Artificial intelligence-based drug recommendation methods and related devices

By generating follow-up question guidelines through a high-order large model, a small-parameter intelligent shopping guide model can be guided to engage in dialogue with users, solving the problem of high computing power consumption in existing technologies and achieving high efficiency and real-time drug recommendation.

CN120632210BActive Publication Date: 2026-11-17ALI HEALTH TECH CO LTD
View PDF 3 Cites 0 Cited by

Patent Information

Application Number
CN202510726903.6
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2025-05-30
Publication Date
2026-11-17
Estimated Expiration
2045-05-30

AI Technical Summary

Technical Problem

Existing intelligent shopping guide models require a large amount of computing resources when recommending medicines, making it difficult to meet real-time and high-concurrency requirements. Furthermore, the large number of parameters leads to high deployment and maintenance complexity.

Method used

A high-order large model is used to generate a follow-up question guide, which guides a smaller-parameter intelligent shopping guide model to ask questions during dialogue with users in order to obtain the information needed for drug recommendations, thereby reducing the consumption of computing resources.

Benefits of technology

While ensuring the accuracy of drug recommendations, the system reduced the consumption of computing resources, improved response speed and real-time performance, and optimized the user experience.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN120632210B_ABST
    Figure CN120632210B_ABST
Patent Text Reader

Abstract

This application provides an artificial intelligence-based drug recommendation method and related apparatus. The method includes: determining a target symptom category corresponding to a user from multiple symptom categories; wherein each of the multiple symptom categories corresponds to a follow-up question guide generated by a high-order large model for that symptom category; the follow-up question guide is used to indicate information that needs to be obtained from the user before recommending a drug to the user; during a dialogue with the user, based on the generated dialogue content and the follow-up question guide corresponding to the target symptom category, an intelligent shopping guide model is invoked to generate a question message to ask the user; wherein the parameter size of the intelligent shopping guide model is smaller than the parameter size of the high-order large model. The artificial intelligence-based drug recommendation method and related apparatus can reduce the consumption of computing resources while ensuring the accuracy of drug recommendations.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The embodiments described in this application relate to the field of computer technology, and in particular to a drug recommendation method and related apparatus based on artificial intelligence. Background Technology

[0002] With the rapid development of internet healthcare, pharmaceutical sales have become one of the main businesses of e-commerce platforms. To improve the efficiency and experience of online drug purchases, e-commerce platforms offer intelligent shopping guides that interact with users and recommend drugs based on these interactions. To enhance the accuracy of these intelligent guides, the platforms train their ability to identify user needs and make drug recommendations based on drug-related information, user descriptions of their conditions, and historical recommendation records. However, invoking these intelligent guides to make drug recommendations consumes significant computing resources. Summary of the Invention

[0003] In view of this, multiple embodiments of this application aim to provide an artificial intelligence-based drug recommendation method and related apparatus, which can reduce the consumption of computing resources while ensuring the accuracy of drug recommendations.

[0004] One embodiment of this application provides an artificial intelligence-based drug recommendation method, comprising: determining a target symptom category corresponding to a user among multiple symptom categories; wherein, each of the multiple symptom categories corresponds to a follow-up question guide generated by a high-order large model for the symptom category; the follow-up question guide is used to indicate the information that needs to be obtained from the user before recommending a drug to the user; during a dialogue with the user, based on the generated dialogue content and the follow-up question guide corresponding to the target symptom category, an intelligent shopping guide model is invoked to generate a question message to ask the user; wherein, the parameter size of the intelligent shopping guide model is smaller than the parameter size of the high-order large model.

[0005] One embodiment of this application provides an artificial intelligence-based drug recommendation method, comprising: acquiring drug applicability information and actual dialogue information corresponding to symptom categories; wherein the actual dialogue information is dialogue information generated between a user and customer service; based on the drug applicability information and the actual dialogue information corresponding to symptom categories, invoking a high-order large model to generate follow-up guidance for each symptom category; wherein the follow-up guidance is used to indicate information that needs to be obtained from the user before recommending the drug to the user; wherein the follow-up guidance is used to provide intelligent shopping guide models that interact with the user, and the parameter size of the intelligent shopping guide model is smaller than the parameter size of the high-order large model.

[0006] One embodiment of this application provides an artificial intelligence-based drug recommendation device, comprising: a category determination module, used to determine a target symptom category corresponding to a user among multiple symptom categories; wherein, each of the multiple symptom categories corresponds to a follow-up question guide generated by a high-order large model for the symptom category; the follow-up question guide is used to indicate information that needs to be obtained from the user before recommending a drug to the user; and a message generation module, used to, during a dialogue with the user, invoke an intelligent shopping guide model based on the generated dialogue content and the follow-up question guide corresponding to the target symptom category to generate a question message to ask the user; wherein, the parameter size of the intelligent shopping guide model is smaller than the parameter size of the high-order large model.

[0007] One embodiment of this application provides an artificial intelligence-based drug recommendation device, comprising: an information acquisition module, used to acquire drug applicability information and actual dialogue information corresponding to symptom categories; wherein the actual dialogue information is dialogue information generated between a user and customer service; and a guide generation module, used to call a high-order large model to generate follow-up guidance for each symptom category based on the drug applicability information and the actual dialogue information corresponding to the symptom categories; wherein the follow-up guidance is used to indicate information that needs to be obtained from the user before recommending the drug to the user; wherein the follow-up guidance is used to provide to an intelligent shopping guide model that interacts with the user, the parameter size of the intelligent shopping guide model being smaller than the parameter size of the high-order large model.

[0008] One embodiment of this application provides a computer device, the computer device including a memory and a processor, the memory storing at least one computer program, the at least one computer program being loaded and executed by the processor to implement the aforementioned artificial intelligence-based drug recommendation method.

