A medical data analysis auxiliary method and system based on artificial intelligence

Abstract

The present invention discloses an artificial intelligence-based medical data analysis auxiliary method and system, which relates to the field of artificial intelligence, including: obtaining medical record information of target patients, parsing symptom description information with an NLP model after denoising and standardization, and then obtaining medication recommendation information based on a large language model, and finally generating a medication analysis report and pushing it to a preset user interaction interface, thereby realizing efficient analysis of medical data and provision of accurate medication recommendations.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence, and in particular to an artificial intelligence-based medical data analysis auxiliary method and system. Background Art

[0002] In the medical field, accurately analyzing patients' medical data and giving reasonable medication recommendations are crucial for diagnosis and treatment. With the development of artificial intelligence, using its technology to assist in medical data analysis has become a trend. However, existing related solutions still have shortcomings in terms of comprehensiveness of data processing and accuracy of medication recommendations, and cannot well meet the growing demand for efficient and precise medical care. Summary of the invention

[0003] The purpose of the present invention is to provide a medical data analysis auxiliary method and system based on artificial intelligence.

[0004] In a first aspect, an embodiment of the present invention provides an artificial intelligence-based medical data analysis auxiliary method, comprising:

[0005] Obtain medical record information of target patients;

[0006] Performing denoising and standardization processing on the medical record information respectively to obtain pre-processed medical record information;

[0007] Calling a preset NLP model to parse the preprocessed medical record information to obtain symptom description information of the target patient;

[0008] Processing the symptom description information based on a preset large language model to obtain medication recommendation information for the symptom description information;

[0009] A medication analysis report for the target patient is generated based on the medication recommendation information, and the medication analysis report is pushed to a preset user interaction interface.

[0010] In a second aspect, an embodiment of the present invention provides a server system, including a server, wherein the server is used to execute the method described in the first aspect.

[0011] Compared with the prior art, the beneficial effects provided by the present invention include: adopting an artificial intelligence-based medical data analysis auxiliary method and system disclosed in the present invention, by acquiring the medical record information of the target patient, parsing the symptom description information with the NLP model after denoising and standardization, and then obtaining the medication recommendation information based on the large language model, and finally generating a medication analysis report and pushing it to the preset user interaction interface, thereby realizing efficient analysis of medical data and providing accurate medication recommendations. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only illustrate certain embodiments of the present invention and should not be regarded as limiting the scope. For those of ordinary skill in the art, other related drawings can also be obtained based on these drawings without creative work.

[0013] Figure 1 A schematic diagram of the steps of the artificial intelligence-based medical data analysis auxiliary method provided in an embodiment of the present invention;

[0014] Figure 2 A schematic block diagram of the structure of a computer device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0015] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.

[0016] The specific implementation modes of the present invention are described in detail below in conjunction with the accompanying drawings.

[0017] In order to solve the technical problems in the aforementioned background technology, Figure 1 A flowchart of an artificial intelligence-based medical data analysis assistance method provided in an embodiment of the present disclosure is provided below. The artificial intelligence-based medical data analysis assistance method is introduced in detail.

[0018] Step S201, obtaining medical record information of a target patient;

[0019] Step S202, performing denoising and standardization processing on the medical record information to obtain pre-processed medical record information;

[0020] Step S203, calling a preset NLP model to parse the preprocessed medical record information to obtain symptom description information of the target patient;

[0021] Step S204, processing the symptom description information based on a preset large language model to obtain medication recommendation information of the symptom description information;

[0022] Step S205: Generate a medication analysis report for the target patient based on the medication recommendation information, and push the medication analysis report to a preset user interaction interface.

