Medical information processing method, information processing method, equipment, storage medium and program product

Abstract

The embodiment of the invention provides a medical information processing method, an information processing method, equipment, a storage medium and a program product. In the embodiment of the specification, an automatic model selection mechanism is provided, and cooperative work of an inference model and a non-inference model is realized through a problem classification model. The problem classification model is used as a'routing center ', and the user requests are intelligently distributed to different medical question and answer models for processing according to the types of the medical problems, so that the advantages of the different medical question and answer models are fully played, and optimal configuration of system resources and improvement of overall performance are realized. Besides, intelligent routing is carried out to the corresponding medical question and answer model for question answering, the answering quality is enhanced, quick response of simple questions and deep analysis of complex questions are ensured, and optimal configuration of system resources is realized. In addition, the user can obtain the medical answer information without selecting knowledge by a professional model, so that the answering efficiency and the user satisfaction of the medical question-answering system are improved.

Description

Technical Field

[0001] This specification relates to the field of artificial intelligence technology, and in particular to a medical information processing method and information processing method, device, storage medium and program product. Background Art

[0002] With the rapid development of artificial intelligence (AI) technology, AI models are being applied in various fields. For example, in the medical field, AI models are being used in various areas such as auxiliary diagnosis, disease prediction, drug development, and health management, aiming to improve medical efficiency and optimize the quality of medical services.

[0003] One application of AI models in the medical field is a medical question-and-answer system. This system provides a question-and-answer page for users, where they enter a question. The system then uses the associated AI model to output the answer to the question.

[0004] The application of AI models in the medical field has some problems, such as long response time for outputting answer information and poor matching between the output answer information and the question information. Therefore, a new solution is urgently needed. Summary of the Invention

[0005] The embodiments of this specification provide a medical information processing method and information processing method, device, storage medium and program product, which are used to provide an automated model selection mechanism, optimize resource allocation according to the complexity of the problem, intelligently route to a model adapted to the problem for answering the problem, enhance the quality of the answer, ensure rapid response to simple problems, and in-depth analysis of complex problems.

[0006] An embodiment of this specification provides a medical information processing method, including: responding to input operations on a medical question and answer page to obtain medical question information; calling a question classification model based on the medical question information, and in the question classification model, analyzing the fitness information of the medical question information and multiple medical question and answer models based on the semantic information of the medical question information, the multiple medical question and answer models have different reasoning capabilities, and the reasoning capabilities of the medical question and answer models are positively correlated with the complexity of the medical question information; based on the fitness information, selecting a first medical question and answer model from the multiple medical question and answer models; calling the first medical question and answer model to perform question and answer processing on the medical question information to obtain medical answer information corresponding to the medical question information.

[0007] An embodiment of this specification also provides a method for processing question and answer information, including: responding to input operations on a question and answer page to obtain question information; calling a question classification model based on the question information, and in the question classification model, analyzing the fitness information between the question information and multiple question and answer models based on the semantic information of the question information, the multiple question and answer models have different reasoning capabilities, and the reasoning capabilities of the question and answer models are positively correlated with the complexity of the question information; based on the fitness information, selecting a target question and answer model from multiple question and answer models; calling the target question and answer model to perform question and answer processing on the question information to obtain answer information corresponding to the question information.

[0008] An embodiment of this specification also provides a medical information processing device, including: an acquisition module, a first calling module, a selection module and a second calling module; the acquisition module is used to respond to input operations on a medical question and answer page to obtain medical question information; the first calling module is used to call a question classification model based on the medical question information, in which the fitness information of the medical question information and multiple medical question and answer models is analyzed based on the semantic information of the medical question information. The reasoning capabilities of the multiple medical question and answer models are different, and the reasoning capabilities of the medical question and answer models are positively correlated with the complexity of the medical question information; the selection module is used to select a first medical question and answer model from multiple medical question and answer models based on the fitness information; the second calling module is used to call the first medical question and answer model to perform question and answer processing on the medical question information to obtain medical answer information corresponding to the medical question information.

[0009] An embodiment of this specification also provides an electronic device, comprising: a memory and a processor; the memory is used to store a computer program; the processor is coupled to the memory and is used to execute the computer program to implement each step in the medical information processing method and the question-and-answer information processing method provided in the embodiment of this specification.

[0010] The embodiments of this specification also provide a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, the processor implements the steps of the medical information processing method and the question-and-answer information processing method provided in the embodiments of this specification.

[0011] The embodiments of this specification also provide a computer program product, including a computer program / instruction. When the computer program / instruction is executed by a processor, it causes the processor to implement the various steps in the medical information processing method and question-and-answer information processing method provided in the embodiments of this specification.

[0012] In the embodiments of this specification, an automated model selection mechanism is provided, which realizes the collaborative work of the inference model and the non-inference model through the question classification model. With the question classification model as the "routing center", user requests are intelligently distributed to different medical question-answering models for processing according to the type of medical question, giving full play to the advantages of different medical question-answering models, and realizing the optimal allocation of system resources and the improvement of overall performance. In addition, intelligent routing to the corresponding medical question-answering model for question answering enhances the quality of answers, ensures quick response to simple questions, in-depth analysis of complex questions, and realizes the optimal allocation of system resources. In addition, users can obtain medical answer information without professional model selection knowledge, thereby improving the answer efficiency and user satisfaction of the medical question-answering system. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] The drawings described herein are used to provide a further understanding of this specification and constitute a part of this specification. The exemplary embodiments and descriptions of this specification are used to explain this specification and do not constitute an improper limitation of this specification. In the drawings: Figure 1 A flowchart of a medical information processing method provided as an exemplary embodiment of this specification.

[0014] Figure 2 A flowchart of another medical information processing method provided as an exemplary embodiment of this specification.

[0015] Figure 3 This is a diagram of the internal architecture of a question classification model provided in an exemplary embodiment of this specification.

[0016] Figure 4 A schematic diagram of a training process of a question classification model provided in an exemplary embodiment of this specification.

[0017] Figure 5 A schematic diagram of the structure of a medical information processing device provided as an exemplary embodiment of this specification.

[0018] Figure 6 This is a schematic diagram of the structure of an electronic device provided by an exemplary embodiment of this specification. DETAILED DESCRIPTION

[0019] To make the objectives, technical solutions, and advantages of this specification more clear, the following will clearly and completely describe the technical solutions of this specification in conjunction with the specific embodiments of this specification and the corresponding drawings. Obviously, the embodiments described are only part of the embodiments of this specification, not all of the embodiments. Based on the embodiments in this specification, all other embodiments obtained by ordinary technicians in this field without making any creative efforts are within the scope of protection of this specification.

[0020] It should be noted that when the embodiments of this specification involve user information, the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in the embodiments of this specification are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with the relevant laws, regulations and standards of the relevant countries and regions, and provide corresponding operation portals for users to choose to authorize or refuse. In addition, the various models involved in this specification (including but not limited to language models, AI models or large language models (LLM), etc.) are in compliance with relevant laws and standards.

[0021] With the rapid development of artificial intelligence (AI) technology, AI models are increasingly being used in the medical field. Currently, AI models are being applied in a variety of areas, including assisted diagnosis, disease prediction, drug development, and health management. These efforts aim to improve medical efficiency, optimize the quality of medical services, and ultimately enhance the patient experience.

[0022] When AI models are used in the medical field, a medical Q&A system is associated with one AI model. With the advancement of AI technology, medical Q&A systems can now be associated with multiple AI models, allowing users to manually select which AI model to use. The medical Q&A system uses the user-selected AI model to answer user-entered questions. This can lead to issues such as long response times for answer output and poor matching between the output answer and the question.

[0023] The core idea of ​​manually selecting an AI model is to give the decision-making power of AI model selection to the user, who will determine which AI model to choose to answer the question based on his or her understanding and needs of the problem.

[0024] The following is an example description of the method of manually selecting a model.

[0025] (1) Mode switching button or switch: There are multiple buttons or switches corresponding to AI models on the user interface. Before or after asking a question, the user can manually trigger the button or switch corresponding to a certain AI model, and the medical question-answering system can call the AI ​​model corresponding to the triggered button to answer the question.