[0009] One embodiment of this application provides a computer-readable storage medium storing at least one computer program that, when executed by a processor, can implement the aforementioned artificial intelligence-based drug recommendation method.

[0010] One embodiment of this application provides a computer program product for implementing the aforementioned artificial intelligence-based drug recommendation method.

[0011] In several embodiments provided in this application, a high-order large-scale model generates follow-up question guidelines for symptom categories. These guidelines indicate the information that needs to be obtained from the user before recommending medication. During a dialogue with the user, based on the dialogue content and the follow-up question guidelines for the user's target symptom category, a smart shopping guide model with a smaller parameter size than the high-order large-scale model can be invoked to generate a question message to trigger continued dialogue until sufficient information for medication recommendation is obtained. The follow-up question guidelines are generated by the high-order large-scale model based on language understanding and knowledge integration capabilities, and are then used as the basis for medication recommendations by the smart shopping guide model. The smart shopping guide model does not need to have the ability to generate follow-up question guidelines; therefore, it can have a smaller parameter size. Invoking the smart shopping guide model can reduce the consumption of computing resources while ensuring the accuracy of medication recommendations. Attached Figure Description

[0012] Figure 1 This is a system diagram illustrating the application of an artificial intelligence-based drug recommendation method according to one embodiment of this application.

[0013] Figure 2 A flowchart of an artificial intelligence-based drug recommendation method provided for one embodiment of this application.

[0014] Figure 3 A flowchart of an artificial intelligence-based drug recommendation method provided for another embodiment of this application.

[0015] Figure 4 A block diagram of an artificial intelligence-based drug recommendation device provided for one embodiment of this application.

[0016] Figure 5 A block diagram of an artificial intelligence-based drug recommendation device provided for another embodiment of this application.

[0017] Figure 6 A schematic diagram of a computer device provided for one embodiment of this application. Detailed Implementation

[0018] The technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments.

[0019] In the description of the embodiments of this application, it should be understood that the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, features defined with "first" and "second" may explicitly or implicitly include one or more of the stated features. In the description of the embodiments of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0020] In the traditional online drug purchase process, users need to manually formulate symptom keywords, drug types, brands, and names, and then input this information into the e-commerce platform. The platform then sorts and displays the results based on sales volume, customer reviews, or the merchant's advertising strategy. This method requires users to have some medical knowledge or understand drug naming and efficacy; otherwise, it's unlikely that the e-commerce platform will return the expected results in a single search.

[0021] To help users without relevant knowledge purchase medicines online, e-commerce platforms offer intelligent shopping guides that obtain the necessary information for drug recommendations from users through dialogue. These intelligent shopping guides are question-and-answer robots based on question templates and preset rules. They can guide users to provide more information to help make accurate drug recommendations by asking specific questions (such as, "What are your main symptoms?" or "How long have your symptoms lasted?"). The information obtained through dialogue allows for adaptive drug recommendations.

[0022] In related technologies, large-scale deep learning models (e.g., large pre-trained models with hundreds of billions or trillion parameters) are typically used as intelligent shopping guides to handle tasks such as natural language understanding, symptom recognition, and recommendation decisions. Because large-scale models possess better semantic representation and language understanding capabilities, and can absorb knowledge from multiple domains and disciplines, they can combine medical, pharmaceutical, and everyday language to generate richer question-and-answer strategies. When applied to drug recommendation, they possess the expected user demand mining capabilities and accurate drug recommendation capabilities. However, when large-scale models are used as intelligent shopping guides, they suffer from problems such as requiring significant computing resources, high deployment and maintenance complexity, and slow response speeds, making it difficult to meet the real-time and high-concurrency requirements of drug recommendation scenarios.

[0023] Therefore, it is necessary to provide an AI-based drug recommendation method with a high-order large model that generates follow-up question guidelines for symptom categories. These guidelines indicate the information that needs to be obtained from the user before recommending a drug. During the dialogue with the user, based on the dialogue content and the follow-up question guidelines for the user's target symptom category, a smart recommendation model with a smaller parameter size than the high-order large model can be invoked to generate a question message to trigger continued dialogue until sufficient information for drug recommendation is obtained. The follow-up question guidelines are generated by the high-order large model based on language understanding and knowledge integration capabilities, and then used as the basis for drug recommendations by the smart recommendation model. The smart recommendation model does not need to have the ability to generate follow-up question guidelines, thus allowing for a smaller parameter size. Invoking the smart recommendation model can reduce the consumption of computational resources while ensuring the accuracy of drug recommendations.

[0024] Please see Figure 1 In several embodiments provided in this application, the AI-based drug recommendation method can be applied to an AI-based drug recommendation device. The AI-based drug recommendation device can be an electronic device with certain computing power and network access capabilities. This electronic device can be a desktop computer, laptop computer, tablet computer, smartphone, or a server. The server can also be a distributed server, including multiple processors, memory, network communication modules, etc., working collaboratively to achieve various functions. Alternatively, the server can be a server cluster formed by several servers, possessing higher computing and data processing capabilities. With the development of science and technology, the server can also be implemented using new technological means, such as a new type of "server" based on quantum computing. Of course, in some embodiments, the operating state determination device can also be a program module running in the electronic device.

[0025] In this embodiment, the AI-based drug recommendation device can communicate with the client through dialogue. During the dialogue, it determines the user's target symptom category from multiple symptom categories. Since a high-order large model generates follow-up guidance for each symptom category, and this follow-up guidance indicates the information that needs to be obtained from the user before recommending a drug, a smart shopping guide model with a smaller parameter size than the high-order large model can be invoked based on the generated dialogue content and the follow-up guidance corresponding to the target symptom category to generate a question message for the user. This method enables multi-round dialogue with the client until sufficient information is obtained for drug recommendation, and then drug recommendations are made to the client accordingly.