[0023] In an embodiment of the present invention, illustratively, in the information management system of a large general hospital, the server undertakes the important task of assisting medical data analysis. When a patient comes for a medical consultation or a follow-up visit, the medical staff of each department of the hospital will enter the patient's relevant medical records into the electronic medical record system. For example, a patient named Xiao Li came to the internal medicine department for a cold, fever and cough. After the doctor asked about the condition and conducted various examinations (such as blood routine, temperature measurement, etc.), Xiao Li's basic information (name, age, gender, etc.), symptoms (fever degree, cough frequency, etc.), examination results, preliminary diagnosis and other information were recorded in detail in the electronic medical record. At this time, the server will obtain the medical record information of the target patient Xiao Li from the electronic medical record system of the hospital. This information exists in the form of structured and unstructured data, ready for subsequent analysis and processing. After the server obtains Xiao Li's medical record information, it begins to perform denoising and standardization processing. De-noising: There may be some duplicate information in the medical record information, such as the patient's body temperature recorded by the nurse multiple times at different times, and there may be several times with the same value and almost the same record description. The server uses a text filtering algorithm to automatically identify and remove these duplicate information. At the same time, there may be some irrelevant formatting marks in the medical records, such as some special characters or spaces used for typesetting, which will also be removed. In addition, there are some incorrect statements that do not conform to the medical record specifications. For example, when a doctor enters "cough" in a hurry, he may mistakenly write "cough" as "cough gargle". The server can also correct such incorrect statements by comparing with the medical terminology library and complete the denoising process. Standardization: After denoising, the server then performs standardization. For the date format, different medical staff may record different formats, some write "2023.05.10", and some write "10 / 05 / 2023". The server will unify all date formats into a preset standard format, such as "YYYY-MM-DD". In terms of medical terminology, if the doctor uses abbreviations such as "WBC" (white blood cells) when recording, the server will restore it to the full form "White Blood Cell". In addition, for measurement data in different units, such as some blood pressure is recorded in "mmHg" and some in "kPa", the server will convert them all into a unified standard unit system, such as "mmHg". After this series of denoising and standardization processing, the server obtains the pre-processed medical record information of Xiao Li, making it more standardized and easier for subsequent analysis. The server takes the pre-processed medical record information of Xiao Li for the next step of processing. It will divide this medical record information into multiple text paragraphs according to the preset text segmentation rules. For example, the basic information of the patient is taken as a paragraph, the symptoms are taken as a paragraph, the examination results are taken as a paragraph, and so on. Then, the server calls the preset NLP model to perform part-of-speech tagging on these multiple text paragraphs in turn.Taking the symptom paragraph as an example, for the sentence "fever to 38.5 degrees, cough more frequently", the NLP model will mark "fever" as a verb, "38.5 degrees" as a quantifier, "cough" as a verb, and "more frequently" as an adjective. Based on the results of part-of-speech tagging, the NLP model then performs syntactic analysis on each paragraph to determine the grammatical structure of the sentence. For example, the grammatical structure of the above sentence is "subject (patient implied) + predicate (fever, cough) + complement (to 38.5 degrees, more frequently)". Based on the grammatical structure and part-of-speech tagging information, the NLP model can extract words and phrases related to symptoms, such as "fever" and "cough". After that, the extracted words and phrases related to symptoms are semantically analyzed, and combined with medical knowledge, their exact meanings and relationships are determined. For example, "fever" means that the body temperature is higher than the normal range, and "cough" is a reaction after the respiratory tract is stimulated, and they are symptoms that appear on Xiao Li at the same time. Finally, the server integrates these symptom-related words and phrases, their exact meanings, and their relationships, and generates the symptom description information of the target patient Xiao Li, that is, "Patient Xiao Li has symptoms of fever (body temperature 38.5 degrees) and cough (more frequent)". After the server obtains Xiao Li's symptom description information, it starts to process it based on the preset large language model to obtain medication recommendation information. First, the server obtains Xiao Li's symptom description information "Patient Xiao Li has symptoms of fever (body temperature 38.5 degrees) and cough (more frequent)", as well as the symptom category corresponding to the symptom description information, which belongs to the "respiratory tract infection" symptom category. Next, the server determines the drug database that matches the "respiratory tract infection" symptom category from multiple drug databases. This drug database contains low-dimensional vector representations corresponding to candidate drug information matching the symptom category, such as the commonly used drugs for respiratory tract infections, such as "amoxicillin" and "ibuprofen", which are stored in the form of low-dimensional vectors. Then, the server determines the cosine similarity between the low-dimensional vector representation of Xiao Li's symptom description information and the low-dimensional vector representation in the drug database. For example, for the low-dimensional vectors converted from the symptom description information such as "fever" and "cough", the cosine similarity is calculated with the low-dimensional vector corresponding to "amoxicillin" in the drug