Medical question generation method and apparatus

Abstract

The specification provides a medical question generation method and device, the method comprising: obtaining a topic tree corresponding to a target medical topic, and obtaining a sub-topic corresponding to a target leaf node of the topic tree; inputting medical knowledge related to the sub-topic in a medical knowledge base into a first model, generating a verification type medical question based on the medical knowledge by the first model; inputting the verification type medical question and its corresponding medical knowledge into a second model, analyzing the verification type medical question and its corresponding medical knowledge by the second model, obtaining supplementary knowledge corresponding to the analysis result, and generating an explanation type medical question based on the supplementary knowledge; obtaining clinical information corresponding to the target medical topic, and inputting the clinical information, the verification type medical question and the explanation type medical question into a third model, generating a decision type medical question based on the clinical knowledge when the clinical knowledge is obtained by the third model, and generating a conjecture type medical question when the clinical knowledge is not obtained.

Description

TECHNICAL FIELD

[0001] One or more embodiments of the present specification relate to the field of artificial intelligence, and in particular, to a medical question generation method and device. BACKGROUND

[0002] In modern clinical medical practice, proposing accurate and operable medical questions, especially clinical questions, is the core prerequisite for realizing evidence-based medicine (EBM) decision-making. Clinical questions not only guide the selection of diagnosis and treatment paths and the development of individualized treatment plans, but also play a key role in multidisciplinary team collaboration, guideline updating, research topic selection, and medical quality improvement. In this case, how to obtain reasonable medical questions has become a problem to be solved. SUMMARY

[0003] One or more embodiments of the present specification provide technical solutions as follows:

[0004] According to a first aspect of one or more embodiments of the present specification, a medical question generation method is provided, the method comprising:

[0005] obtaining a topic tree corresponding to a target medical topic and obtaining a sub-topic corresponding to a target leaf node of the topic tree; wherein a root node of the topic tree represents the target medical topic, and a leaf node represents a sub-topic belonging to the target medical topic;

[0006] retrieving medical knowledge related to the sub-topic in a medical knowledge base, and inputting the medical knowledge into a first question generation model, generating a verification-type medical question based on the medical knowledge by the first question generation model;

[0007] inputting the verification-type medical question and its corresponding medical knowledge into a second question generation model, analyzing the verification-type medical question and its corresponding medical knowledge by the second question generation model, obtaining supplementary knowledge corresponding to the analysis result, and generating an explanation-type medical question based on the supplementary knowledge;

[0008] obtaining clinical information corresponding to the target medical topic, and inputting the clinical information, the verification-type medical question, and the explanation-type medical question into a third question generation model, generating a decision-type medical question based on clinical knowledge corresponding to the clinical information when the clinical knowledge is obtained by the third question generation model, and generating a speculative-type medical question when the clinical knowledge is not obtained.

[0009] According to a second aspect of one or more embodiments of the present specification, a medical question generation device is provided, the device comprising:

[0010] a subject acquisition module, configured to acquire a subject tree corresponding to a target medical subject, and acquire a sub-subject corresponding to a target leaf node of the subject tree; wherein a root node of the subject tree represents the target medical subject, and a leaf node represents a sub-subject belonging to the target medical subject;

[0011] a first question generation module, configured to search medical knowledge related to the sub-subject in a medical knowledge base, and input the medical knowledge into a first question generation model, so as to generate a verification-type medical question based on the medical knowledge by the first question generation model;

[0012] a second question generation module, configured to input the verification-type medical question and the corresponding medical knowledge into a second question generation model, so as to analyze the verification-type medical question and the corresponding medical knowledge by the second question generation model, acquire supplementary knowledge corresponding to an analysis result, and generate an explanation-type medical question based on the supplementary knowledge;

[0013] a third question generation module, configured to acquire clinical information corresponding to the target medical subject, and input the clinical information, the verification-type medical question and the explanation-type medical question into a third question generation model, so as to generate a decision-type medical question based on clinical knowledge corresponding to the clinical information by the third question generation model when the clinical knowledge is acquired, and generate a speculation-type medical question when the clinical knowledge is not acquired.

[0014] According to a third aspect of one or more embodiments of the present specification, an electronic device is provided, comprising:

[0015] a processor;

[0016] a memory for storing processor-executable instructions;

[0017] The processor implements the steps of the method according to any one of the above embodiments by running the executable instructions.

[0018] According to a fourth aspect of one or more embodiments of the present specification, a computer-readable storage medium is provided, which stores computer instructions, and the instructions are executed by a processor to implement the steps of the method according to any one of the above embodiments.

[0019] According to a fifth aspect of one or more embodiments of the present specification, a computer program product is provided, comprising computer program / instructions, and the computer program / instructions are executed by a processor to implement the steps of the method according to any one of the above embodiments.

[0020] As can be known from the above embodiments, the subject tree corresponding to the target medical subject can be acquired, and the sub-subject belonging to the target medical subject and corresponding to the target leaf node of the subject tree can be acquired; subsequently, the medical knowledge related to the sub-subject can be first searched in the medical knowledge base, and the medical knowledge is input into the first question generation model, and the first question generation model generates the verification-type medical question based on the medical knowledge; then the generated verification-type medical question and the corresponding medical knowledge can be input into the second question generation model, and the second question generation model analyzes the verification-type medical question and the corresponding medical knowledge, and acquires the supplementary knowledge corresponding to the analysis result, and generates the explanation-type medical question based on the supplementary knowledge; finally, the clinical information corresponding to the target medical subject can be acquired, and the clinical information, the generated verification-type medical question and the explanation-type medical question are input into the third question generation model, and the third question generation model generates the decision-type medical question based on the clinical knowledge corresponding to the clinical information when the clinical knowledge is acquired, and generates the speculation-type medical question when the clinical knowledge is not acquired.

[0021] In the above manner, on the one hand, the generation of medical questions can be driven based on the subject tree corresponding to a specific medical subject, so that the comprehensiveness of the generated medical questions can be improved, and the maintenance and update can be facilitated; on the other hand, different types of medical questions such as verification-type, explanation-type, decision-type and speculation-type can be automatically generated in a multi-level progressive manner, so that the generated medical questions have medical cognitive levels, and thus the quality and usability of the generated medical questions can be improved. BRIEF DESCRIPTION OF DRAWINGS

[0022] Figure 1 FIG. 1 is a schematic diagram of a medical question generation system provided by an example embodiment.

[0023] Figure 2 FIG. 1 is a flowchart of a medical question generation method provided by an example embodiment.

[0024] Figure 3 FIG. 1 is a schematic diagram of a model interaction process provided by an example embodiment.

[0025] Figure 4 FIG. 1 is a structural schematic diagram of an apparatus provided by an example embodiment.

[0026] Figure 5 FIG. 1 is a block diagram of a medical question generation apparatus provided by an example embodiment. DETAILED DESCRIPTION

[0027] In order for those skilled in the art to better understand the technical solutions in the specification, the technical solutions in the specification will be clearly and completely described below in combination with the drawings in the embodiments of the specification. Obviously, the described embodiments are only part of the embodiments of the specification, not all the embodiments. Based on the embodiments in the specification, all other embodiments obtained by those of ordinary skill in the art without creative labor should belong to the protection scope of the specification.