[0026] In this embodiment, the AI-based drug recommendation device can also be used to call a high-level big data model to generate follow-up guidance for each symptom category based on the drug's applicable scope information and the actual dialogue information corresponding to the symptom category, and then provide the follow-up guidance to the intelligent shopping guide model. Optionally, the intelligent shopping guide model and the high-level big data model run on different electronic devices. The electronic device running the intelligent shopping guide model is used to communicate with the client, and the electronic device running the high-level big data model is used to communicate with the electronic device running the intelligent shopping guide model.

[0027] One embodiment of this application provides an example of an application scenario for an AI-based drug recommendation method. This AI-based drug recommendation method is applied to an AI-based drug recommendation device to reduce the consumption of computing resources while ensuring the accuracy of drug recommendations.

[0028] For example, e-commerce platforms may include the following medication: XX brand aluminum magnesium carbonate chewable tablets. The indications for XX brand aluminum magnesium carbonate chewable tablets include: symptoms such as stomach pain, heartburn, and acid reflux caused by excessive stomach acid; indigestion and bloating; adjunctive treatment for acute gastritis and acute exacerbations of chronic gastritis; and prevention of gastric mucosal damage caused by nonsteroidal anti-inflammatory drugs (NSAIDs) or other medications.

[0029] During the conversation with the user, the user asked, "I've been having stomach pain lately, what medicine can I take to relieve it?"

[0030] Furthermore, based on the existing dialogue content and the follow-up questions corresponding to stomach pain, the intelligent shopping guide model is invoked to generate a question for the user: "What is your gender?"

[0031] The follow-up question guide is generated by a high-order large model, whose parameter size is larger than that of the intelligent shopping guide model. The follow-up question guide for stomach pain is represented as follows:

[0032]

[0033] Furthermore, the intelligent shopping guide model can also follow up with questions based on the follow-up question guide and the user's response to "What is your gender?", such as "How long have your symptoms lasted?".

[0034] After obtaining sufficient information such as gender, age, symptoms, and duration of symptoms through repeated questioning to make accurate drug recommendations, and based on existing conversation content such as "So what can I take to relieve the symptoms? I need something that works quickly" and the user's high response speed in the conversation, it was analyzed that the user had a relatively positive willingness to communicate. In addition, it was also detected that the conversation content included "Quickly recommend something" and "Is the cure rate for this disease high?".

[0035] Based on this, and taking into account user needs such as "symptom relief", concerns about drug safety such as "worrying about not being able to use it in combination with lactobacillus tablets", and user preferences such as "wanting to take a less potent medication", we selected XX brand aluminum magnesium carbonate chewable tablets from a large number of drugs for recommendation.

[0036] In this process, there is no need to call a high-order model with a large number of parameters. Instead, a smart shopping guide model with a smaller number of parameters is obtained by distilling and training the high-order model. The smart shopping guide model is then used to perform drug recommendations, which can save computing power.

[0037] Please see Figure 2 This application provides an artificial intelligence-based drug recommendation method. This AI-based drug recommendation method can be applied to an AI-based drug recommendation device. The AI-based drug recommendation method may include the following steps.

[0038] Step S110: Among multiple symptom categories, determine the target symptom category corresponding to the user; wherein, each of the multiple symptom categories corresponds to a follow-up question guide generated by a high-order large model for the symptom category; the follow-up question guide is used to indicate the information that needs to be obtained from the user before recommending medicine to the user.

[0039] Step S120: During the dialogue with the user, the intelligent shopping guide model is invoked based on the generated dialogue content and the follow-up question guide corresponding to the target symptom category to generate a question message to ask the user; wherein, the parameter size of the intelligent shopping guide model is smaller than the parameter size of the high-order large model.

[0040] In this embodiment, symptom categories refer to a series of abnormal changes in function, metabolism, and morphology within the body during the course of illness, resulting in subjective abnormal sensations or objective pathological changes in the user. Symptom categories include, but are not limited to: pain, dizziness, fever, jaundice, dyspnea, mucosal bleeding, abdominal mass, obesity, emaciation, polyuria, and oliguria.

[0041] In this embodiment, the target symptom category corresponding to the user can be determined based on the generated dialogue content, or based on the user's search / browsing history within a unit of time before the dialogue. The target symptom category corresponding to the user (e.g., hair loss) is selected from multiple symptom categories.

[0042] In this embodiment, a high-order large-scale model generates follow-up question guidelines for each symptom category. The high-order large-scale model is a deep learning model pre-trained on a large-scale medical and conversational corpus with a large number of parameters (e.g., hundreds of billions of parameters). It possesses the expected language understanding, knowledge reasoning, and text generation capabilities, and can generate high-quality follow-up question guidelines based on symptom categories. These follow-up question guidelines guide the intelligent purchasing guide model to ask questions based on the generated dialogue content to obtain more comprehensive and accurate disease-related information for precise drug recommendations. Disease-related information includes, but is not limited to: gender, age, allergy history, medical history, self-care ability, symptom duration, severity, previous medications, and risks associated with concomitant medications.

[0043] In this implementation, the follow-up questioning guide includes follow-up questioning sequences for various scenarios. Different scenarios correspond to dialogue content with different amounts of information, avoiding repeated follow-up questions on information already provided by the user in the dialogue (e.g., gender). This helps improve the efficiency of drug recommendations and optimize the user's experience in the dialogue with the sales guide, making the questions output by the intelligent sales guide model more closely resemble real human responses. For example, the follow-up questioning guide includes the following follow-up questioning sequences: "gender-symptom-duration" for scenarios with no known information, "symptom-duration" for scenarios with known gender, and "gender-duration" for scenarios with known symptoms.