database to obtain a specific similarity value. After that, the server determines from the drug database multiple low-dimensional vector representations whose cosine similarity exceeds the preset cosine similarity threshold. Assuming that the preset cosine similarity threshold is 0.6, those low-dimensional vector representations with a calculated similarity greater than 0.6 will be screened out, and the low-dimensional vector representations corresponding to drugs such as "amoxicillin" and "ibuprofen" may be screened out. Based on the feature conversion relationship between the low-dimensional vector representation and the candidate drug information, the candidate drug information that has a feature conversion relationship with these multiple low-dimensional vector representations is determined.That is, through specific conversion rules, the corresponding candidate drug information is restored from the selected low-dimensional vector representation, such as the specific drug names such as "amoxicillin" and "ibuprofen". The candidate drug information that has a feature conversion relationship with multiple low-dimensional vector representations is used as multiple candidate drug information that matches the low-dimensional vector representation of Xiao Li's symptom description information. Next, according to the text description features of Xiao Li's symptom description information and the text description features corresponding to these multiple candidate drug information, the pharmacological adaptation coefficients between Xiao Li's symptom description information and these multiple candidate drug information are determined. For example, "amoxicillin" has a good antibacterial effect on bacterial infections in respiratory tract infections, and has a certain degree of pharmacological adaptation to the situation that Xiao Li's fever may be caused by bacterial infection. A pharmacological adaptation coefficient value is calculated through a specific algorithm. At the same time, according to the keywords of Xiao Li's symptom description information and the keywords corresponding to the target candidate drug information (such as "amoxicillin"), the semantic adaptation between the target candidate drug information and Xiao Li's symptom description information is determined. For example, the keywords of "amoxicillin" include "antibacterial" and "anti-inflammatory", and the semantic correlation with keywords such as "fever" and "cough" in Xiao Li's symptom description information is calculated to obtain a semantic fitness value. In addition, according to the core medical terms in the keywords of Xiao Li's symptom description information (such as "fever" and "cough") and the core medical terms in the keywords corresponding to the target candidate drug information (such as "antibacterial" of "amoxicillin"), the first keyword fitness coefficient between the target candidate drug information and Xiao Li's symptom description information is determined. And according to the non-core medical terms in the keywords of Xiao Li's symptom description information (such as "more frequent") and the non-core medical terms in the keywords corresponding to the target candidate drug information (it may be that "amoxicillin" has no obvious corresponding non-core medical terms, which is assumed to be blank here), the second keyword fitness coefficient between the target candidate drug information and Xiao Li's symptom description information is determined. The semantic fitness, the first keyword fitness coefficient, and the second keyword fitness coefficient corresponding to the target candidate drug information are cumulatively added to obtain the keyword fitness coefficients between Xiao Li's symptom description information and these multiple candidate drug information. Finally, according to the pharmacological adaptation coefficients and keyword adaptation coefficients corresponding to these multiple candidate drug information, the candidate drug information that matches the medication-related features corresponding to Xiao Li's symptom description information is selected from these multiple candidate drug information as the recommended drug information. Assuming that after comprehensive calculation and comparison, "Amoxicillin" performs well in terms of pharmacological adaptation coefficient, keyword adaptation coefficient, etc., it will be recommended to Xiao Li as drug information. After obtaining Xiao Li's medication recommendation information (for example, the recommended drug is "Amoxicillin"), the server starts to generate a medication analysis report.First, the server extracts key information based on the medication recommendation information, including the drug name "Amoxicillin", the dosage (assuming that the doctor prescribes 0.5 grams each time according to Xiao Li's situation), the frequency of medication (three times a day), the expected duration of medication (three days), and adverse reactions (possibly mild gastrointestinal discomfort, etc.). Then, according to the preset report template, these key information are typeset and organized to generate a medication analysis report containing a title (such as "Xiao Li's patient medication analysis report"), basic patient information (name, age, gender, etc.), symptom description (fever and cough symptoms), medication recommendation details (drug name, dosage, frequency, duration, etc.), and medication precautions (such as paying attention to a light diet, and seeking medical treatment in time if gastrointestinal discomfort occurs). Next, add reference links related to the medication recommendation information to the medication analysis report, which point to the corresponding medical literature and drug instructions. For example, add a link to the "Amoxicillin" drug instructions to facilitate medical staff or patients to further review detailed information. Finally, the medication analysis report is pushed to the preset user interaction interface through the preset push method. Assuming that the preset push method is network transmission, the server will transmit the generated medication analysis report through the hospital's internal network to the user interface on the doctor's workstation computer, or to the user interface of the online medical platform that patients can access through mobile phones and other devices, so that medical staff and patients can view and understand medication-related situations in a timely manner.