[0028] The user information (including but not limited to user equipment information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in the specification are information and data authorized by the user or authorized by all parties, and the collection, use and processing of related data need to comply with relevant laws, regulations and standards of relevant countries and regions, and provide corresponding operation portal for user to choose authorization or refusal.

[0029] Evidence-based medicine is a medical practice method, and its core idea is: in the process of clinical decision-making, the best research evidence currently available, the professional skills and clinical experience of doctors, and the values and preferences of patients are organically combined to develop the most scientific and individualized diagnosis and treatment plan. In simple terms, it is "use the best evidence to make the most appropriate decision".

[0030] The five-step method of practicing evidence-based medicine includes: raising clinical problems, that is, converting clinical uncertainties into answerable and structured questions (commonly in PICO or PECO format); searching for evidence, that is, efficiently finding the best evidence currently available; evaluating evidence, that is, evaluating the authenticity (Validity), importance (Importance) and applicability (Applicability) of the research (for example: Is the research design reasonable? Is the sample size sufficient? Is there any bias? Is the result clinically significant? Is it applicable to my patients?); applying evidence, that is, combining evidence with clinical experience and patient willingness to develop individualized programs; evaluating the effect, that is, evaluating the clinical effect and efficiency after applying evidence, summarizing experience and continuously improving.

[0031] As can be seen, raising precise and operable medical problems, especially clinical problems, is the core prerequisite for implementing evidence-based medical decision-making.

[0032] However, the raising and grading of medical problems have long relied on the subjective experience and artificial induction of field experts, and have inherent defects such as limited coverage, strong subjectivity and difficulty in standardization.

[0033] This specification provides one or more embodiments of a technical solution for generating medical questions. In this solution, a topic tree corresponding to a target medical topic can be obtained, and subtopics belonging to the target medical topic and corresponding to the target leaf nodes of the topic tree can be obtained. Subsequently, medical knowledge related to the subtopic can be retrieved from a medical knowledge base, and this medical knowledge is input into a first question generation model. The first question generation model generates a verification-type medical question based on this medical knowledge. Then, the generated verification-type medical question and its corresponding medical knowledge can be input into a second question generation model. The second question generation model analyzes the verification-type medical question and its corresponding medical knowledge, obtains supplementary knowledge corresponding to the analysis results, and generates an explanatory medical question based on this supplementary knowledge. Finally, clinical information corresponding to the target medical topic can be obtained, and this clinical information, the generated verification-type medical question, and the explanatory medical question are input into a third question generation model. When the third question generation model obtains clinical knowledge corresponding to the clinical information, it generates a decision-type medical question based on the clinical knowledge; otherwise, it generates a speculative medical question.

[0034] Using the above approach, on the one hand, medical questions can be generated based on a topic tree corresponding to a specific medical topic, thereby improving the comprehensiveness of the generated medical questions and making them easier to maintain and update; on the other hand, different types of medical questions, such as verification, explanation, decision-making, and speculation, can be automatically generated in a multi-level progressive manner, giving the generated medical questions a medical cognitive level, thereby improving the quality and usability of the generated medical questions.

[0035] refer to Figure 1 , Figure 1 This is a schematic diagram of a medical problem generation system provided in an exemplary embodiment.

[0036] like Figure 1 As shown, the aforementioned intelligent dialogue system may include a server and at least one client that accesses the server via any type of wired or wireless network.

[0037] The aforementioned server can correspond to a server containing a single physical host, or a server cluster consisting of multiple independent physical hosts; alternatively, it can correspond to a virtual server, cloud server, etc., hosted by a host cluster.

[0038] The client can correspond to a terminal device such as a smartphone, a tablet computer, a notebook computer, a desktop computer, a PC (Personal Computer), a PDA (Personal Digital Assistant), a wearable device (for example, smart glasses, a smart watch, etc.), a smart vehicle device, or a game console.

[0039] In actual applications, the client can upload medical information for generating a medical question to the server. The medical information can come from a clinical scenario, and the server can perform specific steps of generating a medical question.

[0040] In some embodiments, the server can be equipped with a large language model (LLM). The large language model can work with various functional components or functional subsystems on the server to generate a medical question.

[0041] A large language model is a natural language processing model based on deep learning technology, with strong language understanding and generation capabilities. A large language model is usually a deep learning model trained using a large amount of text data, which can be used to understand the meaning of natural language text or generate natural language text. A large language model can handle a variety of natural language tasks, such as text classification, named entity recognition (NER), question answering, dialogue, etc., and is an important way to artificial intelligence.

[0042] In the field of natural language processing (NLP), a large-scale text dataset is usually called a corpus. A corpus can contain various types of text data, such as literary works, academic papers, legal documents, news reports, daily conversations, emails, online forum posts, etc. By learning the text data in the corpus, a large language model can acquire and understand the rules and patterns of natural language, and then effectively process and generate human language.

[0043] A large language model usually adopts a Transformer architecture, i.e., a large language model is usually a deep learning model based on a Transformer architecture. A deep learning model based on a Transformer architecture is a type of neural network model that adopts a Transformer architecture, which performs well in natural language processing and other fields.

[0044] A Transformer is a neural network model for sequence-to-sequence modeling. The Transformer does not rely on a recurrent structure, and can parallelize training and inference, accelerating the model processing speed. In deep learning models based on the Transformer architecture, a multi-layer Transformer encoder is usually used to extract features from the input sequence, and a Transformer decoder is used to convert the extracted features into an output sequence. At the same time, such models usually also use a self-attention mechanism to capture long-distance dependencies in the input sequence, and use residual connections and normalization methods to speed up training and improve model performance.

[0045] A pre-trained model is a large language model pre-trained on large-scale unlabeled text data. The pre-trained model is a general-purpose model that is not designed and optimized for a specific task. In order to adapt the pre-trained model to specific application scenarios and task requirements, fine-tuning is needed to improve the model's performance on specific tasks. The final large language model used in practice is usually a model that is further fine-tuned based on pre-trained models, and is a supervised learning model based on labeled text data. Pre-training and fine-tuning are complementary processes, pre-training enables the model to have broad language understanding capabilities, and fine-tuning makes the model more specialized and accurate on specific tasks.

[0046] That is, the training process of a large language model can be divided into two stages: pre-training and fine-tuning. In the pre-training stage, unsupervised learning (e.g., self-supervised learning) can be used to pre-train on large-scale, unlabeled text datasets (e.g., web encyclopedias, web articles, books, etc.). Specifically, the model can predict missing parts or the next word based on context, learn statistical rules and language structures such as semantics and syntax, and iteratively update model parameters by minimizing prediction loss through backpropagation and optimization algorithms (e.g., gradient descent) to gradually improve the model's understanding of language. In the fine-tuning stage, according to the specific application scenario and task requirements, a corresponding supervised learning task (e.g., text classification, named entity recognition, question-answering systems, dialogue systems, etc.) can be selected, and a task-specific text dataset can be prepared. Then, the pre-trained model can be used as the starting point for fine-tuning, and supervised learning can be used to fine-tune on the task-specific text dataset. Specifically, the model can perform the task based on the text dataset, and iteratively update model parameters by minimizing the loss that measures the model's performance on the specific task through backpropagation and optimization algorithms (e.g., gradient descent) to gradually improve the model's performance on the specific task. In practical applications, fine-tuning can flexibly choose supervised learning, unsupervised learning, or semi-supervised learning based on the specific application scenario and the type of available data.