[0044] Furthermore, the order of follow-up questions for different symptom categories may be the same, different, or partially the same in different / same scenarios, and the implementation method of this application does not limit this. For example, for the symptom category of thirst and polyuria, if the patient is known to have gestational diabetes, the questioning order specified in the follow-up guidelines may be "duration of symptoms - other symptoms." Since the sex of patients with gestational diabetes is definitely female, the follow-up guidelines do not inquire about the sex. For example, for the symptom category of baldness, if the duration of the symptom is known to be one year and there are no other symptoms, the questioning order specified in the follow-up guidelines may be "sex - family history." Since the probability of male inheritance of baldness is higher than that of females, it is necessary to inquire about the sex and family history. It can be seen that different symptoms may correspond to different follow-up guidelines to conduct targeted and effective follow-up questions, which is conducive to accurate drug recommendations.

[0045] In this embodiment, the intelligent product recommendation model is an artificial intelligence model built based on deep learning or lightweight machine learning techniques. It has a smaller parameter size than high-order large models and is used to execute follow-up question guidance and provide corresponding drug recommendations during dialogue. The intelligent product recommendation model combines the dialogue context and the requirements of the follow-up question guidance to output specific question messages and the final drug recommendations.

[0046] In this embodiment, during a dialogue with the user, an intelligent product recommendation model can be invoked based on the generated dialogue content (e.g., "I've been having headaches lately, especially in my left temple") and the follow-up question guide corresponding to the target symptom category (e.g., headache). The intelligent product recommendation model can determine the key information needed for recommended medications based on the dialogue content and generate question messages according to the question order corresponding to the follow-up question guide. The question messages (e.g., "How long has your headache lasted?", "Besides your temples, are there any other areas of discomfort?", "Have you taken any painkillers? If so, how effective were they?", "Do you have high blood pressure or other chronic diseases?") seek information from the user that is not included in the existing dialogue content. The purpose of generating question messages is to accurately collect the supplementary information needed for recommendations.

[0047] In some implementations, the question message can contain multiple questions, allowing the user to respond to multiple questions in one round of conversation (e.g., "Persistent for three days, only in the temples, took acetaminophen, only provided temporary relief, no high blood pressure"), improving information collection efficiency and thus improving drug recommendation efficiency.

[0048] In some implementations, users may accidentally input information during the conversation. Therefore, the user's response to the question message and the context can be combined to evaluate the user's corresponding symptom category (e.g., bloating). If the evaluation result is inconsistent with the target symptom category (e.g., stomach pain), the target symptom category is updated. In subsequent conversations, the intelligent shopping guide model is invoked based on the generated conversation content and the follow-up question guide corresponding to the updated target symptom category (e.g., bloating) to generate a question message to ask the user, so as to ensure that the information required for drug recommendations can be accurately obtained.

[0049] In some implementations, when a potential medical risk (e.g., a history of drug allergies) is detected in the conversation, a prompt message is generated to remind the user to seek medical treatment offline (e.g., "It is recommended to seek medical treatment or further examination first").

[0050] In this embodiment, a high-order large model generates follow-up question guidelines for symptom categories. These guidelines indicate the information needed from the user before recommending medication. During dialogue with the user, based on the dialogue content and the follow-up question guidelines for the user's target symptom category, a smart product recommendation model with a smaller parameter size than the high-order large model can be invoked to generate a question message to trigger continued dialogue until sufficient information for medication recommendation is obtained. The follow-up question guidelines are generated by the high-order large model based on language understanding and knowledge integration capabilities, and are used as the basis for medication recommendations by the smart product recommendation model. The smart product recommendation model does not need to have the ability to generate follow-up question guidelines; therefore, it can have a smaller parameter size. Invoking the smart product recommendation model can reduce the consumption of computing resources while ensuring the accuracy of medication recommendations.

[0051] In some implementations, the AI-based drug recommendation device can analyze the user's willingness to communicate based on the generated dialogue content during the conversation; if the willingness to communicate meets the conditions for continuing communication, it can invoke the intelligent shopping guide model based on the generated dialogue content and the follow-up question guide corresponding to the target symptom category to generate a question message to ask the user.

[0052] In this implementation, communication willingness refers to a user's subjective inclination or motivation to continue interacting with the intelligent shopping guide system, answering questions, and obtaining recommendations in the current round of conversation. In some implementations, communication willingness is categorized into several types, including: a user's positive willingness to seek further consultation, a user's ambiguous willingness to hesitate about continuing communication, and a user's negative willingness to dislike the content of the conversation. In some implementations, communication willingness is expressed numerically as the degree to which a user is willing to continue communicating.

[0053] In some implementations, analyzing communication intent based on the generated dialogue content can be done by classifying the intent of user responses in the dialogue. Specifically, this includes classifying communication intent as positive when user responses express questions, detailed descriptions of symptoms, or a desire to learn more; and classifying communication intent as negative when user responses contain perfunctory (e.g., "Never mind"), rejection (e.g., "No"), or complaints (e.g., "That's not smart enough").

[0054] Alternatively, analyzing the communication intention based on the generated dialogue content can be done by analyzing the emotional tendency (e.g., positive, neutral, negative) and tone intensity (e.g., exaggerated sentences, interrogative sentences, imperative sentences) of user responses, identifying higher positive emotions or questioning tones as signals of communication intention.

[0055] Alternatively, the communication intention can be analyzed based on the generated dialogue content as follows: the communication intention is determined based on the response duration and / or response frequency of the user's reply. The degree of positive communication intention is negatively correlated with response duration and positively correlated with response frequency. That is, the shorter the response duration or the higher the response frequency, the higher the communication intention.