[0026] (2) Keyword triggering: The medical question-answering system presets some keywords (for example, "why", "principle", "mechanism", "how to solve", etc.). When the user's question contains these keywords, the medical question-answering system can prompt the user whether a specific AI model needs to be used, or directly switch to the default AI model.

[0027] (3) Model selection menu: When a model selection drop-down menu or radio button is provided on the user question interface, the user can clearly select the AI ​​model to be used before asking the question.

[0028] The above method has some shortcomings: (1) Poor user experience and cumbersome operation: a. Increased user cognitive burden: Users need to understand the differences between AI models and determine the type of questions they ask. This creates a certain cognitive barrier for non-expert users. Users need to learn and understand the characteristics of the models to make the right choices, increasing the learning cost and user difficulty.

[0029] b. Complicated operation steps: Users need to perform additional model selection operations before or after asking questions, such as clicking buttons, toggling switches, or selecting menus. This increases the number of steps and reduces the convenience and smoothness of users' questioning. Especially in high-frequency question-and-answer scenarios, frequent manual selection can significantly reduce the user experience.

[0030] (2) Low efficiency and affected response speed: Decision-making time: Users need to spend time thinking about and determining the question type and selecting a model, which prolongs the overall time it takes to obtain an answer. This is especially true when users are unfamiliar with the model's features or are hesitant.

[0031] b. Risk of misselection of models leading to reduced efficiency: If a user misjudges the question type and selects an inappropriate AI model, for example, choosing an AI model with a large number of parameters for a simple question, the system may call a model that consumes more computing resources, extending response time and reducing overall efficiency. Conversely, choosing an AI model with a small number of parameters for a complex question may result in an unsatisfactory answer, requiring the user to re-ask the question and select a model, further reducing efficiency.

[0032] (3) The accuracy of model selection is difficult to guarantee, which affects the quality of the answer: a. Insufficient user expertise: Users are not professional model users and may not accurately judge the complexity of the problem and the capabilities of the model. This is especially true in the medical field, where users' understanding of medical knowledge varies widely, making misjudgment more likely and leading to incorrect model selection.

[0033] b. Model selection relies entirely on subjective user judgment: Model selection relies entirely on subjective judgment, lacking objective standards and automated mechanisms. This is easily influenced by factors such as user experience and cognitive bias, making it difficult to guarantee accurate model selection. Incorrect model selection directly impacts the quality of the final answer and reduces user confidence in the system's professionalism and intelligence.

[0034] In the embodiments of this specification, an automated model selection mechanism is provided, which realizes the collaborative work of the inference model and the non-inference model through the question classification model. With the question classification model as the "routing center", user requests are intelligently distributed to different medical question-answering models for processing according to the type of medical question, giving full play to the advantages of different medical question-answering models, and realizing the optimal allocation of system resources and the improvement of overall performance. In addition, intelligent routing to the corresponding medical question-answering model for question answering enhances the quality of answers, ensures quick response to simple questions, in-depth analysis of complex questions, and realizes the optimal allocation of system resources. In addition, users can obtain medical answer information without professional model selection knowledge, thereby improving the answer efficiency and user satisfaction of the medical question-answering system.

[0035] Throughout the entire process, users can enter their questions into the medical Q&A system, and the system automatically completes model selection and routing in the background. This greatly simplifies the user experience, lowers the barrier to entry, and improves the convenience and fluidity of the user experience. Users no longer need to consider complex concepts such as question type and model characteristics, which reduces their cognitive burden and decision-making time, allowing them to focus more on the question itself rather than the details of model selection. This is particularly user-friendly for non-expert users and improves the system's ease of use.

[0036] A solution provided by an embodiment of this specification is described in detail below with reference to the accompanying drawings.

[0037] Figure 1 This is a flowchart of a medical information processing method provided by an exemplary embodiment of this specification. Figure 1 As shown, the method includes: 101. Respond to input operations on the medical question and answer page to obtain medical question information.

[0038] 102. A question classification model is called based on the medical question information. In the question classification model, the compatibility information between the medical question information and multiple medical question-answering models is analyzed based on the semantic information of the medical question information. The reasoning capabilities of multiple medical question-answering models are different, and the reasoning capabilities of the medical question-answering models are positively correlated with the complexity of the medical question information.

[0039] 103. Based on the fitness information, select a first medical question-answering model from multiple medical question-answering models.

[0040] 104. Call the first medical question-answering model to perform question-answering processing on the medical question information to obtain medical answer information corresponding to the medical question information.

[0041] In this embodiment, the execution subject may be a medical question-and-answer system, a medical question-and-answer application, a medical question-and-answer applet, etc., without limitation. The following description will take the medical question-and-answer system as an example, but the present invention is not limited thereto.

[0042] The medical question system may display a medical question and answer page to the user. The user may perform input operations on the medical question and answer page. Input operations may include, but are not limited to, text input operations, selection operations, and file upload operations.

[0043] Among them, (1) Text input: Question information can be entered in a text box or text area. (2) Selection operation: a. Drop-down menu. The user opens the menu by clicking or touching. The menu includes multiple medical question information and selects a medical question information from the list. b. Radio button: Multiple medical question information is displayed on the medical Q&A page. Multiple medical question information corresponds to multiple buttons. The user selects a button from the multiple buttons, which means entering the question corresponding to this button. c. Check box: Multiple medical question information is displayed on the medical Q&A page. Each medical question information corresponds to a check box, allowing the user to select zero, one or more check boxes. The question corresponding to the check box selected by the user is the question information entered by the user. (3) File upload: The user can select a local file and upload it to the medical Q&A page. The local file includes medical question information.

[0044] In this embodiment, medical question information can be obtained in response to input operations on the medical Q&A page. Medical question information describes information related to personal health status, disease diagnosis, treatment process, medication use and side effects, and medical advice. For example, medical question information includes, but is not limited to, symptom descriptions, medical history, diagnosis results, treatment plans, medication instructions, rehabilitation and health care, and preventive health care.

[0045] In this embodiment, multiple medical question-answering models are provided. Medical question-answering models are artificial intelligence systems used to process and answer healthcare-related questions. Different medical question-answering models have varying reasoning capabilities. This capability is positively correlated with the complexity of the medical question information. For example, a medical question-answering model with stronger reasoning capabilities can handle more complex medical question information.

[0046] The following is an illustrative introduction to multiple medical question-answering models and their reasoning capabilities.

[0047] For example, the various medical question-answering models, ranked from lowest to highest in terms of reasoning power, are: rule-based models, machine learning models, deep learning models, hybrid models, and reinforcement learning models. Rule-based models answer questions based on a predefined medical knowledge base and set of rules. They are typically used to address relatively straightforward questions, such as drug interaction queries or basic symptom diagnosis. However, their reliance on pre-defined rules limits their reasoning capabilities when faced with complex or unforeseen situations. Machine learning models, such as decision trees and support vector machines, learn patterns from large amounts of labeled data to make predictions or classifications. These models are suitable for answering questions of a certain degree of complexity and can handle more complex medical questions than rule-based models. Deep learning models, such as convolutional neural networks and recurrent neural networks, excel at processing unstructured data, such as medical images and medical records, automatically extracting features and recognizing complex patterns. They excel at addressing highly complex medical problems, such as diagnosing diseases through imaging and analyzing gene sequences. For example, convolutional neural networks are used to screen for early lung cancer from X-rays. Hybrid models (e.g., combining deep learning with rule-based approaches) aim to combine the powerful feature extraction capabilities of deep learning with the logical reasoning capabilities of rule-based models to improve model accuracy and interpretability. They can handle highly complex medical information and provide a degree of decision transparency, making them suitable for applications requiring high accuracy and good interpretability. For example, deep learning can be used for preliminary screening, followed by a refined assessment using rule-based models to assist doctors in making a final diagnosis. Reinforcement learning models, which learn optimal strategies through interaction with the environment, are suitable for dynamically adjusted medical scenarios, such as personalized treatment recommendation. They are particularly well-suited for long-term management of chronic disease patients, optimizing treatment plans based on real-time feedback. For example, reinforcement learning can be used to develop personalized blood sugar control strategies for diabetic patients. Reinforcement learning models can be implemented as artificial intelligence models or large language models.