[0024] In the embodiment of the present invention, the symptom description information is processed based on a preset large language model to obtain medication recommendation information of the symptom description information, which can be implemented through the following examples.

[0025] Obtaining symptom description information and the symptom category corresponding to the symptom description information;

[0026] Determine a drug database matching the symptom category from multiple drug databases; the drug database includes a low-dimensional vector representation corresponding to candidate drug information matching the symptom category;

[0027] According to the low-dimensional vector representation in the drug database, selecting a plurality of candidate drug information that matches the low-dimensional vector representation of the symptom description information from the candidate drug information corresponding to the low-dimensional vector representation in the drug database;

[0028] According to the medication-related features respectively corresponding to the plurality of candidate drug information, selecting, from the plurality of candidate drug information, candidate drug information that matches the medication-related features corresponding to the symptom description information as recommended drug information;

[0029] According to the recommended drug information and the symptom description information, a guide word corresponding to a large language model is obtained, and through the large language model, medication recommendation information of the symptom description information is obtained according to the guide word.

[0030] In the embodiment of the present invention, illustratively, after a series of processing of its medical record information, the server has generated Xiao Li's symptom description information: "Patient Xiao Li has a fever (body temperature 38.5 degrees) and is accompanied by coughing (more frequent), a slight sore throat, physical fatigue and other symptoms". At the same time, the server will determine the symptom category corresponding to the symptom description information according to the pre-set symptom classification rules. In the hospital's medical knowledge base, there is a detailed symptom classification system. For example, a series of symptoms such as Xiao Li's fever, cough, sore throat, physical fatigue, etc., are determined to belong to the "respiratory tract infection" symptom category after comparison and analysis between the server and the knowledge base. This step is crucial for the subsequent accurate screening of appropriate drug databases and obtaining accurate medication advice information, because different symptom categories often correspond to different therapeutic drugs and plans. The hospital's information system maintains multiple drug databases, which cover the commonly used drug information for various diseases, and in order to facilitate efficient data analysis and matching, the relevant information of each drug is stored in the form of low-dimensional vector representation. When the server determines that Xiao Li's symptoms belong to the "respiratory tract infection category", it begins to search and screen in many drug databases. For example, there is a drug database specifically for common internal medicine diseases, which stores a large amount of commonly used drug information for respiratory-related diseases such as colds, pneumonia, and bronchitis. The server quickly locates this drug database that matches the symptom category of "respiratory tract infection" through the database index and classification identification. In this specific drug database, each candidate drug information has its corresponding low-dimensional vector representation. Taking "amoxicillin", a commonly used drug for treating respiratory tract infections, as an example, its drug name, pharmacological effects, applicable symptoms, adverse reactions and many other information are converted into a low-dimensional vector representation through a specific vector generation algorithm. This low-dimensional vector representation is like a unique "fingerprint" of "amoxicillin" in this database, which can facilitate subsequent operations such as similarity calculation with the patient's symptom description information in order to find the most matching drug. After locating the drug database of "respiratory tract infection", the server will then filter out multiple candidate drug information that matches the low-dimensional vector representation of Xiao Li's symptom description information based on the low-dimensional vector representation. First, the server will also convert Xiao Li's symptom description information