[0056] In this embodiment, when the willingness to communicate is identified as a positive intention or the value of the willingness to communicate is greater than a specified value, the willingness to communicate is deemed to meet the conditions for continuing communication. The conditions for continuing communication are used to limit the intensity of the willingness required to continue communication.

[0057] In this implementation, analyzing the willingness to communicate can avoid excessive questioning or awkward silences. When the willingness to communicate is low, i.e., when the willingness to communicate does not meet the conditions for continuing communication, the conversation can be ended in a timely manner, switching to a concise recommendation mode or switching to a human customer service mode. This avoids the user feeling disturbed or wasting time, and also saves on model calls, thereby conserving computing power. When the willingness to communicate is high, i.e., when the willingness to communicate meets the conditions for continuing communication, the intelligent shopping guide model can be invoked based on the generated conversation content and the follow-up question guide corresponding to the target symptom category to generate question messages for the user, enabling appropriately in-depth questioning to obtain more information that helps in accurately recommending medicines.

[0058] In some implementations, the AI-based drug recommendation device can generate drug recommendation information based on the dialogue content when it detects that the degree of information collection meets expectations, or when it detects that the dialogue content contains information requesting drug recommendations from the user, or when it analyzes that the user's acceptance of drug recommendations meets expectations; wherein, the degree of information collection is used to indicate the extent to which the dialogue content conforms to the follow-up guidelines corresponding to the target symptom category.

[0059] In this embodiment, to make timely drug recommendations, the information collection level is deemed to meet expectations when multiple types of information as indicated by the follow-up guidelines are collected. Based on this, drug recommendation information is generated according to the dialogue content. This avoids inappropriate recommendations due to insufficient information and allows for more personalized and relevant drug recommendations, preventing the need for secondary communication due to insufficient initial communication. Specifically, the information collection level refers to the matching ratio or completeness between the user information obtained in the current dialogue (such as symptom details, medication history, allergy history, etc.) and the required items listed in the follow-up guidelines corresponding to that symptom category. The expectation refers to the limitation on the types of information that should be collected (e.g., gender, age, duration, symptoms, previous medications, etc.) / matching ratio (e.g., 80%).

[0060] In this embodiment, if a user requests a drug recommendation during the conversation, drug recommendation information is generated promptly based on the conversation content to respect the user's initiative. In some embodiments, the information collected from the existing conversation content may be insufficient, but drug recommendation information can still be generated based on the information already contained in the conversation content. In some embodiments, if the information collected from the existing conversation content is insufficient, a question message is generated by calling the intelligent shopping guide model according to the follow-up question guide to prompt the user to provide more information for accurate drug recommendations.

[0061] In this implementation, the user's acceptance of drug recommendations can be analyzed based on one or more of the following information: dialogue content (e.g., containing positive sentiment words), historical dialogue records (e.g., records of receiving drug recommendations), user profiles, and behavioral characteristics (e.g., frequent proactive inquiries, fast response times, and lengthy responses). Drug recommendation acceptance measures a user's attitude towards receiving drug recommendations as quickly as possible. When the drug recommendation acceptance meets the expected acceptance level, drug recommendation information is generated based on the dialogue content. When the drug recommendation acceptance does not meet the expected acceptance level, forced recommendations can be avoided to prevent negative user emotions. For users with high drug recommendation acceptance, multiple rounds of in-depth dialogue can be conducted; for users with low drug recommendation acceptance, a concise response or transfer to a human agent is provided to rationally allocate the resources of the product recommendation system.

[0062] In this embodiment, drug recommendation information can be represented as text, interactive cards, links, controls, etc., and includes information such as drug name, dosage, usage, and precautions. For example, the drug recommendation information could be: "Based on your current headache, you may consider taking ibuprofen extended-release tablets, 200mg each time, twice a day; if you experience stomach discomfort, you can also supplement with a gastric mucosal protectant."

[0063] In some implementations, the AI-based drug recommendation device is also equipped with a timed follow-up function. This function sends follow-up information to the user at specified times according to the medication cycle corresponding to the drug recommendation information (e.g., how is the pain relief after medication? Are there any adverse reactions?). This allows for continuous optimization of the intelligent product recommendation model based on the user's feedback content and timing, thereby improving the recommendation accuracy of the intelligent product recommendation model. The specified time is related to the medication cycle, and the specified time may differ for different drug recommendation information.

[0064] In some implementations, the AI-based drug recommendation device can extract reference information from the generated dialogue content during the dialogue with the user, and call the intelligent shopping guide model based on the reference information to generate drug recommendation information; wherein, the reference information includes at least one of the following: user needs, drug safety-related doubts, and user preferences.

[0065] In this embodiment, to make accurate drug recommendations, user needs can be analyzed using a multi-class intent classifier, a named entity recognition model, and a contextual association analysis model to categorize user input dialogue information into user intentions such as "symptom relief," "preventive medication," and "chronic maintenance." User needs are used to indicate, to a certain extent, the user's expectations for drug recommendation information.

[0066] In this embodiment, the system can also identify user concerns about medication safety based on questions expressed in the conversation (e.g., "Are there any side effects? Can it be used with XXX drug?") and negative emotional content in the analyzed conversation (e.g., "This drug has serious side effects!"). These concerns about medication safety represent user worries about medication safety and can be used to guide the purchasing system to provide targeted responses.

[0067] In this embodiment, user preferences can also be analyzed based on historical purchase records, click-through rates, historical drug reviews, user profiles, and user self-reports (e.g., "I like xxx", "I don't want to eat XXX"). User preferences can be updated in real time with each round of conversation to ensure that the user preferences used to make the final drug recommendation accurately represent the user.