[0048] The medical question-answering model provided in the embodiments of this specification can be a model with a large parameter scale, such as a large language model, or a traditional neural network model with a relatively small parameter scale. If the model has relatively more parameters, the scale of the medical question-answering model will be relatively large, and the model reasoning ability will be relatively better. Of course, it will consume more time and resources during the reasoning and training process. If the model has relatively fewer parameters, the scale of the medical question-answering model will be relatively small. If the performance meets the requirements, the model will be more lightweight and consume relatively less time and resources during the reasoning and training process. In terms of model implementation, medical question-answering models can include but are not limited to: Generative Adversarial Networks (GAN), Convolutional Neural Networks (CNN), Recurrent Neural Networks (RNN), Deep Neural Networks (DNN), Residual Networks, Encoder-Only Architecture, Decoder-Only Architecture, Encoder-Decoder Architecture, Architecture based on self-attention mechanism, etc.

[0049] Alternatively, we will use multiple medical question-answering models, including both inference and non-inference models, as examples for illustrative introduction. Non-inference models lack multi-step thinking capabilities, but offer rapid responses and information retrieval, excelling at handling simple, straightforward questions. For example, when a user asks, "What is the value of low blood pressure?" a non-inference model can quickly retrieve and return an answer from a knowledge base. The advantages of non-inference models lie in their fast response speed and low computational resource consumption, enabling them to quickly meet users' immediate needs for simple questions. Inference models, on the other hand, possess multi-step thinking capabilities and possess stronger logical analysis and problem-solving abilities, enabling them to handle complex questions that require in-depth understanding and reasoning. For example, when a user asks, "Why can't I smoke after myopia surgery?" a reasoning model can logically reason based on medical knowledge, surgical principles, and the impact of smoking on postoperative recovery, providing a more comprehensive and explanatory answer. Inference models offer the advantage of being able to handle complex questions and provide more in-depth answers, but they typically have higher computational costs and relatively longer response times.

[0050] In this embodiment, for medical question information, automatic model selection can be achieved based on the question classification model, which avoids resource waste and unnecessary computing overhead and improves the overall answering efficiency of the system. Compared with the user's subjective judgment, the question classification model can more objectively and accurately identify the type of medical problem, thereby improving the accuracy of model selection. During the entire process, there is no need for manual intervention by the user, and the operation is more convenient and smooth, which lowers the threshold for users to use the medical question-answering system, improves the user experience, and enhances the user's trust in the professionalism and intelligence of the system. Among them, the question classification model can be flexibly integrated into the medical question-answering system as an independent module, has good scalability and compatibility, and is convenient for deployment and upgrading on the medical question-answering system.

[0051] There are no restrictions on the implementation of the question classification model. From the perspective of model scale, the question classification model can be a model with a large parameter scale, such as a large language model, or a traditional neural network model with a relatively small parameter scale. There are no restrictions on this. The internal architecture of the question classification model is also not limited. For example, the question classification model can include but is not limited to: generative adversarial networks, convolutional neural networks, recurrent neural networks, deep neural networks, residual networks, encoder architectures, decoder architectures, encoder-decoder architectures, and architectures based on self-attention mechanisms.

[0052] Specifically, a question classification model is invoked based on the medical question information. Within the question classification model, the compatibility between the medical question information and multiple medical question-answering models is analyzed based on the semantic information of the medical question information. This compatibility information is used to quantitatively or qualitatively measure the degree of semantic match between the user's medical question information and the functions, knowledge scope, and expertise of a specific medical question-answering model. A higher compatibility indicates a more suitable medical question-answering model for answering the medical question information.

[0053] Exemplarily, multiple medical question and answer models include medical question and answer model A, medical question and answer model B, and medical question and answer model C. The fitness information between the medical question information and the medical question and answer model A is 60%, the fitness information between the medical question information and the medical question and answer model A is 95%, and the fitness information between the medical question information and the medical question and answer model A is 40%.

[0054] Based on the fitness information, a first medical question-answering model is selected from the multiple medical question-answering models. For example, the medical question-answering model with the highest fitness information is selected from the multiple medical question-answering models as the first medical question-answering model. In another example, candidate medical question-answering models whose fitness information exceeds a set fitness threshold are selected from the multiple medical question-answering models, and a medical question-answering model is randomly selected from the candidate medical question-answering models as the first medical question-answering model.

[0055] In this embodiment, the first medical question-answering model is called to perform question-answering processing on the medical question information to obtain medical answer information corresponding to the medical question information.

[0056] For example, medical problem information can be structured and encapsulated as model input. Model inputs may include, but are not limited to, medical problem information, contextual information, and metadata tags. For example, an example of medical problem information might be, "I've been feeling dizzy lately. Is this anemia?"; an example of contextual information might be conversation history or user profiles, such as a history of chronic illness; and an example of metadata tags might be, for example, problem type (symptom category), involved system (neurological system, hematological system, etc.), and urgency.

[0057] The first medical question-answering model can be called based on an application programming interface (API) or a local inference engine. For example, the API call is applicable to a medical question-answering model deployed on a server, while the local inference engine is applicable to a lightweight model deployed locally.

[0058] Within the first medical question-answering model, one or more of the following steps, including semantic understanding, knowledge retrieval, reasoning generation, and fitness assessment, can be performed on the medical question information to obtain the corresponding medical answer information. Optionally, to ensure the security, compliance, and readability of the answer, the medical answer information output by the medical question-answering model can be post-processed. Post-processing may include, but is not limited to: 1) security filtering, such as replacing "definitely not cancer" with "low likelihood, further examination required"; and removing unproven treatments (such as "folk remedies for diabetes"). 2) terminology normalization, such as converting "heart palpitations" to "heart palpitations (medical term)" while retaining the spoken explanation. 3) adding a medical disclaimer, automatically appending: "This information is based on current medical knowledge. Please consult a doctor for individual circumstances." 4) multilingual adaptation, such as translating the medical answer information into the user's language, such as a dialect or minority language.

[0059] For example, an example of medical question information is: "I am 45 years old and have had severe headaches recently. Could it be a brain tumor?" An example of medical answer information is: "There are many reasons for headaches, and brain tumors are just one of them. For middle-aged people with new severe headaches, it is recommended to undergo a head imaging examination as soon as possible to rule out organic lesions. Other common causes include tension headaches, migraines, and high blood pressure. Please consult a neurology or general internal medicine doctor as soon as possible, and let the doctor assess whether further examination is needed. This answer does not replace professional diagnosis, please seek medical attention in time."

[0060] For example, in Figure 2In this paper, we take multiple medical question answering models implemented as inference models and non-inference models as an example to illustrate the process of the medical question answering system processing the medical question information input by the user. Figure 2 In the process, medical question information is obtained and question preprocessing is performed on the medical question information, such as performing word segmentation, noise filtering, and normalization. The preprocessed medical question information is input into the question classification model, and the question classification model outputs the question category of the medical question information. The question category can be determined based on the fitness information of the medical question information and the reasoning model, and the fitness information of the medical question information and the non-reasoning model. For example, the fitness information of the medical question information and the reasoning model is 60%, and the fitness information of the medical question information and the non-reasoning model is 90%, and 90%>60%, the question category can be a non-reasoning question. For another example, the fitness information of the medical question information and the reasoning model is 95%, and the fitness information of the medical question information and the non-reasoning model is 20%, and 95%>20%, the question category can be a reasoning question. If the question category is a reasoning question, the preprocessed medical question information can be input into the reasoning model for question-answering processing to obtain the medical answer information corresponding to the medical question information. If the question category is a non-inferential question, the pre-processed medical question information can be input into the non-inferential model for question-answering processing to obtain medical answer information corresponding to the medical question information.

[0061] In the embodiments of this specification, an automated model selection mechanism is provided, which realizes the collaborative work of the inference model and the non-inference model through the question classification model. With the question classification model as the "routing center", user requests are intelligently distributed to different medical question-answering models for processing according to the type of medical question, giving full play to the advantages of different medical question-answering models, and realizing the optimal allocation of system resources and the improvement of overall performance. In addition, intelligent routing to the corresponding medical question-answering model for question answering enhances the quality of answers, ensures quick response to simple questions, in-depth analysis of complex questions, and realizes the optimal allocation of system resources. In addition, users can obtain medical answer information without professional model selection knowledge, thereby improving the answer efficiency and user satisfaction of the medical question-answering system.