[0047] The language understanding ability learned by a large language model in the pre-training and fine-tuning stages enables the large language model to understand, analyze, and synthesize text information, make logical deductions, perform knowledge reasoning, or solve problems when faced with complex problems or tasks. This ability is often referred to as the reasoning ability of a large language model.

[0048] In practical applications, a pre-trained large language model is often referred to as a base model of a large language model, and a fine-tuned large language model is often referred to as a service model of a large language model.

[0049] Large language models are usually prompted or primed with a prompt, which can be an initial text or text snippet provided to the model, such as a sentence, a question, a dialogue, etc., to guide or prime the model to generate corresponding output. The prompt is a key tool to guide the model output, which can be very simple or designed quite complex, such as including instructions, examples, descriptions of expected output format, etc. Through the prompt, the large language model can be explicitly told what task it is expected to perform, such as answering a question, simulating a dialogue, writing an article, translating text, etc. At the same time, the prompt can provide the necessary background information and context for the large language model, so that the large language model can understand the logic, style, theme or position that should be followed when generating content. In addition, the prompt can also stimulate the large language model to exhibit its inherent knowledge reserve or specific language ability, such as explaining complex concepts, citing regulations, imitating the writing style of a specific author, etc.

[0050] Since large language models are mainly used for text-based processing to understand and generate human language, prompts are usually in the form of text. However, in practical applications, large language models can also accept other forms of input as prompts, such as images, audio, and even video, provided that the large language model is designed or trained to handle multi-modal data (e.g., text, images, audio, video, etc.).

[0051] In practical applications, the large language model carried on the server can be one and only one, which can be used to perform multiple types of medical question generation tasks, knowledge retrieval tasks, topic expansion tasks, medical accuracy verification tasks, answer accessibility verification tasks, standardization processing tasks, scoring rule generation tasks, etc. Alternatively, the large language model carried can be multiple, which can include a large language model for performing verification medical question generation tasks, a large language model for performing explanatory medical question generation tasks, a large language model for performing decision-making / presumptive medical question generation tasks, a large language model for performing knowledge retrieval tasks, a large language model for performing topic expansion tasks, a large language model for performing medical accuracy verification tasks, a large language model for performing answer accessibility verification tasks, a large language model for performing standardization processing tasks, a large language model for performing scoring rule generation tasks, etc.

[0052] In addition, the server can also maintain a topic tree for driving medical question generation.

[0053] It should be noted that the server can also carry various functional components or functional subsystems. These components or subsystems can work together with the large language model carried on the server to jointly implement medical question generation.

[0054] Reference Figure 2 , Figure 2 is a flowchart of a medical question generation method provided by an example embodiment.

[0055] In this embodiment, the medical question generation method described above can be applied to a server as shown in FIG. 2. The medical question generation method can specifically include the following steps: Figure 1

[0056] Step 202: Obtain a topic tree corresponding to a target medical topic, and obtain a sub-topic corresponding to a target leaf node of the topic tree; wherein the root node of the topic tree represents the target medical topic, and the leaf node represents a sub-topic belonging to the target medical topic.

[0057] In this embodiment, the generation of medical questions can be driven based on a topic tree corresponding to a specific medical topic. The specific medical topic can be a medical topic selected according to actual conditions and actual needs.

[0058] A topic tree is a structured method for organizing, presenting, and exploring hierarchical relationships between topics. It is usually presented in the form of a tree diagram, with the root node representing a main topic, and the child nodes gradually refining into more specific sub-topics related to the main topic. Topic trees are widely used in knowledge management, content planning, education design, product demand analysis, AI dialogue systems, and other fields.

[0059] In a topic tree, the root node can represent the top-level main topic, the branch node can represent the intermediate-level sub-topic, and the leaf node can represent the most specific and non-divisible fine-grained sub-topic.

[0060] In this specification, the root node of a topic tree corresponding to a certain medical topic can represent the medical topic, and the leaf node of the topic tree can represent a sub-topic belonging to the medical topic. Taking the medical topic of "type 2 diabetes management" as an example, the topic tree corresponding to this medical topic can be as follows:

[0061] Type 2 Diabetes Management

[0062] ├ Disease Staging

[0063] │ ├ Chronic Stable Phase

[0064] │ ├ Acute Complication Phase

[0065] │ │ ├ Diabetic Ketoacidosis (DKA)

[0066] │ │ └ Hyperosmolar Hyperglycemic State (HHS)

[0067] │ └ Perioperative Management ​

[0068] ├Diagnosis and treatment

[0069] │ ├Screening and risk assessment

[0070] │ │ ├Identification of high-risk groups (obesity, family history, etc.)

[0071] │ │ └Screening tools (HbA1c, OGTT)

[0072] │ ├Diagnostic criteria

[0073] │ │ ├Fasting blood glucose ≥ 7.0 mmol / L

[0074] │ │ ├Random blood glucose ≥ 11.1 mmol / L + symptoms

[0075] │ │ └HbA1c ≥ 6.5%

[0076] │ ├Treatment strategy

[0077] │ │ ├Lifestyle intervention

[0078] │ │ │ ├Medical nutrition therapy

[0079] │ │ │ ├Exercise prescription

[0080] │ │ │ └Weight management

[0081] │ │ ├Drug therapy

[0082] │ │ │ ├First-line drug: metformin

[0083] │ │ │ ├Second-line options: SGLT2 inhibitors, GLP-1 receptor agonists

[0084] │ │ │ └Insulin therapy (basal / prandial)

[0085] │ │ └Complication prevention

[0086] │ │ │ ├Cardiovascular risk management (blood pressure, blood lipids)

[0087] │ │ │ └Kidney / retinopathy screening

[0088] As an example of any medical topic (which can be referred to as a target medical topic), each leaf node in the topic tree corresponding to the target medical topic can serve as an independent medical question generation unit, that is, a medical question can be generated based on the sub-topic represented by the leaf node. In this case, the topic tree corresponding to the target medical topic can be obtained, and the sub-topic corresponding to a certain leaf node (which can be referred to as a target leaf node) of the topic tree can be obtained for subsequent medical question generation.

[0089] In some embodiments, since each leaf node in the topic tree corresponding to the above-mentioned target medical topic can serve as an independent medical question generation unit, when generating a medical question based on the topic tree for the first time, the above-mentioned target leaf node can be any leaf node of the topic tree.

[0090] However, in the subsequent process of using the generated medical question, if one or more leaf nodes in the above-mentioned topic tree change due to the modification of medical guidelines, medical standards, etc., each changed leaf node can be taken as a target leaf node, and a medical question can be re-generated based on the sub-topic represented by the leaf node, and the newly generated medical question can be used to replace the medical question originally generated based on the sub-topic represented by the leaf node.

[0091] Or, if an error is found in a certain medical question in the process of using the generated medical question, the leaf node related to the medical question (for example: the leaf node used to generate the medical question) can be taken as a target leaf node, and a medical question can be re-generated based on the sub-topic represented by the leaf node, and the newly generated medical question can be used to replace the medical question originally generated based on the sub-topic represented by the leaf node.

[0092] That is, the medical question generated based on a certain sub-topic can be re-generated and replaced on demand, without the need to completely rebuild all medical questions based on medical topics. In this way, version control, change tracking and gray release of medical questions can be achieved.