[0068] In this embodiment, the reference information including user needs, drug safety concerns, and user preferences refers to the data extracted and structured from each input of the user during multi-round dialogues, which can be used to guide drug recommendation decisions.

[0069] Please see Figure 3 This application provides an artificial intelligence-based drug recommendation method. This AI-based drug recommendation method can be applied to an AI-based drug recommendation device. The AI-based drug recommendation method may include the following steps.

[0070] Step S210: Obtain the applicable scope information of the medicine and the actual dialogue information corresponding to the symptom category; wherein, the actual dialogue information is the dialogue information generated between the user and customer service.

[0071] Step S220: Based on the drug's applicable scope information and the actual dialogue information corresponding to the symptom category, call the high-order large model to generate follow-up question guidelines for each symptom category; wherein, the follow-up question guidelines are used to indicate the information that needs to be obtained from the user before recommending the drug to the user; wherein, the follow-up question guidelines are used to provide to the intelligent shopping guide model that is interacting with the user, and the parameter size of the intelligent shopping guide model is smaller than the parameter size of the high-order large model.

[0072] In this embodiment, the drug's scope of application information includes indications, dosage, contraindications, and population restrictions, indicating in which disease scenarios the drug can be used safely and effectively. This scope of application information can be extracted from the drug's instructions and / or a medical knowledge base.

[0073] In this implementation, the actual dialogue information corresponding to the symptom categories records the dialogue records generated in real consultation scenarios, specifically including the symptom descriptions raised by the user, follow-up questions, customer service answers, sales guide scripts, and details of multiple rounds of dialogue, which can express the user's true language expression habits, concerns, and needs.

[0074] For example, the symptom category is excessive stomach acid. Sales pitches may include, but are not limited to, the following:

[0075]

[0076]

[0077] In this implementation, the higher-order large-scale model with a larger parameter scale possesses higher semantic understanding and knowledge integration capabilities compared to the intelligent shopping guide model with a smaller parameter scale. By invoking the higher-order large-scale model based on the applicable scope information and actual dialogue information, follow-up question guidelines corresponding to each symptom category can be generated. Invoking the intelligent shopping guide model based on these follow-up question guidelines eliminates the need for the strategy generation task that the intelligent shopping guide model originally required. Invoking the intelligent shopping guide model within the shopping guide system can more efficiently obtain the required shopping guide sessions and drug recommendation results.

[0078] In some implementations, an AI-based drug recommendation device can generate a thought chain by calling a high-order large model based on the actual dialogue information corresponding to the symptom category; wherein, the thought chain is used to provide the intelligent shopping guide model as a basis for recommending at least one drug among user needs, drug safety-related doubts, and user preferences.

[0079] In this implementation, the "thought chain" refers to the intermediate reasoning steps presented in readable text form before the high-order large model generates the final reasoning result. These steps can represent the entire thinking process from input (such as user symptom descriptions and dialogue context) to output (such as recommendation basis), making the reasoning path transparent and easy to trace. Calling the high-order large model to generate thought chains leverages its stronger language processing capabilities to mine causal relationships and medication logic in the corpus and represent them as thought chains. These thought chains serve as the basis for the distillation training of the intelligent shopping guide model, constraining its reasoning process and ensuring its rationality and accuracy, thus reducing the illusion rate of the intelligent shopping guide model.

[0080] In some implementations, the AI-based drug recommendation device can invoke the higher-order large model based on the actual dialogue information corresponding to the symptom category, so that the higher-order large model generates reference inference results based on the thought chain; the reference inference results are used to distill and train the intelligent shopping guide model; the similarity evaluation is performed on the output of the intelligent shopping guide model based on the reference inference results, and the parameters of the intelligent shopping guide model are iterated based on the similarity evaluation; wherein, the similarity evaluation is used to guide the intelligent shopping guide model to output reference inference results that approximate the higher-order large model; the similarity evaluation is generated based on the thought chain loss value and the model parameter loss value between the reference inference results and the output of the intelligent shopping guide model; the weight of the thought chain loss value is less than the weight of the model parameter loss value.

[0081] In this embodiment, in order to obtain an intelligent shopping guide model with reasoning capabilities close to those of a high-order large model, the intelligent shopping guide model can be trained by distillation based on the reference reasoning results output by the high-order large model. In the distillation training, the high-order large model is the teacher model and the intelligent shopping guide model is the student model. The training objective of the intelligent shopping guide model is to imitate the behavior of the teacher model. The intelligent shopping guide model trained in this way can reduce the parameter size and reasoning cost while maintaining performance.

[0082] In this embodiment, the reference reasoning result refers to the information output by the higher-order large model, based on the thought chain, which includes the complete reasoning trajectory and final decision, given the input. By iterating the parameters of the intelligent shopping guide model, the output of the intelligent shopping guide model is made to approximate the reasoning trajectory and final decision of the higher-order large model.

[0083] In this embodiment, similarity evaluation is used to quantify the output gap between the intelligent shopping guide model and the high-order large model. Optionally, the similarity evaluation is specifically a weighted sum of the thought chain loss value and the model parameter loss value. The thought chain loss value characterizes the difference in intermediate inference steps between the intelligent shopping guide model and the high-order large model, while the model parameter loss value characterizes the difference in the final output between the intelligent shopping guide model and the high-order large model. The thought chain loss value and the model parameter loss value can be calculated based on methods such as cross-entropy and mean squared error. The weight of the thought chain loss value is less than the weight of the model parameter loss value, aiming to guide the training results to first ensure the matching degree between the final product recommendation and user needs, which is conducive to obtaining an intelligent shopping guide model more suitable for application in shopping guide systems.