into a low-dimensional vector representation. For Xiao Li's description of "fever (body temperature 38.5 degrees) accompanied by coughing (more frequently), slight sore throat, physical fatigue and other symptoms", a series of text processing and vector generation algorithms are used to convert it into a low-dimensional vector representation that can be compared with the vectors in the drug database in the same dimension. Then, the server starts to calculate the cosine similarity between the low-dimensional vector representation of Xiao Li's symptom description information and the low-dimensional vector representation of each candidate drug information in the drug database.For example, for the low-dimensional vector representation of "amoxicillin", the server calculates the cosine similarity value between it and the low-dimensional vector representation of Xiao Li's symptom description information through a specific mathematical algorithm. Similarly, for other candidate drugs in the drug database, such as "azithromycin", "ibuprofen", "cough syrup", etc., the cosine similarity between them and the low-dimensional vector representation of Xiao Li's symptom description information is also calculated. Assume that after calculation, the cosine similarity value of "amoxicillin" is 0.65, the cosine similarity value of "azithromycin" is 0.58, the cosine similarity value of "ibuprofen" is 0.6, and the cosine similarity value of "cough syrup" is 0.55. The server will set a preset cosine similarity threshold, such as 0.5. Then, all candidate drug information whose cosine similarity exceeds this threshold will be screened out as multiple candidate drug information that matches the low-dimensional vector representation of Xiao Li's symptom description information. In this example, the cosine similarity of "amoxicillin", "azithromycin" and "ibuprofen" all exceeds 0.5, so the candidate drug information corresponding to these three drugs is screened out as a set of candidate drugs that preliminarily match Xiao Li's symptom description information. After the server screens out the preliminary matching candidate drug information of "amoxicillin", "azithromycin" and "ibuprofen", it will further screen them according to their corresponding medication-related features to determine the final recommended drug information. These medication-related features include multiple aspects, such as the pharmacological effects of each drug, the applicable population, the dosage range, the adverse reactions, the drug storage time node, etc. Taking the pharmacological effect as an example, "amoxicillin" mainly exerts its antibacterial effect by inhibiting the synthesis of bacterial cell walls, and has a good effect on some respiratory diseases caused by bacterial infections. Among Xiao Li's symptoms, fever may be caused by bacterial infection, so from the perspective of pharmacological effects, "amoxicillin" is compatible with Xiao Li's symptoms to a certain extent. The server will calculate the pharmacological adaptation coefficient between "Amoxicillin" and Xiao Li's symptom description information based on the pre-set pharmacological adaptation evaluation algorithm, assuming that this coefficient is 0.7. For "Azithromycin", its antibacterial spectrum is different from that of "Amoxicillin". Although it can also be used for respiratory tract infections, it may not be as targeted as "Amoxicillin" for Xiao Li's situation. After calculation, its pharmacological adaptation coefficient may be 0.5. "Ibuprofen" mainly has antipyretic, analgesic and anti-inflammatory effects. It can relieve Xiao Li's fever and physical fatigue symptoms, but from an antibacterial point of view, it is not as good as "Amoxicillin" to directly target the cause of the disease. Its pharmacological adaptation coefficient is assumed to be 0.6. Looking at the dosage range of medication, the doctor may be more inclined to choose a medication with a dosage range that is more suitable for Xiao Li based on Xiao Li's age, weight and other factors. Assume that Xiao Li's condition is suitable for a smaller dose of medication, and "Amoxicillin" has a variety of dosage specifications to choose from, and can meet Xiao Li's medication needs. In this regard, its adaptability is relatively high, and a medication dosage adaptation coefficient can be calculated through the corresponding algorithm, such as 0.8.As for adverse reactions, "Amoxicillin" may cause some mild gastrointestinal discomfort, "Azithromycin" may have some gastrointestinal reactions and other less common adverse reactions, and "Ibuprofen" may also have some effects on the stomach. The server will comprehensively consider the degree of impact of these