[0093] In some embodiments, the topic tree corresponding to each medical topic can be constructed in advance. In the process of constructing the topic tree, the topic expansion model can be used to quickly and accurately complete the construction of the topic tree.

[0094] Taking the above-mentioned target medical topic as an example, the target medical topic can be input into the topic expansion model, and the topic expansion model can perform multi-dimensional semantic analysis on the target medical topic to generate sub-topics belonging to multiple dimensions of the target medical topic. The multi-dimensional semantic analysis refers to not only understanding the literal meaning of the initial medical topic, but also deeply and structurally understanding and disassembling the topic from multiple preset semantic dimensions with clinical significance.

[0095] In actual applications, for a medical topic, the key clinical dimensions can include disease stages (acute stage, chronic stage), related departments (endocrinology department, nephrology department), diagnosis and treatment links (screening, diagnosis, treatment, follow-up), population characteristics (old age, pregnancy, children), etc.

[0096] It should be noted that the above subject expansion model can adopt a RAG (Retrieval-Augmented Generation) mode when performing multi-dimensional semantic analysis on the above target medical subject, and the knowledge related to the target medical subject retrieved in the medical knowledge base (including medical guidelines, medical literature, medical textbooks, etc.) can be used as auxiliary information.

[0097] In the case where a plurality of dimensional sub-topics belonging to the target medical subject are obtained, a subject tree corresponding to the target medical subject can be constructed based on the target medical subject and the plurality of dimensional sub-topics.

[0098] The above subject expansion model can be a service model of a large language model for executable subject expansion tasks obtained by fine-tuning a base model of a large language model. In this case, a Prompt for guiding the model to perform a subject expansion task can be constructed based on the target medical subject, and the Prompt can be input into the subject expansion model to guide the subject expansion model to perform multi-dimensional semantic analysis on the target medical subject to generate a plurality of dimensional sub-topics belonging to the target medical subject.

[0099] Step 204: retrieving medical knowledge related to the sub-topic in the medical knowledge base, and inputting the medical knowledge into the first question generation model to generate a verification-type medical question based on the medical knowledge by the first question generation model.

[0100] In the present embodiment, in the case where the above sub-topic is obtained, knowledge (i.e. medical knowledge) related to the sub-topic can be retrieved in the medical knowledge base, and the retrieved medical knowledge can be input into the first question generation model to generate a medical question based on the retrieved medical knowledge by the first question generation model.

[0101] It should be noted that the evidence-based question type usually includes G1-G4. Among them, G1 refers to abstract fact and standard alignment (verification type): verification around definition / standard / process / grading / threshold, etc. Abstract facts do not explain the reasons and do not involve decision-making. G2 refers to mechanism and evidence interpretation (explanation type): answering "why", comparing evidence differences, evaluating extrapolation and bias, and not giving clinical decision-making. G3 refers to direct evidence support for clinical decision-making (decision-making type): contains specific patient information, gives clear and executable action plan (do / not do / when do / how far), contains threshold / time window / contraindication and applicable boundary. G4 refers to clinical decision-making without direct evidence support (speculative type): in the case of insufficient / evidence conflict / population mismatch, give semi-quantitative / intervalized conclusions, extrapolate hypotheses and uncertainties, alternative paths and trigger conditions.

[0102] The medical question generated by the first question generation model based on the medical knowledge can be a medical question of the evidence-based question type G1, i.e., a verification type medical question. The generated verification type medical question can be used to test the memory and retrieval ability of the existing medical facts, and the answer can be directly queried in medical guidelines, medical literature, medical textbooks, etc.

[0103] The first question generation model can be a service model of a large language model for an executable verification type medical question generation task, which is obtained by fine-tuning a base model of a large language model. In this case, a Prompt for guiding the model to perform the verification type medical question generation task can be constructed based on the medical knowledge, and the Prompt is input into the first question generation model to guide the first question generation model to generate the verification type medical question based on the medical knowledge.

[0104] Step 206: inputting the verification type medical question and its corresponding medical knowledge into the second question generation model, analyzing the verification type medical question and its corresponding medical knowledge by the second question generation model, obtaining supplementary knowledge corresponding to the analysis result, and generating an explanation type medical question based on the supplementary knowledge.

[0105] In this embodiment, in the case where the verification type medical question is obtained, the medical knowledge corresponding to the verification type medical question can be determined, for example, the medical knowledge used when generating the verification type medical question.

[0106] Further, the verification type medical question and its corresponding medical knowledge can be input into the second question generation model, and the second question generation model can analyze the verification type medical question and its corresponding medical knowledge, for example, analyze the physiological mechanism or causal relationship requirement involved therein. Subsequently, the second question generation model can obtain supplementary knowledge corresponding to the analysis result, and generate a medical question based on the obtained supplementary knowledge.

[0107] The medical question generated by the second question generation model based on the supplementary knowledge can be a medical question of the evidence-based question type G2, i.e., an explanation type medical question. The generated explanation type medical question can be used to test the understanding and explanation ability of physiological mechanisms, pathological processes or treatment principles, and requires causal deduction combined with multi-source evidence.

[0108] The second question generation model can be a service model of a large language model for an executable explanatory medical question generation task, which is obtained by fine-tuning a base model of a large language model. In this case, a Prompt for guiding the model to perform the explanatory medical question generation task can be constructed based on the supplementary knowledge, and the Prompt is input into the second question generation model to guide the second question generation model to generate the explanatory medical question based on the medical knowledge.

[0109] Step 208: Obtain clinical information corresponding to the target medical topic, and input the clinical information, the verification-type medical question, and the explanatory medical question into a third question generation model. When the clinical knowledge corresponding to the clinical information is obtained, the third question generation model generates a decision-type medical question based on the clinical knowledge, and when the clinical knowledge is not obtained, the third question generation model generates a speculative-type medical question.

[0110] In this embodiment, after obtaining the verification-type medical question and the explanatory medical question, the clinical information corresponding to the target medical topic can be obtained first. The clinical information can be a real case and / or a diagnosis and treatment plan that has been strictly privacy-protected, all direct identifiers (for example, name, ID number, address, etc.) and indirect identifiers (for example, accurate birth date, rare disease combination, etc.) are removed, and key clinical information (for example, age interval, gender, diagnosis, laboratory index, medication history, etc.) is retained.

[0111] Subsequently, the clinical information, the verification-type medical question, and the explanatory medical question can be input into a third question generation model, and the third question generation model generates a medical question according to whether the clinical knowledge corresponding to the clinical information is obtained.

[0112] Specifically, when the clinical knowledge is obtained, the third question generation model generates a medical question of the evidence-based question type G3, i.e., a decision-type medical question based on the clinical knowledge, and when the clinical knowledge is not obtained, the third question generation model generates a medical question of the evidence-based question type G4, i.e., a speculative-type medical question. The generated decision-type medical question can be used to test the ability to make reasonable diagnosis and treatment decisions based on direct evidence in a specific clinical context, and usually involves individualized treatment selection. The generated speculative-type medical question can be used to test the ability to make reasonable clinical judgments based on medical consensus and logical extrapolation in the absence of direct evidence, reflecting the reasoning level under uncertainty.