[0062] In an optional embodiment, the question classification model analyzes the compatibility information between the medical question information and multiple medical question-answering models based on the semantic information of the medical question information. The following example provides a method for extracting features from the medical question information, performing semantic analysis based on the extracted features to obtain semantic features, performing complexity analysis based on the semantic features along at least one of the question and answer dimensions to obtain complexity information corresponding to the medical question information, and analyzing the compatibility information between the medical question information and the multiple medical question-answering models based on the complexity information and the reasoning capabilities of the multiple medical question-answering models. Based on this approach, the question classification model can accurately determine the type of medical question information (i.e., compatibility information) and route the medical question information to the appropriate medical question-answering model for answering, thereby optimizing the allocation of system resources.

[0063] Among them, the implementation method of feature extraction of medical problem information is not limited. For example, multi-dimensional feature extraction can be performed on medical problem information to obtain structured input. The multi-dimensionality may include but is not limited to: surface language features, medical entity recognition, question type classification, emotion and urgency, and context dependency. For example, surface language features include but are not limited to: number of words, number of sentences, whether there are question marks, and punctuation usage; medical entity recognition includes but is not limited to: symptoms, diseases, drugs, parts, and populations; question type classification includes but is not limited to: etiology, diagnosis, treatment, prognosis, prevention, and drug consultation; emotion and urgency include but are not limited to: whether anxiety is expressed, and whether acute symptoms are described; context dependency includes but is not limited to: whether it depends on the previous context, such as "Can I still take the medicine you mentioned last time?"

[0064] Semantic analysis is performed based on the extracted feature information to obtain high-level semantic features. For example, semantic analysis may include, but is not limited to, semantic intent, medical concept graph mapping, semantic depth, and knowledge dependency breadth. Semantic intent can be understood as the user's actual needs. For example, the intent of a medical question may include seeking an explanation, worrying about a serious illness, or seeking comfort. For example, if the medical question is "Is my headache a brain tumor?", the semantic intent may be "rule out major illnesses." Medical concept graph mapping describes the mapping of the question to standard medical terminology, for example, mapping the word "palpitation" in the medical question to "heart palpitations." Semantic depth describes whether the medical question contains multiple layers of logic and whether a chain of reasoning is required. For example, if the question is "I have high blood pressure and now I've been diagnosed with poor kidney function. Is this related?", the semantic depth may be: establishing a causal inference from "high blood pressure to kidney damage." Knowledge dependency breadth describes whether the medical question involves multiple medical systems. For example, if the medical question involves "diabetes, vision loss, and numbness in the feet," it may involve endocrinology, ophthalmology, and neurology.

[0065] Among them, for semantic features, complexity analysis is performed from at least one dimension of the question dimension and the answer dimension to obtain complexity information corresponding to the medical problem information. The problem dimension is used to describe the difficulty of understanding and processing the medical problem information itself. For example, the problem dimension may include multiple sub-dimensions, and the sub-dimensions may include but are not limited to: the number of medical entities, multi-organ system associations, causal reasoning requirements, fuzziness, contextual dependence, etc. For example, a medical entity can be understood as a key information unit with medical meaning, which can be used to represent concepts or objects in the medical field. The entity types of medical entities may include but are not limited to: symptom class, disease class, surgical operation class, and test and examination class, etc. The following is an exemplary explanation of the sub-dimensions under the problem dimension based on Table 1.

[0066] Table 1 For example, an output example of complexity level is provided: medical problem information C1: single symptom + clear entity ("What medicine should I take for a cold?") -> low complexity; medical problem information C2: multiple symptoms + preliminary reasoning ("Can diabetes cause blurred vision?") -> medium complexity; medical problem information C3: high: cross-organ system + implicit concerns + vague description ("I have been feeling weak and losing weight recently. Is this cancer?") -> high complexity.

[0067] The Response dimension measures the breadth of knowledge and depth of reasoning required to generate compliant, secure, and complete answers. The following provides an illustrative description of the subdimensions within the Response dimension based on Table 2.

[0068] Table 2 For example, consider a medical question like "What are some antipyretic medications?" with a low response complexity. For example, a simple list of antipyretic medications would suffice. Another example of a medical question like "I'm taking medication D1. Can I also take medication D2?" would have a high response complexity. For example, drug interaction analysis and bleeding risk warnings could be performed.

[0069] In this optional embodiment, when determining the complexity information corresponding to the medical problem information, the problem dimension may be considered separately, the answer dimension may be considered separately, or both the problem dimension and the answer dimension may be considered simultaneously. Specifically, when considering the problem dimension or the answer dimension separately, the complexity obtained from the problem dimension or the answer dimension may be directly used as the complexity information corresponding to the medical problem information. When considering both the problem dimension and the answer dimension simultaneously, the complexity obtained from the problem dimension and the answer dimension may be weighted to comprehensively consider the complexity of the problem dimension and the answer dimension, and the weighted result may be used as the complexity information corresponding to the medical problem information.

[0070] Based on the complexity information corresponding to the medical question information and the reasoning ability information of multiple medical question-answering models, the compatibility information between the medical question information and the multiple medical question-answering models is analyzed. The reasoning ability information of multiple medical question-answering models can be maintained through a reasoning ability profile, which is a multidimensional description system used to describe the cognitive and logical deduction capabilities of the medical question-answering model when processing medical questions. The reasoning ability profile may include but is not limited to: reasoning depth, number of model parameters, required computing resources, and supported question types. The compatibility information between the complexity information corresponding to the medical question information and the reasoning ability profile of each medical question-answering model is calculated.

[0071] Optionally, the internal structure of the question classification model is not limited. In this embodiment, an internal structure of a question classification model is exemplarily provided. Figure 3 As shown in the figure, the question classification model includes: question input layer, text embedding layer, multi-layer conversion layer and classification layer.

[0072] The question input layer is the input to the model, receiving the user's medical question information. It converts the medical question information into a form that the model can process. For example, it performs word segmentation, noise filtering, and normalization.

[0073] The text embedding layer converts the input medical question information into a vector representation (i.e., an embedded vector), converting discrete text symbols into a continuous vector representation to facilitate subsequent neural network processing. For example, a word or character can be mapped into a fixed-dimensional vector space so that the model can understand its semantics and context.

[0074] The multi-layer transformation layer performs deep processing on the input embedding vector to extract complex semantic information and contextual relationships. Specifically, the multi-layer transformation layer performs the following steps: feature extraction of the medical question information, semantic analysis based on the extracted features to obtain semantic features; and, based on the semantic features, complexity analysis along at least one of the question and answer dimensions to obtain complexity information corresponding to the medical question information. The multi-layer transformation layer includes a multi-head attention network, a self-attention network, and a feed-forward neural network (FFN). The multi-head attention mechanism and feed-forward neural network capture long-range dependencies. The multi-head attention network allows the model to simultaneously focus on different parts of the input sequence, thereby better understanding the context. The feed-forward neural network further transforms and enhances the feature representation. The stacked structure forms a deep network capable of learning more complex patterns. The self-attention network enables the model to dynamically assign importance to different input elements, which is crucial for handling the complex context of medical questions.

[0075] Among them, the classification layer is mainly used to analyze the adaptability information of medical question information and multiple medical question answering models based on complexity information and reasoning ability information of multiple medical question answering models. The classification layer can be implemented based on the pooling layer and the fully connected layer. Figure 3 In the figure, the categories of medical question information output by the classification layer are used as an example for illustration. The categories of medical question information can reflect the adaptability information of medical question information and multiple medical question-answering models. Figure 3 In the figure, the classification layer is implemented as a binary classification, and the output results include inference categories and non-inference categories. The inference category indicates the degree of fit between the medical question information and the inference model, which is higher than the degree of fit between the medical question information and the non-inference model. Therefore, the medical question information can be answered using the inference model. In this case, the first medical question-answering model is an inference model. The non-inference category indicates the degree of fit between the medical question information and the inference model, which is lower than the degree of fit between the medical question information and the non-inference model. The medical question information can be answered using the non-inference model. In this case, the first medical question-answering model is a non-inference model.