adverse reactions on the patient and calculate the adaptation coefficient of each drug to Xiao Li's situation in terms of adverse reactions. Assume that the coefficient of "Amoxicillin" in this regard is 0.6, "Azithromycin" is 0.5, and "Ibuprofen" is 0.55. In addition, considering the entry time node of the drug, the newly entered drugs may have updated research results in terms of efficacy, safety, etc. Assume that "Amoxicillin" is the most recently entered drug, and its entry time node is close to the current execution time node, indicating that it may be more in line with the current medical standards and research progress. The server can determine the effectiveness coefficient corresponding to each drug based on the difference between the entry time node and the execution time node. Assume that the effectiveness coefficient of "Amoxicillin" is 0.8, "Azithromycin" is 0.6, and "Ibuprofen" is 0.7. Then, the server will comprehensively consider the various coefficients corresponding to the above medication-related features, and calculate a total fitness for each candidate drug information through a specific comprehensive evaluation algorithm. For example, for "Amoxicillin", the pharmacological fitness coefficient, dosage fitness coefficient, adverse reaction fitness coefficient, effectiveness coefficient, etc. are weighted and summed (assuming that the weights of each coefficient are the same), and a total fitness value of 0.725 is obtained. Similarly, such calculations are performed for "Azithromycin" and "Ibuprofen". Assuming that the total fitness value of "Azithromycin" is 0.525, the total fitness value of "Ibuprofen" is 0.625. Finally, the server will set a recommended fitness threshold, such as 0.6. Any candidate drug information whose total fitness exceeds this threshold will be selected as the recommended drug information that matches the medication-related features corresponding to Xiao Li's symptom description information. In this example, the total fitness value of "Amoxicillin" exceeds 0.6, so "Amoxicillin" is determined to be the drug information recommended to Xiao Li. When the server determines that the recommended drug information is "Amoxicillin", it begins to prepare to obtain the guide words corresponding to the large language model, so as to further obtain more detailed and accurate medication recommendation information through the large language model. First, the server obtains the guide word frame corresponding to the large language model. This guide word frame is pre-set, with a specific format and placeholders, which are used to fill in relevant information to generate guide words. For example, the guide word frame may have a storage location corresponding to the "symptom identifier" and a storage location corresponding to the "drug identifier". Then, the server stores Xiao Li's symptom description information in the storage location corresponding to the symptom identifier in the guide word frame. That is, the information that "Patient Xiao Li has a fever (body temperature 38.5 degrees) and is accompanied by coughing (more frequent), slight sore throat, physical fatigue and other symptoms" is accurately filled in the corresponding position.Next, the recommended drug information "Amoxicillin" is placed in the storage location corresponding to the drug identifier in the guide word framework. In this way, the server obtains the guide word corresponding to the large language model. For example, the guide word may be "For patient Xiao Li who has a fever (body temperature 38.5 degrees) and is accompanied by coughing (more frequent), a slight sore throat, and physical fatigue, amoxicillin is recommended. What is the medication recommendation?" Finally, the server inputs this guide word into the large language model. Based on its powerful language understanding and generation capabilities, the large language model will analyze the specific needs based on this guide word, and combine its internal medical knowledge and medication experience to generate medication recommendation information about Xiao Li's symptom description information. For example, the large language model may output "For patient Xiao Li, the medication recommendations for amoxicillin are as follows: the dosage is 0.5 grams each time, three times a day, and the expected duration of medication is five days. During the medication period, you should pay attention to a light diet and avoid drinking. If you experience adverse reactions such as gastrointestinal discomfort, you should seek medical attention in time." Such detailed medication recommendation information can provide important reference for doctors' diagnosis and treatment decisions and patients' medication.