[0113] The third question generation model can be a service model of a large language model for an executable decision-type / suspect-type medical question generation task, which is obtained by fine-tuning a base model of a large language model. In this case, a Prompt for guiding the model to perform a decision-type medical question generation task can be constructed based on the clinical information and the corresponding clinical knowledge, and the Prompt can be input into the third question generation model to guide the third question generation model to generate a decision-type medical question; or a Prompt for guiding the model to perform a suspect-type medical question generation task can be constructed based on the clinical information, and the Prompt can be input into the third question generation model to guide the third question generation model to generate a suspect-type medical question.

[0114] As can be seen, the verification-type medical question and the explanation-type medical question are medical questions generated on the basis of authoritative evidence (i.e., knowledge in a medical knowledge base), and the decision-type medical question and the suspect-type medical question are medical questions generated on the basis of desensitized clinical information and authoritative evidence. In this way, the comprehensiveness of the generated medical questions can be improved, and the privacy security risk brought by directly using real information can be avoided.

[0115] In actual applications, the generated medical questions can be used to guide the selection of a diagnosis and treatment path, the development of an individualized treatment plan, multi-disciplinary team collaboration, guideline updating, research topic selection, and medical quality improvement.

[0116] In some embodiments, to achieve fast and accurate knowledge retrieval, a knowledge retrieval model can be used to retrieve knowledge on demand from one or more medical knowledge bases.

[0117] Reference Figure 3 , Figure 3 FIG. 1 is a schematic diagram of a model interaction process according to an example embodiment.

[0118] Retrieving medical knowledge related to the sub-topic from the medical knowledge base can specifically include inputting the sub-topic into a knowledge retrieval model, and retrieving medical knowledge related to the sub-topic from the medical knowledge base by the knowledge retrieval model.

[0119] Specifically, to reduce the complexity of calculating the association degree between texts and improve efficiency, the calculation of the association degree between texts can be converted into the calculation of the similarity between embedding vectors corresponding to the texts. The embedding process (Embedding) is used to map high-dimensional sparse data to a low-dimensional dense vector space, for example, a word can be converted into a vector representation, so that the similarity between words can be measured in a mathematical space.

[0120] The knowledge retrieval model can pre-generate first embedding vectors corresponding to each piece of knowledge in the medical knowledge base, and upon obtaining the sub-topic, generate a second embedding vector corresponding to the sub-topic, so that the similarity between the second embedding vector corresponding to the sub-topic and the first embedding vectors corresponding to each piece of knowledge can be calculated, and based on the calculated similarity, the medical knowledge related to the sub-topic can be determined from the medical knowledge base.

[0121] In one example, first embedding vectors with a similarity greater than a preset threshold to the second embedding vector corresponding to the sub-topic can be determined first, and then the medical knowledge corresponding to the determined first embedding vectors can be obtained from the medical knowledge base as the medical knowledge related to the sub-topic.

[0122] In another example, a preset number (i.e., Top K, K is the preset number) of first embedding vectors with the greatest similarity to the second embedding vector corresponding to the sub-topic can be determined first, and then the medical knowledge corresponding to the determined first embedding vectors can be obtained from the medical knowledge base as the medical knowledge related to the sub-topic.

[0123] The knowledge retrieval model can be a service model of a large language model for an executable knowledge retrieval task obtained by fine-tuning a base model of the large language model. In this case, a Prompt for guiding the model to perform a topic expansion task can be constructed based on the sub-topic, and the Prompt can be input into the topic expansion model, and the target medical topic can be multi-dimensionally semantically analyzed by the topic expansion model to generate sub-topics belonging to multiple dimensions of the target medical topic.

[0124] After the knowledge retrieval model retrieves the medical knowledge, the medical knowledge can be passed to the first question generation model. For example, the medical knowledge output by the knowledge retrieval model can be obtained by a prompt generation component in the medical question generation system, and after a Prompt for guiding the model to perform a verification-type medical question generation task is constructed based on the medical knowledge, the Prompt can be input into the first question generation model.

[0125] In some embodiments, after the second question generation model obtains the analysis result for the verification-type medical question and the corresponding medical knowledge, the analysis result can be passed to the knowledge retrieval model. For example, the analysis result output by the second question generation model can be obtained by the prompt generation component, and after a Prompt for guiding the model to perform a knowledge retrieval task is constructed based on the analysis result, the Prompt can be input into the knowledge retrieval model.

[0126] At this time, the knowledge retrieval model can retrieve, in a medical knowledge base (which can be the same as or different from the aforementioned medical knowledge base), supplementary knowledge corresponding to the analysis result, and deliver the supplementary knowledge to the second question generation model. For example, the prompt generation component can obtain the supplementary knowledge output by the knowledge retrieval model, and after constructing a Prompt for guiding the model to perform an explanatory medical question generation task based on the supplementary knowledge, input the Prompt into the second question generation model.

[0127] In some embodiments, the third question generation model can analyze the clinical information, the verification-type medical question, and the explanatory medical question to obtain corresponding analysis results. Subsequently, the third question generation model can deliver the analysis results to the knowledge retrieval model. For example, the prompt generation component can obtain the analysis results output by the third question generation model, and after constructing a Prompt for guiding the model to perform a knowledge retrieval task based on the analysis results, input the Prompt into the knowledge retrieval model.

[0128] At this time, the knowledge retrieval model can retrieve, in a medical knowledge base (which can be the same as or different from the aforementioned medical knowledge base), clinical knowledge corresponding to the analysis result, such as a matched treatment recommendation. In the case where the knowledge retrieval model retrieves the clinical knowledge corresponding to the analysis result, the knowledge retrieval model can deliver the clinical knowledge to the third question generation model. For example, the prompt generation component can obtain the clinical knowledge output by the knowledge retrieval model, and after constructing a Prompt for guiding the model to perform a decision-type medical question generation task based on the clinical knowledge, input the Prompt into the third question generation model. Accordingly, in the case where the knowledge retrieval model does not retrieve the clinical knowledge corresponding to the analysis result, the knowledge retrieval model can send a notification message to the third question generation model that the clinical knowledge is not retrieved, so that the third question generation model can directly generate a speculative medical question based on the clinical information.

[0129] In some embodiments, in order to ensure the quality and availability of the generated medical questions, the generated medical questions can be checked.

[0130] Specifically, the generated medical questions can be first verified for medical accuracy. The medical accuracy verification can include checking whether the content of the medical question conforms to the current medical consensus, whether there are term errors, logical contradictions, or safety red line problems (such as recommended contraindicated drugs), and whether the content is consistent with the knowledge or clinical information used when generating the medical question.

[0131] If the medical accuracy verification passes, the medical question can be further subjected to answer accessibility verification. The answer accessibility verification can include simulating an answer to check whether the medical question has a solution, whether the answer to the medical question can be retrieved from an existing knowledge base, and whether the answer is clear and unambiguous.

[0132] If the answer accessibility verification passes, the medical question can be further subjected to standardization processing to obtain a standardized medical question. The standardization processing can include language optimization of the medical question, use of a standard WH question (i.e., a special interrogative sentence) format, removal of colloquial expressions, addition of option structures (A / B / C / D), and insertion of meta tags (e.g., "#G3", "#Cardiology", "#Decision") for classification management.

[0133] In some embodiments, the generated medical question can be input into a first verification model, and the medical question can be subjected to medical accuracy verification by the first verification model to improve the efficiency and correctness of the medical accuracy verification.