[0076] In an optional embodiment, a training method for a problem classification model is provided. An exemplary introduction is given below. Among them, the sample medical problem information can be labeled based on a combination of manual labeling and model labeling, which can improve the accuracy of the labeling on the one hand and reduce the labor cost on the other hand. For the convenience of distinction and description, the manually labeled sample medical problem information is referred to as the first sample medical problem information, and the labeling result of the first sample medical problem information is referred to as the first labeling result. The model-labeled sample medical problem information is referred to as the second sample medical problem information, and the labeling result of the second sample medical problem information is referred to as the second labeling result. Among them, based on the first sample medical problem information and its first labeling result, an artificial intelligence-based labeling model is used to label the second sample medical problem information to be labeled.

[0077] There are no restrictions on the labeling model. From a model scale perspective, the labeling model can be a model with a large parameter scale, such as a large language model, or a traditional neural network model with a relatively small parameter scale. There are no restrictions on this. The internal architecture of the labeling model is also not limited. For example, the question classification model can include but is not limited to: generative adversarial networks, convolutional neural networks, recurrent neural networks, deep neural networks, residual networks, encoder architectures, decoder architectures, encoder-decoder architectures, and architectures based on self-attention mechanisms.

[0078] The annotation model maintains annotation rule information, which describes a set of structured, executable instructions or technical specifications that guide how to identify, classify, and label specific information units (such as entities, relationships, intent, and complexity) from sample medical problem information. During the initial annotation phase, the annotation model maintains this annotation rule information. During the annotation process, the model continuously learns and optimizes this information. In the initial phase, the annotation rule information can be empty, or some initial annotation rule information can be maintained.

[0079] Specifically, a first sample medical question information and its first annotation result, as well as a second sample medical question information to be annotated, are obtained; the first annotation result represents a medical question-answering model adapted to the first sample medical question information; the annotation model is called based on the first sample medical question information and its first annotation result, and the annotation rule information relied on by the annotation model is optimized to obtain optimized annotation rule information; the annotation model is called so that the annotation model annotates the second annotation result corresponding to the second sample medical question information according to the optimized annotation rule information; and the pre-trained model is fine-tuned based on the first sample medical question information and its first annotation result, as well as the second annotation result corresponding to the second sample medical question information, to obtain a question classification model. The pre-trained model can be trained using a large amount of text data and has strong language understanding and representation capabilities.

[0080] The labeling rule information may include information objects under multiple information dimensions. For example, these dimensions may include, but are not limited to, all or part of the following: task type, medical entity, entity relationship, user intent, question complexity, and answer quality. Labeling is performed based on these multiple dimensions. For example, the task type dimension may include, but is not limited to, symptom identification, medication identification, disease identification, examination item identification, and physiological indicator identification. For example, information objects under the medical entity dimension are used to identify medical entities in sample medical question information. Information objects under the entity relationship dimension are used to describe the semantic relationship between two medical entities, such as the symptom-disease association: "Headache -> May be caused by hypertension" or the medication-indication association: "Drug D3 -> Used to prevent myocardial infarction." Information objects under the user intent dimension are used to describe the actual purpose of the user's question. For example, user intent may include, but is not limited to, etiology consultation, diagnosis advice, treatment plan, medication consultation, and prevention advice. The question complexity and answer quality dimensions can be found in the previous descriptions and will not be further elaborated here.

[0081] For example, the sample medical question information is: "What are the symptoms of a cold?" Based on the labeling rule information, each information dimension is identified. The task type is: disease symptom list extraction; medical entity: identification of disease (cold), attribute (symptoms); entity relationship: extraction of standard symptom list from the knowledge base; reasoning requirement judgment: It belongs to static knowledge query, no personalization, no causal chain, and no reasoning is required; the labeling result of this sample medical question information is a non-inference category, and the question can be answered based on a non-inference model.

[0082] For example, the sample medical question information is: "I have diabetes and my feet have been numb recently. Is it neuropathy?" Based on the labeling rule information, each information dimension is identified. The task type is: causal relationship identification; medical entities: disease (diabetes), symptoms (numb feet), suspected complications (neuropathy); entity relationship: if there is a medical mechanism path of "chronic disease -> long-term complications", then a "may cause" relationship is established; reasoning requirement judgment: it involves a multi-hop knowledge chain (diabetes -> hyperglycemia -> peripheral nerve damage -> numb feet) -> multi-hop reasoning is possible. The labeling result of the sample medical question information is the reasoning category, and the question can be answered based on the reasoning model.

[0083] Optionally, the method for obtaining the first sample medical question information and its first annotation result, as well as the second sample medical question information to be annotated, is not limited. For example, a large amount of user question data in the medical field, i.e., sample medical question information, can be collected from various data sources to obtain an original data set. The data sources may include but are not limited to: user question records of medical question-answering platforms, medical-related user search questions on search engines, user question logs of medical consulting systems, and various questions asked by simulated users in medical scenarios, etc. Figure 4 shown.

[0084] The acquired raw data is labeled to obtain the labeling results corresponding to the sample medical question information. Multiple medical question-answering models can correspond to multiple question categories, and different medical question-answering models are used to answer medical question information of different question categories. Figure 4 In this example, the annotation results include both reasoning-based and non-reasoning-based questions. Data annotation can include manual annotation and model annotation. For detailed descriptions, please refer to the aforementioned embodiments. Data annotation is performed on the original dataset to obtain an annotated dataset. For example, regarding the annotation results of the first sample medical question information mentioned above, annotators with a medical background or professional training were hired to manually annotate the collected sample medical questions. The annotators can determine the required reasoning ability based on the semantics and complexity of the medical question information and accurately annotate it. To ensure annotation quality, mechanisms such as multi-person cross-annotation and expert review can be implemented. For example, multiple question categories can be pre-set for multiple medical question-answering models, ranked from high to low in reasoning ability. For example, taking three medical question-answering models as an example, if medical question-answering model E1 has low reasoning ability, medical question-answering model E2 has medium reasoning ability, and medical question-answering model E3 has high reasoning ability, then the question categories may include: low reasoning-based questions, medium reasoning-based questions, and high reasoning-based questions. Furthermore, taking the medical question-answering model including the reasoning model and the non-reasoning model as an example, it can be determined whether the medical question information requires reasoning ability. If reasoning ability is required, it is marked as a reasoning question; if no reasoning ability is required, it is marked as a non-reasoning question.

[0085] Optionally, based on the first sample medical problem information and its first annotation result, the annotation model is called, and the annotation rule information relied on by the annotation model is optimized to obtain the optimized annotation rule information. The implementation method is not limited. The following provides an example. The annotation model maintains a first prompt word template. The first prompt word template may include annotation rule information. The annotation rule information may be empty or may include some initial annotation rule information. The initial annotation rule information may be some basic, simple annotation rules. The initial annotation rule information may be added, deleted, modified, or removed later to obtain the optimized annotation rule information. The first sample medical problem information and its first annotation result are added to the first prompt word template corresponding to the annotation model to obtain the first prompt word. The first prompt word template includes the annotation rule information; based on the first prompt word, the annotation model is called so that the annotation model annotates the third annotation result corresponding to the first sample medical problem information according to the annotation rule information; based on the third annotation result and the first annotation result, the annotation rule information is optimized to obtain the optimized annotation rule information. Among them, by preserving data for part of the data through the annotation model, the quality of the data set and the accuracy of the annotation are guaranteed, providing a data foundation for the effective training of the problem classification model.

[0086] For example, based on the difference information between the first and third annotation results, the annotation rule information in the first prompt word template can be optimized so that the annotation model can make the third annotation result of the first sample medical problem information as close as possible to the first annotation result according to the optimized annotation rule information.

[0087] Optionally, the method of invoking the annotation model so that the annotation model annotates a second annotation result corresponding to the second sample medical problem information according to the optimized annotation rule information is not limited. The following example provides an example in which a first prompt word template including the optimized annotation rule information is used as a second prompt word template, and the second sample medical problem information is added to the second prompt word template corresponding to the annotation model to obtain a second prompt word. Based on the second prompt word, the annotation model is invoked, and the second annotation result corresponding to the second sample medical problem information is annotated in the annotation model according to the optimized annotation rule information.

