearest clustering.

4. In Paragraph 3, The above second prompting technique is a method comprising Chain-of-Thought (CoT) prompting.

5. In Paragraph 4, The above chain of thought prompting is, The operation of extracting a first morpheme having the meaning of a condition from the above user text; The operation of extracting a second morpheme having the meaning of a response from the above user text; and A method comprising the operation of determining whether there is depression based on at least one of the first morpheme or the second morpheme.

6. In Paragraph 5, The above first morpheme is a method, which is a morpheme having the meaning of the above condition including situation, way of thinking, and attitude.

7. In Paragraph 6, A method in which the second morpheme is a morpheme having the meaning of the reaction, including emotion, opinion, and intention regarding the condition.

8. In Paragraph 7, The above chain of thought prompting is, The operation of extracting a third morpheme having the meaning of depression intensity from the above user text; and A method further comprising an action of determining the intensity of depression based on the third morpheme above.

9. In Paragraph 8, The above-mentioned depression intensity is classified into mild, moderate, and severe, and The above third morpheme is, A morpheme having the meaning of the above degree, including fatigue and motivation; The above-mentioned moderate meaning including anxiety and nervousness; and A method comprising the above-mentioned morphemes having high meaning, including suicide, hallucination, and delusion.

10. In Paragraph 9, A method for generating an alarm on at least one of the user terminal or the expert terminal when the depression intensity obtained as a result of the generative artificial intelligence model is moderate or higher.

11. In Paragraph 8, A method in which the above user text is text entered as a diary at the above user terminal.

12. In electronic devices, One or more processors; and It includes one or more memories that store at least one instruction executed by the above one or more processors, and An electronic device configured such that one or more processors perform the method of any one of claims 1 to 11 by executing at least one instruction.

13. A non-transient computer-readable recording medium comprising at least one instruction that, when executed by one or more processors, causes said one or more processors to perform the method of any one of claims 1 to 11.

Images