[0134] The first verification model can be a service model of a large language model that is executable for a medical accuracy verification task, and is obtained by fine-tuning a base model of the large language model.

[0135] In some embodiments, a medical question that passes the medical accuracy verification can be input into a second verification model, and the medical question can be subjected to answer accessibility verification by the second verification model to improve the efficiency and correctness of the answer accessibility verification.

[0136] The second verification model can be a service model of a large language model that is executable for an answer accessibility verification task, and is obtained by fine-tuning a base model of the large language model.

[0137] In some embodiments, a medical question that passes both the medical accuracy verification and the answer accessibility verification can be input into a standardization processing model, and the medical question can be subjected to standardization processing by the standardization processing model to obtain a standardized medical question, to improve the efficiency and correctness of the standardization processing.

[0138] The standardization processing model can be a service model of a large language model that is executable for a standardization processing task, and is obtained by fine-tuning a base model of the large language model.

[0139] In some embodiments, a scoring rule corresponding to each medical question can also be generated based on a scoring rule generation strategy. That is, a structured scoring rule can be generated for each medical question.

[0140] For a medical question, the scoring rules corresponding to the medical question can include three aspects of scoring points, deduction points, and general deduction rules.

[0141] The scoring points refer to key correct information that must be included in the answer (for example: "mentioning that SGLT2i has cardiorenal protective effect"). The deduction points refer to the wrong suggestions that should be deducted if they appear in the answer (for example: "recommendation of thiazolidinediones for heart failure patients"). The general deduction rules refer to safety norms applicable across questions (for example: violation of contraindications, use of non-standard terms, etc.).

[0142] By generating scoring rules for the generated medical questions, the medical questions can be used to evaluate the model effect of the medical large model. Specifically, the medical large model can answer the medical question, and according to the scoring rules corresponding to the medical question, the score of the answer generated by the medical large model can be obtained, which can reflect the model effect of the medical large model. Moreover, the generated medical questions and their corresponding scoring rules can also be used as a supervised signal for reward modeling of reinforcement learning.

[0143] In some embodiments, each medical question and its corresponding knowledge (including medical knowledge, supplementary knowledge, and clinical knowledge) can be input into a scoring rule generation model, and the scoring rule generation model can generate scoring rules corresponding to each medical question, so as to improve the efficiency and correctness of scoring rule generation.

[0144] The scoring rule generation model can be a service model of a large language model for executable scoring rule generation tasks, which is obtained by fine-tuning a base model of a large language model.

[0145] In the technical solutions provided by one or more embodiments of the present application, a subject tree corresponding to a target medical subject can be acquired, and a sub-subject corresponding to a target leaf node of the subject tree and belonging to the target medical subject can be acquired; subsequently, medical knowledge related to the sub-subject can be first searched in a medical knowledge base, and the medical knowledge can be input into a first question generation model, and the first question generation model can generate a verification-type medical question based on the medical knowledge; then, the generated verification-type medical question and the corresponding medical knowledge can be input into a second question generation model, the second question generation model can analyze the verification-type medical question and the corresponding medical knowledge, and acquire supplementary knowledge corresponding to the analysis result, and generate an explanation-type medical question based on the supplementary knowledge; finally, clinical information corresponding to the target medical subject can be acquired, and the clinical information, the generated verification-type medical question and the explanation-type medical question can be input into a third question generation model, the third question generation model can generate a decision-type medical question based on clinical knowledge corresponding to the clinical information when the clinical knowledge is acquired, and generate a speculation-type medical question when the clinical knowledge is not acquired.

[0146] In the above manner, on the one hand, the generation of medical questions can be driven based on the subject tree corresponding to a specific medical subject, so that the comprehensiveness of the generated medical questions can be improved, and the maintenance and update can be facilitated; on the other hand, verification-type, explanation-type, decision-type and speculation-type medical questions of different types can be automatically generated in a multi-level progressive manner, so that the generated medical questions have medical cognitive levels, and thus the quality and usability of the generated medical questions can be improved.

[0147] Corresponding to the foregoing embodiments of the medical question generation method, the present specification also provides embodiments of a medical question generation device.

[0148] Reference Figure 4 , Figure 4 is a structural schematic diagram of an apparatus provided in an example embodiment. As shown in Figure 4As shown, the device 400 is comprised primarily of a communication interface 402, a user interface 404, a processor 406, and a data store 408, which are interconnected and communicate with each other through a system bus, network, or other connection mechanism 410. The communication interface 402 enables the device 400 to communicate with other devices, access networks, and transport networks through analog or digital modulation. For example, the communication interface 402 can include a chipset and antenna for wireless communication with a radio access network or access point. In addition, the communication interface 402 can be a wired interface such as an Ethernet, Token Ring, or USB port, or a wireless interface such as Wifi, Bluetooth, Global Positioning System (GPS), or a wide-area wireless interface (e.g., WiMAX or LTE). Of course, the communication interface 402 can support other forms of physical layer interface and standard or proprietary communication protocol. The communication interface 402 can also include multiple physical communication interfaces, such as a Wifi interface, a Bluetooth interface, and a wide-area wireless interface.

[0149] The user interface 404 includes receiving user input and providing output to a user. Thus, the user interface 404 can include input components such as a keypad, keyboard, touch- sensitive or presence-sensitive panel, computer mouse, trackball, joystick, microphone, still camera, and video camera, and output components such as a display screen (which can be combined with a touch-sensitive panel), CRT, LCD, LED, display using DLP technology, printer, other known or future developed equivalent devices. The user interface 404 can also generate audible output through a speaker, speaker jack, audio output port, audio output device, earphones, and other known or future developed equivalent devices. In some embodiments, the user interface 404 can include software, circuitry, or other form of logic that is capable of transmitting data to and receiving data from external user input / output devices. Additionally or alternatively, the device 400 can support remote access from other devices through the communication interface 402 or another physical interface (not shown). The user interface 404 can be configured to receive user input, the location and movement of which can be indicated by an indicator or cursor described herein. The user interface 404 can also be configured as a display device for rendering or displaying a text segment.

[0150] The processor 406 can include one or more general processors and / or specialized processors.

[0151] The data store 408 can include one or more volatile and / or non-volatile storage components, and can be integrated in whole or in part with the processor 406. The data store 408 can include removable and / or non-removable components.

[0152] The processor 406 can execute program instructions stored in the data storage 408 (e.g., compiled or interpreted program logic and / or machine code) to implement various functionality described herein. The data storage 408 can include a non-transitory computer-readable medium having stored thereon program instructions that, when executed by the device 400, enable the device 400 to perform any of the methods, processes, or functions disclosed in the specification and / or drawings. Execution of the program instructions 418 by the processor 406 can cause the processor 406 to utilize the data 412.

[0153] For example, the program instructions 418 can include an operating system 422 (e.g., an operating system kernel, device drivers, and / or other modules) installed on the device 400 as well as one or more application programs 420 (e.g., a browser, a social application, or a game application). Similarly, the data 412 can include operating system data 416 and application data 414. The operating system data 416 is primarily accessible to the operating system 422, while the application data 414 is primarily accessible to the one or more application programs 420. The application data 414 can be located in a file system that is visible or hidden to a user of the device 400.