[0088] Optionally, the annotation model is implemented as a second medical question-and-answer model among multiple medical question-and-answer models whose reasoning capabilities meet preset conditions. For example, to improve the accuracy of sample annotation, the second medical question-and-answer model with the strongest reasoning capabilities among multiple medical question-and-answer models can be used as the annotation model. For another example, to balance the accuracy and efficiency of sample annotation, a second medical question-and-answer model with intermediate reasoning capabilities can be selected from multiple medical question-and-answer models as the annotation model. In the case where multiple medical question-and-answer models include both reasoning models and non-reasoning models, the reasoning model can be used as the annotation model.

[0089] Furthermore, the second medical question-answering model annotates the second sample medical question information to obtain second annotated result information, which is compatible with the reasoning capabilities of the second medical question-answering model. When the annotation model is implemented as a second medical question-answering model with high reasoning capabilities (such as an inference model), the second medical question-answering model generates high-quality annotation results and sample medical question information for training downstream question classification models. This entire process forms an advanced AI architecture characterized by "strong-leading weak, self-reinforcement, and closed-loop optimization." This design offers several significant technical advantages and system-level improvements.

[0090] Optionally, the pre-trained model is fine-tuned based on the first sample medical problem information and its first annotation result, and the second annotation result corresponding to the second sample medical problem information to obtain the problem classification model. The implementation method is not limited.

[0091] Among them, supervised fine-tuning (SFT) can be used to fine-tune the pre-trained model so that the trained question classification model focuses on the medical question classification task.

[0092] Among them, the model architecture of the pre-trained model is the same as the model architecture of the question classification model. For the introduction of the internal architecture of the model, please refer to the aforementioned embodiment and will not be repeated here.

[0093] Regarding the choice of loss function, the cross-entropy loss function (Cross-Entropy Loss), suitable for classification tasks, was used as the optimization objective. This loss function measures the difference between the model's predicted class probability distribution and the actual annotation results (e.g., the first and second annotation results). The goal of model training can be to ensure that the cross-entropy loss meets certain conditions. For example, the cross-entropy loss function must remain unchanged over several iterations, or its value must be less than a set threshold, such as 0.001 or 0.015.

[0094] In terms of optimization algorithms and parameter adjustment, adaptive optimization algorithms can be used. For example, adaptive optimization algorithms may include, but are not limited to, weight decay optimization algorithms (Adaptive Moment Estimation). Hyperparameters such as the learning rate, batch size, and number of training rounds should be adjusted according to actual conditions. During training, using a labeled medical Q&A dataset, the model parameters are continuously updated iteratively, enabling the pre-trained model to learn to analyze the compatibility of medical question information with multiple medical Q&A models. For example, if multiple medical Q&A models include both inference and non-inference models, the pre-trained model can learn to distinguish between "inference" and "non-inference" questions.

[0095] exist Figure 4 In this paper, the process of model training and model evaluation is demonstrated. Figure 4 During the model evaluation process, model performance is assessed. Model performance describes the comprehensive performance of the model in terms of accuracy, stability, response speed, and generalization ability when processing input data and completing classification tasks. Accuracy indicates the degree of consistency between the model's output and the true answer. Reliability indicates the stability and consistency of the model under different inputs and is measured by the error rate of the output answer information. Response speed indicates the time it takes for the model to process a request and return a result, such as average response time or throughput. Generalization ability indicates the model's ability to adapt to unprocessed medical problem information.

[0096] In an optional embodiment, the implementation method of calling the first medical question-answering model to perform question-answering processing on the medical question information to obtain the medical answer information corresponding to the medical question information is not limited. Considering that multiple medical question-answering models include: non-inference models that do not have multi-step thinking capabilities and inference models that have multi-step thinking capabilities. The following provides an implementation method of calling the first medical question-answering model to perform question-answering processing on the medical question information to obtain the medical answer information corresponding to the medical question information. If the first medical question-answering model is implemented as an inference model, the inference model is called according to the medical question information to generate the medical answer information corresponding to the medical question information according to at least part of the problem decomposition step, knowledge call step, causal reasoning step, induction and summary step, and evaluation step. At this time, the medical answer information is more in-depth and more logical. If the first medical question-answering model is implemented as a non-inference model, the non-inference model is called according to the medical question information to obtain the medical answer information corresponding to the medical question information according to at least one of pattern matching, information retrieval or predefined rules. At this time, the medical answer information is more concise and direct.

[0097] The reasoning model may execute some or all of the problem decomposition steps, knowledge retrieval steps, causal reasoning steps, summarization steps, and evaluation steps. For example, the reasoning model may generate medical answer information corresponding to the medical question information according to the problem decomposition steps, causal reasoning steps, and evaluation steps. Alternatively, the reasoning model may generate medical answer information corresponding to the medical question information according to the problem decomposition steps, knowledge retrieval steps, summarization steps, and evaluation steps.

[0098] The problem decomposition step is used to break down complex questions into several manageable sub-questions or sub-tasks. For example, the first medical question-answering model can identify the core intent of a medical question, such as "Explain the medical reasons for not smoking after myopia surgery." It breaks down the medical question into multiple logical sub-questions: 1) What is the principle of myopia surgery? 2) What is the post-operative eye recovery process? 3) What are the effects of smoking on the body? 4) How does smoking affect the healing of eye tissue? 5) What is the causal relationship between the two?

[0099] The knowledge retrieval step extracts relevant domain knowledge from the medical knowledge base, integrates it with the context, and establishes connections between knowledge. The causal reasoning step constructs a causal or logical chain based on known facts, forming an explanatory multi-level reasoning path. The inductive summary step integrates the results of multi-step reasoning into coherent, concise, and explanatory medical answers. The evaluation step assesses the rationality and completeness of the reasoning. For example, it determines whether key factors have been omitted, checks whether the logic is self-consistent, avoids contradictions, and uses words such as "maybe" and "usually" to reflect uncertainty.

[0100] Among them, non-inferential models that do not have the ability to think in multiple steps can obtain medical answer information through pattern matching, information retrieval or predefined rules.

[0101] For example, in a pattern matching approach, a non-inferential model can maintain a structured or unstructured knowledge base containing a large amount of predefined medical question information and its corresponding answer information. When a user enters a medical question information, the first medical question-answering model extracts segmentation words and keywords from the medical question information. Based on the extracted segmentation words and keywords, the model matches the predefined medical answer information corresponding to the medical question information from the knowledge base and uses the medical answer information of the predefined medical answer information as the medical answer information corresponding to the medical question information.

[0102] For another example, regarding information retrieval, a non-inferential model can maintain a structured or unstructured knowledge base containing various predefined factual medical knowledge. For a user's input medical question, the knowledge base can be searched for medical knowledge related to the "medical question information" as a medical answer. For example, if the medical question is about low blood pressure, the knowledge base can retrieve the following relevant knowledge: "systolic blood pressure below 90 mmHg (millimeters of mercury) and / or diastolic blood pressure below 60 mmHg in adults." This information can then be returned to the user as the medical answer.

[0103] For another example, answer generation using predefined rules can be template-based. For example, relevant medical knowledge can be retrieved from a knowledge base based on the medical question information, and a user-provided answer template can be obtained. Medical answer information can then be generated based on the retrieved medical knowledge and the answer template. For example, the answer template could be: "According to medical standards, hypotension typically refers to systolic blood pressure below __ mmHg and / or diastolic blood pressure below __ mmHg." The relevant medical knowledge retrieved is "systolic blood pressure below 90 mmHg and / or diastolic blood pressure below 60 mmHg in adults." By filling this medical knowledge into the answer template, the resulting medical question information could be: "According to medical standards, hypotension typically refers to systolic blood pressure below 60 mmHg and / or diastolic blood pressure below 90 mmHg."

[0104] The above embodiments of this specification use the application of AI models in the medical field as an example to illustrate the process of answering questions on a medical Q&A page. The method of answering question information in this specification is not limited to the medical field and can also be applied in other fields, such as e-commerce, tutoring, legal, and daily life. Based on this, in an optional embodiment, a method for processing question and answer information is also provided, which includes: S101. Respond to input operations on the question-and-answer page and obtain question information.