[0031] In an embodiment of the present invention, according to the low-dimensional vector representation in the drug database, multiple candidate drug information matching the low-dimensional vector representation of the symptom description information is selected from the candidate drug information corresponding to the low-dimensional vector representation in the drug database, which can be implemented through the following examples.

[0032] Determining the cosine similarity between the low-dimensional vector representation of the symptom description information and the low-dimensional vector representation in the drug database;

[0033] Determining, from the drug database, a plurality of low-dimensional vector representations whose cosine similarity exceeds a preset cosine similarity threshold;

[0034] Determining, according to the feature conversion relationship between the low-dimensional vector representation and the candidate drug information, the candidate drug information having feature conversion relationships with the multiple low-dimensional vector representations respectively;

[0035] The candidate drug information having feature conversion relationships with the multiple low-dimensional vector representations respectively is used as multiple candidate drug information matched with the low-dimensional vector representation of the symptom description information.

[0036] In the embodiment of the present invention, for example, taking the patient Xiao Li mentioned above as an example, the server has obtained Xiao Li's detailed symptom description information: "Patient Xiao Li has a fever (body temperature 38.5 degrees) and is accompanied by coughing (more frequent), a slight sore throat, physical fatigue and other symptoms." In order to screen out suitable candidate drug information from the drug database, the server must first convert Xiao Li's symptom description information into a low-dimensional vector representation. This conversion process involves a series of complex text processing and vector generation algorithms. By analyzing and quantifying each word, phrase and the relationship between them in the symptom description, Xiao Li's symptom description information is finally converted into a low-dimensional vector representation of a specific dimension. At the same time, the drug database connected to the server stores a large number of candidate drug information for various diseases, and each candidate drug information has its corresponding low-dimensional vector representation. Taking "Amoxicillin", a commonly used drug for the treatment of respiratory infections, as an example, its related information in the drug database, such as drug name, pharmacological action, applicable symptoms, adverse reactions, etc., are all processed by a professional vector generation algorithm to form a unique low-dimensional vector representation. Next, the server starts to calculate the cosine similarity between the low-dimensional vector representation of Xiao Li's symptom description information and the low-dimensional vector representation of each candidate drug information in the drug database. For each candidate drug in the drug database, the server will use specific mathematical formulas and algorithms to perform this calculation. For example, for the low-dimensional vector representation of "amoxicillin", the server will compare the low-dimensional vector representation of Xiao Li's symptom description information with it, and determine the cosine similarity between them by calculating the cosine value of the angle between the two in the vector space. Assume that after precise calculation, the cosine similarity between the low-dimensional vector representation of Xiao Li's symptom description information and the low-dimensional vector representation of "amoxicillin" is 0.65. Similarly, the server will calculate the cosine similarity between the low-dimensional vector representation of Xiao Li's symptom description information and other candidate drugs in the drug database, such as "azithromycin", "ibuprofen", "cough syrup", etc., respectively. After a series of calculations, the cosine similarity of "azithromycin" is 0.58, the cosine similarity of "ibuprofen" is 0.6, the cosine similarity of "cough syrup" is 0.55, and so on. After completing the above cosine similarity calculation, the server will filter according to the preset cosine similarity threshold. In the hospital's medical data analysis system, this preset cosine similarity threshold is determined based on a large amount of clinical experience and data analysis, aiming to ensure that the screened candidate drug information has a high correlation with the patient's symptom description information. Assume that the cosine similarity threshold set by the hospital is 0.5. The server will compare the cosine similarity values ​​between each previously calculated candidate drug information and the low-dimensional vector representation of Xiao Li's symptom description information one by one. For "amoxicillin", the cosine similarity is 0.65, which is greater than the preset threshold of 0.5, so the low-dimensional vector representation corresponding to "amoxicillin" meets