[0084] Please see Figure 4This application also provides an artificial intelligence-based drug recommendation device, comprising: a category determination module, used to determine a target symptom category corresponding to a user among multiple symptom categories; wherein, each of the multiple symptom categories corresponds to a follow-up question guide generated by a high-order large model for the symptom category; the follow-up question guide is used to indicate information that needs to be obtained from the user before recommending a drug to the user; and a message generation module, used to, during a dialogue with the user, invoke an intelligent shopping guide model based on the generated dialogue content and the follow-up question guide corresponding to the target symptom category to generate a question message to ask the user; wherein, the parameter size of the intelligent shopping guide model is smaller than the parameter size of the high-order large model.

[0085] The specific functions and effects of the AI-based drug recommendation device can be explained by referring to other embodiments of this application, and will not be repeated here. Each module in the AI-based drug recommendation device can be implemented entirely or partially through software, hardware, or a combination thereof. Each module can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0086] Please see Figure 5 This application also provides an artificial intelligence-based drug recommendation device, comprising: an information acquisition module for acquiring drug applicability information and actual dialogue information corresponding to symptom categories; wherein the actual dialogue information is dialogue information generated between a user and customer service; and a guide generation module for calling a high-order large model to generate follow-up guidance for each symptom category based on the drug applicability information and the actual dialogue information corresponding to the symptom categories; wherein the follow-up guidance is used to indicate information that needs to be obtained from the user before recommending the drug to the user; and wherein the follow-up guidance is used to provide to an intelligent shopping guide model that interacts with the user, wherein the parameter size of the intelligent shopping guide model is smaller than the parameter size of the high-order large model.

[0087] The specific functions and effects of the AI-based drug recommendation device can be explained by referring to other embodiments of this application, and will not be repeated here. Each module in the AI-based drug recommendation device can be implemented entirely or partially through software, hardware, or a combination thereof. Each module can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0088] This application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, causes the processor to implement the aforementioned artificial intelligence-based drug recommendation method.

[0089] This application also provides a computer program product containing instructions that, when executed by a processor, cause the computer to implement the aforementioned artificial intelligence-based drug recommendation method.

[0090] Please see Figure 6 This description describes an embodiment that can provide a computer device, the computer device including: a memory, and one or more processors communicatively connected to the memory; the memory stores instructions executable by the one or more processors, the instructions being executed by the one or more processors to cause the one or more processors to implement the aforementioned artificial intelligence-based drug recommendation method.

[0091] In some embodiments, the computer device may include a processor connected to a system bus, a non-volatile storage medium, internal memory, a communication interface, a display device, and an input device. The non-volatile storage medium may store an operating system and related computer programs.

[0092] The user information or user account information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, etc.) involved in various embodiments of this application are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, use and processing of the relevant data must comply with the relevant laws and regulations and standards of the relevant countries and regions, and corresponding operation entry points are provided for users to choose to authorize or refuse.

[0093] It is understood that the specific examples in this document are only intended to help those skilled in the art better understand the embodiments of this application, and are not intended to limit the scope of the invention.

[0094] It is understood that in the various embodiments of this application, the sequence number of each process does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0095] It is understood that the various implementation methods described in this application can be implemented individually or in combination, and the implementation methods in this application are not limited in this respect.

[0096] Unless otherwise stated, all technical and scientific terms used in the embodiments of this application have the same meaning as commonly understood by one of ordinary skill in the art. The terminology used in this application is for the purpose of describing particular embodiments only and is not intended to limit the scope of this application. The term "and / or" as used in this application includes any and all combinations of one or more of the associated listed items. The singular forms "a," "the," and "the" as used in the embodiments of this application and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise.

[0097] It is understood that the processor in the embodiments of this application can be an integrated circuit chip with signal processing capabilities. During implementation, each step of the above method embodiments can be completed by the integrated logic circuits in the processor's hardware or by instructions in software form. The processor can be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this application. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this application can be directly embodied in the execution of a hardware decoding processor, or executed by a combination of hardware and software modules in the decoding processor. The software modules can be located in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. This storage medium is located in memory; the processor reads information from the memory and, in conjunction with its hardware, completes the steps of the above method.

[0098] It is understood that the memory in the embodiments of this application may be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. Specifically, non-volatile memory may be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. Volatile memory may be random access memory (RAM). It should be noted that the memory in the systems and methods described herein is intended to include, but is not limited to, these and any other suitable types of memory.

[0099] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0100] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the aforementioned method implementations, and will not be repeated here.

[0101] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0102] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0103] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0104] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0105] The above description is merely a specific embodiment of this application, but the scope of protection of this invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this invention should be determined by the scope of the claims.

Claims

1. A drug recommendation method based on artificial intelligence, characterized in that, The method includes: Among multiple symptom categories, the target symptom category corresponding to the user is determined; wherein, each of the multiple symptom categories corresponds to a follow-up question guide pre-generated by a high-order large model for the symptom category; the follow-up question guide is used to indicate the information that needs to be obtained from the user before recommending medicine to the user; the follow-up question guide also includes the follow-up question order in multiple scenarios, with different scenarios corresponding to dialogue content with different amounts of information; During the dialogue with the user, the intelligent shopping guide model is invoked based on the generated dialogue content and the follow-up question guide corresponding to the target symptom category. The intelligent shopping guide determines the key information required for the known drug recommendation based on the dialogue content, and generates a question message to ask the user according to the follow-up question order corresponding to the key information in the follow-up question guide. The parameter size of the intelligent shopping guide model is smaller than the parameter size of the higher-order large model.