[0154] The application programs 420 can communicate with the operating system 422 through one or more application programming interfaces (APIs). These APIs facilitate the application programs 420 in reading and / or writing application data 414, communicating or receiving information via the communication interface 402, receiving or displaying information on the user interface 404, and the like.

[0155] In some terminology, the application programs 420 can be referred to simply as “apps.” Furthermore, the application programs 420 can be downloaded to the device 400 through one or more online application stores or application markets. However, the application programs can also be installed on the device 400 through other means, such as through a web browser or a physical interface (e.g., a USB port) on the device 400.

[0156] Reference Figure 5 , Figure 5 is a block diagram of a medical question generation apparatus according to an example embodiment. The medical question generation apparatus can be applied to a device as shown in Figure 4 to implement the technical solutions of the present specification.

[0157] The medical question generation apparatus described above can include:

[0158] The subject obtaining module 502 obtains a subject tree corresponding to a target medical subject, and obtains a sub-subject corresponding to a target leaf node of the subject tree; wherein a root node of the subject tree represents the target medical subject, and a leaf node represents a sub-subject belonging to the target medical subject;

[0159] The first question generation module 504 retrieves medical knowledge related to the sub-theme in a medical knowledge base, and inputs the medical knowledge into a first question generation model, and generates a verification-type medical question based on the medical knowledge by the first question generation model;

[0160] The second question generation module 506 inputs the verification-type medical question and its corresponding medical knowledge into a second question generation model, analyzes the verification-type medical question and its corresponding medical knowledge by the second question generation model, obtains supplementary knowledge corresponding to the analysis result, and generates an explanation-type medical question based on the supplementary knowledge;

[0161] The third question generation module 508 obtains clinical information corresponding to the target medical theme, and inputs the clinical information, the verification-type medical question and the explanation-type medical question into a third question generation model, generates a decision-type medical question based on clinical knowledge corresponding to the clinical information by the third question generation model when the clinical knowledge is obtained, and generates a speculation-type medical question when the clinical knowledge is not obtained.

[0162] In some embodiments, the retrieving medical knowledge related to the sub-theme in a medical knowledge base comprises:

[0163] Inputting the sub-theme into a knowledge retrieval model, and retrieving medical knowledge related to the sub-theme in a medical knowledge base by the knowledge retrieval model.

[0164] In some embodiments, the analyzing the verification-type medical question and its corresponding medical knowledge by the second question generation model, and obtaining supplementary knowledge corresponding to the analysis result, and generating an explanation-type medical question based on the supplementary knowledge comprises:

[0165] Analyzing the verification-type medical question and its corresponding medical knowledge by the second question generation model, and inputting the analysis result into the knowledge retrieval model.

[0166] Retrieving supplementary knowledge corresponding to the analysis result in a medical knowledge base by the knowledge retrieval model, and inputting the supplementary knowledge into the second question generation model.

[0167] Generating an explanation-type medical question based on the supplementary knowledge by the second question generation model.

[0168] In some embodiments, the generating a decision-type medical question based on clinical knowledge corresponding to the clinical information by the third question generation model when the clinical knowledge is obtained, and generating a speculation-type medical question when the clinical knowledge is not obtained comprises:

[0169] The third question generation model analyzes the clinical information, the verification-type medical question and the explanation-type medical question, and inputs an analysis result into the knowledge retrieval model;

[0170] The knowledge retrieval model retrieves clinical knowledge corresponding to the analysis result in a medical knowledge base, and inputs the clinical knowledge into the third question generation model when the clinical knowledge is retrieved;

[0171] The third question generation model generates a decision-type medical question based on the clinical knowledge when the clinical knowledge corresponding to the clinical information is obtained, and generates a speculative-type medical question when the clinical knowledge is not obtained.

[0172] In some embodiments, the apparatus further comprises a topic tree generation module configured to:

[0173] input the target medical topic into a topic expansion model, and perform multi-dimensional semantic analysis on the target medical topic by the topic expansion model to generate sub-topics belonging to multiple dimensions of the target medical topic;

[0174] construct a topic tree corresponding to the target medical topic based on the target medical topic and the sub-topics of the multiple dimensions.

[0175] In some embodiments, the apparatus further comprises a verification module configured to:

[0176] perform medical accuracy verification on the generated medical question;

[0177] if the medical accuracy verification is passed, perform answer accessibility verification on the medical question;

[0178] if the answer accessibility verification is passed, perform standardization processing on the medical question to obtain a standardized medical question.

[0179] In some embodiments, the medical accuracy verification on the generated medical question comprises:

[0180] input the generated medical question into a first verification model, and perform medical accuracy verification on the medical question by the first verification model.

[0181] In some embodiments, the answer accessibility verification on the medical question comprises:

[0182] input the medical question into a second verification model, and perform answer accessibility verification on the medical question by the second verification model.

[0183] In some embodiments, the standardization processing on the medical question to obtain a standardized medical question comprises:

[0184] inputting the medical question into a standardization processing model, and performing standardization processing on the medical question by the standardization processing model to obtain a standardized medical question.

[0185] In some embodiments, the apparatus further includes a scoring rule generation module configured to:

[0186] generate a scoring rule corresponding to each medical question based on a scoring rule generation strategy.

[0187] In some embodiments, the generating a scoring rule corresponding to each medical question based on a scoring rule generation strategy includes:

[0188] inputting each medical question and its corresponding knowledge into a scoring rule generation model, and generating a scoring rule corresponding to each medical question by the scoring rule generation model.

[0189] In some embodiments, the target leaf node includes a leaf node that has changed; and / or a leaf node related to a medical question that has an error.

[0190] For the convenience of description, the above apparatus is described in various modules or units respectively according to functions. Of course, the functions of each module or unit can be implemented in the same or multiple software and / or hardware when implementing one or more of the present specification, and the modules implementing the same function can be combined or integrated into another system, or some features can be ignored or not executed.

[0191] Based on the same idea as the above method, the present specification also provides an electronic device, including: a processor; a memory for storing processor executable instructions; wherein the processor implements the steps of the method according to any one of the above embodiments by running the executable instructions.

[0192] Based on the same idea as the above method, the present specification also provides a computer readable storage medium having computer instructions stored thereon, which are executed by a processor to implement the steps of the method according to any one of the above embodiments.

[0193] Based on the same idea as the above method, the present specification also provides a computer program product, including computer program / instructions, which are executed by a processor to implement the steps of the method according to any one of the above embodiments.

[0194] Those skilled in the art can understand that:

[0195] In this specification, the term "comprising" or "including" or any other variation thereof is intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can also include other elements not expressly listed or inherent to such process, method, article, or apparatus. Nothing herein is intended to be a disclaimer of any other elements not expressly claimed.

[0196] In this specification, "one", "a", and "the" do not refer to the singular, but can include the plural as well, unless expressly stated otherwise.

[0197] In this specification, the ordinal numbers first, second, etc. do not necessarily indicate the order of sequence, and many times are used for the purpose of distinguishing objects. For example, the first server and the second server usually refer to two servers. In order to distinguish the two servers, they are expressed as the first server and the second server. Of course, sometimes the two servers can also be the same server.

[0198] In this specification, unless specifically stated, "receiving and sending data" is not necessarily direct receiving and sending, and can be indirect receiving and sending. For example, A receiving data sent by B can be understood as A directly receiving data sent by B, or can be understood as A indirectly receiving data sent by B through C and other subjects. Similarly, B sending data to A can be understood as B directly sending data to A, or can be understood as B indirectly sending data to A through C and other subjects. Here, C can be one subject, or two or more subjects.