[0105] S102. Call a question classification model based on the question information. In the question classification model, analyze the compatibility information between the question information and multiple question-answering models based on the semantic information of the question information. The reasoning capabilities of multiple question-answering models are different, and the reasoning capabilities of the question-answering models are positively correlated with the complexity of the question information.

[0106] S103. Based on the fitness information, select a target question answering model from multiple question answering models.

[0107] S104: Call the target question-answering model to perform question-answering processing on the question information to obtain answer information corresponding to the question information.

[0108] In this embodiment, the question information varies depending on the application scenario. For example, in an e-commerce scenario, the question information may include but is not limited to: "When will my order be shipped?", "How do I reset my password?", "What should I do if the app crashes?", and "What is the return and exchange process?" In a learning tutoring scenario, the question information may include but is not limited to: "What is Newton's first law?", "The root formula of a quadratic equation?", "What are the rules of the English past tense?", and "What is the chemical formula of water?" In a legal scenario, the question information may include but is not limited to: "What is the maximum probation period for a labor contract?", "How is the responsibility for a traffic accident divided?", "What procedures are required for divorce?", and "Will a juvenile offender have a criminal record?" In a daily life scenario, the question information may include but is not limited to: "Can a microwave oven heat metal?", "Will bananas turn black in the refrigerator?", "How much water should I drink a day?", "How to remove oil stains from clothes?", etc.

[0109] In this case, a question classification model is trained using sample question information and sample answer information in an application scenario to obtain a question classification model for that application scenario. Accordingly, the application scenario corresponds to multiple question-answering models for answering the question information in that application scenario, and different question-answering models correspond to different reasoning capabilities.

[0110] Alternatively, a question classification model can be trained using sample question information and scenario category information from multiple different scenarios to obtain a question classification model suitable for various application scenarios. The question classification model is used to classify question information into different scenario labels, and then question answering models for different application scenarios are used to answer questions. Accordingly, multiple question answering models can process question information from different application scenarios. Multiple question answering models can be question answering models for different application scenarios, and different question answering models are used to process question information from different application scenarios.

[0111] Among them, the detailed introduction of the question classification model and multiple question-answering models can be found in the aforementioned embodiments, which will not be repeated here.

[0112] In the embodiments of this specification, an automated model selection mechanism is provided. This mechanism, through a question classification model, optimizes resource allocation based on the complexity of the question information. It automatically determines the compatibility of question information with the question-answering model and intelligently routes the question to the appropriate question-answering model for answering. This enhances the quality of answers, ensuring rapid responses to simple questions and in-depth analysis of complex ones. Furthermore, users can obtain answers without specialized model selection knowledge, improving the efficiency of the question-answering system.

[0113] Regarding the embodiments of this specification Figure 1The detailed implementation and beneficial effects of each step in the method have been described in detail in the aforementioned embodiments and will not be elaborated on here.

[0114] It should be noted that the execution entity of each step of the method provided in the above embodiment can be the same device, or the method can be executed by different devices. For example, the execution entity of steps 101 to 103 can be device A; for another example, the execution entity of steps 101 and 102 can be device A, and the execution entity of step 103 can be device B; and so on.

[0115] In addition, some of the processes described in the above embodiments and the accompanying drawings include multiple operations that appear in a specific order, but it should be clearly understood that these operations may not be executed in the order in which they appear in this article or may be executed in parallel. The serial numbers of the operations, such as 101, 102, etc., are only used to distinguish between different operations, and the serial numbers themselves do not represent any order of execution. In addition, these processes may include more or fewer operations, and these operations may be executed in sequence or in parallel. It should be noted that the descriptions of "first", "second", etc. in this article are used to distinguish different messages, devices, modules, etc., and do not represent a sequential order, nor do they limit "first" and "second" to different types.

[0116] Figure 5 This is a schematic diagram of a medical information processing device provided by an exemplary embodiment of this specification. Figure 5 As shown, the device includes: an acquisition module 51, a first calling module 52, a selection module 53 and a second calling module 54.

[0117] An acquisition module 51 is used to respond to input operations on the medical question and answer page and acquire medical question information; A first calling module 52 is configured to call a question classification model based on the medical question information. In the question classification model, based on the semantic information of the medical question information, the compatibility information between the medical question information and multiple medical question-answering models is analyzed. The multiple medical question-answering models have different reasoning capabilities, and the reasoning capabilities of the medical question-answering models are positively correlated with the complexity of the medical question information. A selection module 53, configured to select a first medical question-answering model from a plurality of medical question-answering models based on the fitness information; The second calling module 54 is used to call the first medical question-answering model to perform question-answering processing on the medical question information to obtain medical answer information corresponding to the medical question information.

[0118] Regarding the embodiments of this specification Figure 5 The detailed implementation and beneficial effects of each step in the illustrated device have been described in detail in the aforementioned embodiments and will not be elaborated on here.

[0119] Figure 6 This is a schematic diagram of the structure of an electronic device provided by an exemplary embodiment of this specification, such as Figure 6 As shown, the device includes a memory 64 and a processor 65 .

[0120] The memory 64 is used to store computer programs and can be configured to store various other data to support operations on the electronic device. Examples of such data include instructions for any application or method used to operate on the electronic device.

[0121] The processor 65 is coupled to the memory 64 and is used to execute the computer program in the memory 64 for: executing the medical information processing method and problem information processing method recorded in the aforementioned embodiments. Please refer to the records of the aforementioned embodiments for details and will not be described further here.

[0122] Regarding the embodiments of this specification Figure 6 The detailed implementation and beneficial effects of each step in the illustrated device have been described in detail in the aforementioned embodiments and will not be elaborated on here.

[0123] Further, if Figure 6 As shown, the electronic device also includes: a communication component 66, a display 67, a power component 68, an audio component 69 and other components. Figure 6 Only some components are shown schematically, which does not mean that the electronic device only includes Figure 6 In addition, Figure 6 The components in the dotted box are optional components, not mandatory components, and the specific components may depend on the product form of the electronic device. The electronic device of this embodiment can be implemented as a terminal device such as a desktop computer, a laptop computer, a smart phone or an IOT (Internet of Things) device and a smart wearable device (such as a smart watch, a smart bracelet), or a conventional server, a cloud server or a server array. If the electronic device of this embodiment is implemented as a terminal device such as a desktop computer, a laptop computer, a smart phone, etc., it can include Figure 6 If the electronic device of this embodiment is implemented as a conventional server, cloud server or server array and other server-side devices, it may not include Figure 6 Components within the dotted box.

[0124] The above-mentioned memory can be implemented by any type of volatile or non-volatile memory device or a combination thereof, such as static random-access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0125] The above-mentioned communication component is configured to facilitate wired or wireless communication between the device where the communication component is located and other devices. The device where the communication component is located can access a wireless network based on a communication standard, such as WiFi, 2G, 3G, 4G / LTE, 5G and other mobile communication networks, or a combination thereof. In an exemplary embodiment, the communication component receives a broadcast signal or broadcast-related information from an external broadcast management system via a broadcast channel. In an exemplary embodiment, the communication component also includes a near field communication (NFC) module to facilitate short-range communication. For example, the NFC module can be implemented based on Radio Frequency Identification (RFID) technology, Infrared Data Association (IrDA) technology, Ultra Wide Band (UWB) technology, Bluetooth (BT) technology and other technologies.

[0126] The display includes a screen, which may include a liquid crystal display (LCD) and a touch panel (TP). If the screen includes a touch panel, it may be implemented as a touch screen to receive input signals from the user. The touch panel includes one or more touch sensors to sense touches, slides, and gestures on the touch panel. The touch sensors can detect not only the boundaries of a touch or slide action, but also the duration and pressure associated with the touch or slide operation.

[0127] The power supply assembly provides power to various components of the device in which the power supply assembly is located. The power supply assembly may include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power to the device in which the power supply assembly is located.