2. The method according to claim 1, characterized in that, During a dialogue with a user, the intelligent shopping guide model is invoked based on the generated dialogue content and the follow-up question guide corresponding to the target symptom category. The intelligent shopping guide determines the key information required for known drug recommendations based on the dialogue content, and generates a question message to ask the user according to the follow-up question order corresponding to the key information in the follow-up question guide. This process includes: During conversations with users, analyze their willingness to communicate based on the content of the conversations. If the willingness to communicate meets the conditions for continued communication, the intelligent shopping guide model is invoked based on the generated dialogue content and the follow-up question guide corresponding to the target symptom category to generate a question message to ask the user.

3. The method according to claim 1, characterized in that, The method further includes: If the level of information collection is found to be in line with expectations, or if the dialogue content contains information requesting drug recommendations from the user, or if the user's acceptance of drug recommendations is found to be in line with expectations, drug recommendation information is generated based on the dialogue content; wherein, the level of information collection is used to indicate the degree to which the dialogue content conforms to the follow-up questioning guidelines corresponding to the target symptom category.

4. The method according to claim 3, characterized in that, The steps for generating drug recommendation information based on the dialogue content include: During the dialogue with the user, reference information for drug recommendation is extracted from the generated dialogue content, and the intelligent shopping guide model is invoked based on the reference information to generate drug recommendation information; wherein, the reference information includes at least one of the following: user needs, drug safety-related doubts, and user preferences.

5. A drug recommendation method based on artificial intelligence, characterized in that, The method includes: Obtain information on the applicable scope of the medicine and the actual dialogue information corresponding to the symptom category; wherein, the actual dialogue information is the dialogue information generated between the user and customer service. Based on the drug's applicable scope information and the actual dialogue information corresponding to the symptom categories, a high-order large model is invoked to pre-generate follow-up question guidelines for each symptom category. These follow-up question guidelines indicate the information that needs to be obtained from the user before recommending the drug. The guidelines also include follow-up question sequences for various scenarios, with different scenarios corresponding to different amounts of dialogue content. These follow-up question guidelines are provided to an intelligent shopping guide model that interacts with the user, and the parameter size of the intelligent shopping guide model is smaller than that of the high-order large model.

6. The method according to claim 5, characterized in that, The method further includes: Based on the actual dialogue information corresponding to the symptom categories, a high-order large model is invoked to generate a thought chain; wherein, the thought chain is provided to the intelligent shopping guide model as a basis for recommending at least one drug among user needs, drug safety-related doubts, and user preferences.

7. The method according to claim 6, characterized in that, The method further includes: The higher-order big model is invoked based on the actual dialogue information corresponding to the symptom category, so that the higher-order big model generates reference reasoning results based on the thought chain; The intelligent shopping guide model is trained using the reference inference results; The similarity evaluation of the output of the intelligent shopping guide model is performed based on the reference inference results, and the parameters of the intelligent shopping guide model are iterated based on the similarity evaluation. The similarity evaluation is used to guide the intelligent shopping guide model to output a reference inference result that approximates the higher-order large model. The similarity evaluation is generated based on the thought chain loss value and the model parameter loss value between the reference inference result and the output of the intelligent shopping guide model. The weight of the thought chain loss value is less than the weight of the model parameter loss value.

8. A drug recommendation device based on artificial intelligence, characterized in that, The device includes: The category determination module is used to determine the target symptom category corresponding to the user among multiple symptom categories; wherein, each of the multiple symptom categories corresponds to a follow-up question guide pre-generated by a high-order large model for the symptom category; the follow-up question guide is used to indicate the information that needs to be obtained from the user before recommending medicine to the user; the follow-up question guide also includes the follow-up question order in multiple scenarios, with different scenarios corresponding to dialogue content with different amounts of information; The message generation module is used to invoke the intelligent shopping guide model based on the generated dialogue content and the follow-up question guide corresponding to the target symptom category during a dialogue with the user. The intelligent shopping guide determines the key information required for the known drug recommendation based on the dialogue content, and generates a question message to ask the user according to the follow-up question order corresponding to the key information in the follow-up question guide. The parameter size of the intelligent shopping guide model is smaller than the parameter size of the higher-order large model.

9. A drug recommendation device based on artificial intelligence, characterized in that, The device includes: The information acquisition module is used to acquire information on the applicable scope of the medicine and actual dialogue information corresponding to the symptom category; wherein, the actual dialogue information is the dialogue information generated between the user and customer service. The guide generation module is used to pre-generate follow-up guidance for each symptom category by calling a high-order large model based on the drug's applicable scope information and the actual dialogue information corresponding to the symptom category. The follow-up guidance indicates the information that needs to be obtained from the user before recommending the drug. The follow-up guidance also includes follow-up questioning sequences for various scenarios, with different scenarios corresponding to different amounts of dialogue content. The follow-up guidance is provided to an intelligent shopping guide model that interacts with the user, and the parameter size of the intelligent shopping guide model is smaller than the parameter size of the high-order large model.

10. A computer device, characterized in that, The computer device includes a memory and a processor, wherein the memory stores at least one computer program, which is loaded and executed by the processor to implement the artificial intelligence-based drug recommendation method as described in any one of claims 1 to 7.

11. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores at least one computer program, which, when executed by a processor, is capable of implementing the artificial intelligence-based drug recommendation method as described in any one of claims 1 to 7.

12. A computer program product, characterized in that, The computer program product is used to implement the artificial intelligence-based drug recommendation method as described in any one of claims 1 to 7.

Citation Information

Patent Citations

  • Drug recommendation method and device based on artificial intelligence, equipment and storage medium

    CN114974501A

  • Man-machine conversation system and data processing method thereof, electronic equipment and storage medium

    CN116108155A

  • Dialogue information extraction method and device based on large model, equipment and storage medium

    CN119311786A