[0199] In this specification, unless specifically stated, the association relationship generated between structures can be a direct association relationship or an indirect association relationship. For example, when describing "A is connected with B", unless it is specifically stated that A is directly connected with B, it should be understood that A can be directly connected with B, or A can be indirectly connected with B; for another example, when describing "A is on B", unless it is specifically stated that A is directly above B (AB is adjacent and A is above B), it should be understood that A can be directly above B, or A can be indirectly above B (there are other elements between AB and A is above B). By analogy.

[0200] This specification uses specific terms to describe embodiments thereof. Terms such as "an embodiment," "one embodiment," and / or "some embodiments" refer to a particular feature, structure, or characteristic associated with at least one embodiment of this specification. Therefore, it should be emphasized and noted that references to "an embodiment," "one embodiment," or "an alternative embodiment" in different locations throughout this specification do not necessarily refer to the same embodiment. Furthermore, those skilled in the art can combine and integrate the different embodiments or examples described herein, as well as the features of those different embodiments or examples, without contradiction.

[0201] Although one or more embodiments of this specification provide method steps as described in the embodiments or flowcharts, it is understood that the order of steps listed in the embodiments or flowcharts is only one of many possible execution orders and does not represent the only execution order. Therefore, when the claims involve method steps, any changes or adjustments to the order of such steps, or the parallelism between steps, are also within the scope of protection of the claims.

Claims

1. A method for generating medical questions, the method comprising: Obtain a topic tree corresponding to the target medical topic, and obtain the subtopics corresponding to the target leaf nodes of the topic tree; wherein, the root node of the topic tree represents the target medical topic, and the leaf nodes represent subtopics belonging to the target medical topic; Retrieve medical knowledge related to the subtopic from the medical knowledge base, and input the medical knowledge into the first question generation model, which then generates a verification-type medical question based on the medical knowledge. The verification-type medical question and its corresponding medical knowledge are input into the second question generation model. The second question generation model analyzes the verification-type medical question and its corresponding medical knowledge, obtains supplementary knowledge corresponding to the analysis results, and generates explanatory medical questions based on the supplementary knowledge. Acquire clinical information corresponding to the target medical topic, and input the clinical information, the verification medical question, and the explanation medical question into a third question generation model. When the third question generation model acquires clinical knowledge corresponding to the clinical information, it generates a decision-making medical question based on the clinical knowledge; when it does not acquire the clinical knowledge, it generates a speculative medical question. Each medical problem and its corresponding knowledge are input into the scoring rule generation model, which then generates scoring rules corresponding to each medical problem.

2. The method according to claim 1, wherein retrieving medical knowledge related to the subtopic from the medical knowledge base includes: The subtopic is input into the knowledge retrieval model, which then retrieves medical knowledge related to the subtopic from the medical knowledge base.

3. The method according to claim 2, wherein the step of analyzing the verification-type medical question and its corresponding medical knowledge by the second question generation model, obtaining supplementary knowledge corresponding to the analysis results, and generating an explanatory medical question based on the supplementary knowledge, includes: The second question generation model analyzes the verification-type medical question and its corresponding medical knowledge, and inputs the analysis results into the knowledge retrieval model; The knowledge retrieval model retrieves supplementary knowledge corresponding to the analysis results from the medical knowledge base, and inputs the supplementary knowledge into the second question generation model. The second question generation model generates explanatory medical questions based on the supplementary knowledge.

4. The method according to claim 2, wherein when the third question generation model obtains clinical knowledge corresponding to the clinical information, it generates a decision-making medical question based on the clinical knowledge, and when it does not obtain the clinical knowledge, it generates a speculative medical question, comprising: The third question generation model analyzes the clinical information, the verification-type medical questions, and the explanatory-type medical questions, and inputs the analysis results into the knowledge retrieval model. The knowledge retrieval model retrieves clinical knowledge corresponding to the analysis results from the medical knowledge base, and when the clinical knowledge is retrieved, it is input into the third question generation model. When the third question generation model obtains clinical knowledge corresponding to the clinical information, it generates decision-making medical questions based on the clinical knowledge; when it does not obtain the clinical knowledge, it generates speculative medical questions.

5. The method according to claim 1, further comprising: The target medical topic is input into the topic expansion model, which performs multi-dimensional semantic parsing on the target medical topic to generate sub-topics belonging to the target medical topic in multiple dimensions. Based on the target medical topic and the subtopics of the multiple dimensions, a topic tree corresponding to the target medical topic is constructed.

6. The method according to claim 1, further comprising: Verify the medical accuracy of the generated medical questions; If the medical accuracy verification passes, then the accessibility of the answer to the medical question is verified. If the accessibility verification of the answer passes, the medical question is standardized to obtain a standardized medical question.

7. The method according to claim 6, wherein the step of verifying the medical accuracy of the generated medical problem includes: The generated medical question is input into the first verification model, which then verifies the medical accuracy of the medical question.

8. The method according to claim 6, wherein the step of verifying the accessibility of the answer to the medical question includes: The medical question is input into the second verification model, which then verifies the accessibility of the answer to the medical question.

9. The method according to claim 6, wherein the standardization process of the medical problem to obtain a standardized medical problem includes: The medical problem is input into a standardized processing model, which then performs standardized processing on the medical problem to obtain a standardized medical problem.

10. The method of claim 1, wherein the target leaf node includes a leaf node that has changed; and / or a leaf node associated with a medical problem in which an error exists.

11. A medical problem generation device, the device comprising: The topic acquisition module acquires a topic tree corresponding to the target medical topic and acquires subtopics corresponding to the target leaf nodes of the topic tree; wherein, the root node of the topic tree represents the target medical topic, and the leaf nodes represent subtopics belonging to the target medical topic; The first question generation module retrieves medical knowledge related to the subtopic from the medical knowledge base and inputs the medical knowledge into the first question generation model, which then generates verification-type medical questions based on the medical knowledge. The second question generation module inputs the verification-type medical question and its corresponding medical knowledge into the second question generation model. The second question generation model analyzes the verification-type medical question and its corresponding medical knowledge, obtains supplementary knowledge corresponding to the analysis results, and generates explanatory medical questions based on the supplementary knowledge. The third question generation module acquires clinical information corresponding to the target medical topic and inputs the clinical information, the verification medical question, and the explanation medical question into the third question generation model. When the third question generation model acquires clinical knowledge corresponding to the clinical information, it generates a decision-making medical question based on the clinical knowledge. When it does not acquire the clinical knowledge, it generates a speculative medical question. The scoring rule generation module inputs each medical question and its corresponding knowledge into the scoring rule generation model, which then generates scoring rules corresponding to each medical question.

12. An electronic device, comprising: processor; Memory used to store processor-executable instructions; The processor implements the steps of the method as described in any one of claims 1 to 10 by executing the executable instructions.

13. A computer-readable storage medium having stored thereon computer instructions that, when executed by a processor, implement the steps of the method as claimed in any one of claims 1 to 10.

14. A computer program product comprising a computer program / instructions that, when executed by a processor, implement the steps of the method as claimed in any one of claims 1 to 10.

Images