[0128] The above-mentioned audio component can be configured to output and / or input audio signals. For example, the audio component includes a microphone (MIC). When the device where the audio component is located is in an operating mode, such as call mode, recording mode, and voice recognition mode, the microphone is configured to receive external audio signals. The received audio signal can be further stored in the memory or sent via the communication component. In some embodiments, the audio component also includes a speaker for outputting audio signals.

[0129] Accordingly, an embodiment of the present specification also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, enables the processor to implement each step in the above method embodiment. The computer-readable storage medium includes volatile or non-volatile or a combination thereof, and may be removable or non-removable. Examples of computer-readable storage media include, but are not limited to, phase-change random access memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), flash memory or other memory technology, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassettes, tape disk storage or other magnetic storage devices or any other non-transmission medium. Accordingly, the embodiments of this specification also provide a computer program product, which includes a computer program or instructions. When the computer program or instructions are executed by a processor, the processor is enabled to implement the steps in the above-mentioned method embodiments. It should be understood that each process or a combination of multiple processes in the above-mentioned method flow can be implemented by a computer program or instruction. In addition, these computer programs or instructions can be applied to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device, so that the processor of the general-purpose computer, the special-purpose computer, the embedded processor, or other programmable data processing device can be implemented as a device for implementing the corresponding functions in the above-mentioned method embodiments.

[0130] Those skilled in the art will appreciate that the embodiments of this specification may be provided as methods, systems, or computer program products. Thus, this specification may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, this specification may take the form of a computer program product implemented on one or more computer-readable storage media (including but not limited to magnetic disk storage, compact disc read-only memory (CD-ROM), optical storage, etc.) containing computer-usable program code.

[0131] This specification is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of this specification. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0132] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0133] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0134] In a typical configuration, a computing device includes one or more processors (Central Processing Unit, CPU), input / output interfaces, network interfaces, and memory.

[0135] Memory may include non-permanent storage in a computer-readable medium, random access memory (RAM), and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. Memory is an example of a computer-readable medium.

[0136] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can be implemented using any method or technology for information storage. Information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase-change random access memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassettes, magnetic disk storage or other magnetic storage devices, or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer-readable media does not include transitory computer-readable media, such as modulated data signals and carrier waves.

[0137] It should also be noted that the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, commodity, or apparatus that includes a series of elements includes not only those elements but also other elements not explicitly listed, or includes elements inherent to such process, method, commodity, or apparatus. In the absence of further limitations, an element defined by the phrase "comprises a ..." does not exclude the presence of other identical elements in the process, method, commodity, or apparatus that includes the element.

[0138] The above are merely examples of the present invention and are not intended to limit the present invention. Various modifications and variations are possible for those skilled in the art. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention are intended to be included within the scope of the claims of the present invention.

Claims

1. A medical information processing method, characterized in that: include: Respond to input operations on the medical question and answer page to obtain medical question information; A question classification model is called based on the medical question information. In the question classification model, based on the semantic information of the medical question information, the compatibility information between the medical question information and multiple medical question-answering models is analyzed. The multiple medical question-answering models have different reasoning capabilities, and the reasoning capabilities of the medical question-answering models are positively correlated with the complexity of the medical question information. selecting a first medical question-answering model from the plurality of medical question-answering models based on the fitness information; The first medical question-answering model is called to perform question-answering processing on the medical question information to obtain medical answer information corresponding to the medical question information.

2. The method according to claim 1, characterized in that Based on the semantic information of the medical question information, analyzing the compatibility information between the medical question information and multiple medical question-answering models, including: Extracting features from the medical problem information, and performing semantic analysis based on the extracted feature information to obtain semantic features; Based on the semantic features, performing complexity analysis from at least one of a question dimension and a response dimension to obtain complexity information corresponding to the medical question information; Based on the complexity information and the reasoning ability information of the multiple medical question-answering models, the compatibility information between the medical question information and the multiple medical question-answering models is analyzed.

3. The method according to claim 1, characterized in that Also includes: Obtaining first sample medical problem information and a first annotation result thereof, as well as second sample medical problem information to be annotated; Calling the annotation model based on the first sample medical problem information and the first annotation result thereof, and optimizing the annotation rule information on which the annotation model relies to obtain optimized annotation rule information; calling the annotation model so that the annotation model annotates a second annotation result corresponding to the second sample medical problem information according to the optimized annotation rule information; Based on the first sample medical problem information and its first annotation result, and the second annotation result corresponding to the second sample medical problem information, the pre-trained model is fine-tuned to obtain a problem classification model.

4. The method according to claim 3, characterized in that The annotation model is called based on the first sample medical problem information and the first annotation result thereof, and the annotation rule information on which the annotation model relies is optimized to obtain optimized annotation rule information, including: Adding the first sample medical question information and the first annotation result thereof to a first prompt word template corresponding to the annotation model to obtain a first prompt word, wherein the first prompt word template includes annotation rule information; calling the annotation model based on the first prompt word, so that the annotation model annotates a third annotation result corresponding to the first sample medical problem information according to the annotation rule information; Based on the third labeling result and the first labeling result, the labeling rule information is optimized to obtain optimized labeling rule information.

5. The method according to claim 3, characterized in that Calling the annotation model so that the annotation model annotates a second annotation result corresponding to the second sample medical problem information according to the optimized annotation rule information includes: Adding the second sample medical question information to a second prompt word template corresponding to the labeling model to obtain a second prompt word, wherein the second prompt word template includes the optimized labeling rule information; The annotation model is called based on the second prompt word to annotate a second annotation result corresponding to the second sample medical problem information according to the optimized annotation rule information.

6. The method according to claim 3, characterized in that The annotation model is implemented as a second medical question-answering model whose reasoning ability meets preset conditions among the multiple medical question-answering models, and the second annotation result information is adapted to the reasoning ability of the second medical question-answering model.

7. The method according to any one of claims 1 to 6, characterized in that The multiple medical question-answering models include: a non-inference model without multi-step thinking capability and an inference model with multi-step thinking capability; calling the first medical question-answering model to perform question-answering processing on the medical question information to obtain medical answer information corresponding to the medical question information, including: If the first medical question-answering model is implemented as a reasoning model, calling the reasoning model according to the medical question information to generate medical answer information corresponding to the medical question information according to at least part of the question decomposition step, the knowledge retrieval step, the causal reasoning step, the induction and summary step, and the evaluation step; If the first medical question-and-answer model is implemented as a non-inference model, the non-inference model is called according to the medical question information to obtain the medical answer information corresponding to the medical question information in accordance with at least one of pattern matching, information retrieval or predefined rules.

8. A question-answer information processing method, characterized in that: include: Respond to input operations on the Q&A page and obtain question information; A question classification model is called based on the question information. In the question classification model, based on the semantic information of the question information, the compatibility information between the question information and multiple question-answering models is analyzed. The multiple question-answering models have different reasoning capabilities, and the reasoning capability of the question-answering model is positively correlated with the complexity of the question information. Based on the fitness information, selecting a target question answering model from the multiple question answering models; The target question-answering model is called to perform question-answering processing on the question information to obtain answer information corresponding to the question information.

9. A medical information processing device, characterized in that: include: An acquisition module, a first calling module, a selection module, and a second calling module; The acquisition module is used to respond to input operations on the medical question and answer page and acquire medical question information; The first calling module is configured to call a question classification model based on the medical question information, wherein the question classification model analyzes, based on semantic information of the medical question information, the compatibility information between the medical question information and multiple medical question-answering models, wherein the multiple medical question-answering models have different reasoning capabilities, and the reasoning capabilities of the medical question-answering models are positively correlated with the complexity of the medical question information; The selection module is configured to select a first medical question-answering model from the plurality of medical question-answering models based on the fitness information; The second calling module is used to call the first medical question and answer model to perform question and answer processing on the medical question information to obtain medical answer information corresponding to the medical question information.

10. An electronic device, characterized in that: include: memory and processor; The memory is used to store a computer program; the processor is coupled to the memory and is used to execute the computer program to implement the steps in the method according to any one of claims 1 to 7 and claim 8.

11. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the processor is caused to implement the steps of the method according to any one of claims 1 to 7 and claim 8.

12. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instructions are executed by a processor, the processor is caused to implement the steps of the method according to any one of claims 1 to 7 and claim 8.

Images