the screening conditions.Similarly, the cosine similarity of "azithromycin" is 0.58, which is also greater than 0.5, and its corresponding low-dimensional vector representation also meets the screening requirements. The cosine similarity of "ibuprofen" is 0.6, which also exceeds the preset threshold, and its corresponding low-dimensional vector representation is also within the screening range. The cosine similarity of "cough syrup" is 0.55. Although it is close to the threshold, it is still greater than 0.5, so its corresponding low-dimensional vector representation is also determined to meet the screening conditions. Through such a comparison and screening process, the server determines from the drug database that multiple low-dimensional vector representations whose cosine similarity exceeds the preset cosine similarity threshold. In this example, these are the low-dimensional vector representations corresponding to the drugs "amoxicillin", "azithromycin", "ibuprofen" and "cough syrup". After the server screens out those low-dimensional vector representations whose cosine similarity exceeds the preset threshold, it needs to restore the corresponding candidate drug information based on the feature conversion relationship between the low-dimensional vector representation and the candidate drug information. In the process of building the drug database, in order to facilitate data storage and processing, candidate drug information is converted into low-dimensional vector representations, and clear feature conversion relationships are established. Taking the low-dimensional vector representation corresponding to "amoxicillin" selected previously as an example, the server performs a series of parsing and restoration operations on the low-dimensional vector representation based on the pre-set feature conversion relationship. These operations may involve the interpretation of the values ​​of each dimension of the vector, the comparison with the stored standard vector template, etc. After these fine operations, the server can accurately restore the low-dimensional vector representation corresponding to "amoxicillin" to specific candidate drug information, that is, it is clear that this low-dimensional vector representation corresponds to the drug "amoxicillin", including its detailed drug information, such as drug name, pharmacological action, applicable symptoms, adverse reactions, etc. Similarly, for the low-dimensional vector representations selected such as "azithromycin", "ibuprofen" and "cough syrup", the server will also perform restoration operations according to their corresponding feature conversion relationships. By parsing and restoring their respective low-dimensional vector representations, the corresponding candidate drug information is determined to be "azithromycin", "ibuprofen" and "cough syrup", as well as their respective complete drug information. After the previous steps, the server has successfully restored the low-dimensional vector representations whose cosine similarity exceeds the preset threshold to specific candidate drug information. Now, the server will formally use these candidate drug information that have feature conversion relationships with multiple low-dimensional vector representations as multiple candidate drug information that matches the low-dimensional vector representation of Xiao Li's symptom description information. In this example, through the previous calculation and restoration operations, it is determined that the multiple candidate drug information that matches the low-dimensional vector representation of Xiao Li's symptom description information is "amoxicillin", "azithromycin", "ibuprofen" and "cough syrup".The information of these candidate drugs will serve as the basis for further analysis and screening, so as to ultimately determine the recommended drug information that is most suitable for patient Xiao Li, thereby providing a more accurate and scientific basis for doctors' diagnosis and treatment decisions and patients' medication.

[0037] In an embodiment of the present invention, the medication-related features corresponding to the candidate drug information include text description features and keywords of the candidate drug information, and the medication-related features corresponding to the symptom description information include text description features and keywords of the symptom description information;

[0038] The selecting, from the multiple candidate drug information, candidate drug information matching the medication related features corresponding to the symptom description information as recommended drug information based on the medication related features respectively corresponding to the multiple candidate drug information, can be implemented through the following examples.

[0039] Determining pharmacological compatibility coefficients between the symptom description information and the plurality of candidate drug information, respectively, according to the text description features of the symptom description information and the text description features respectively corresponding to the plurality of candidate drug information;

[0040] Determining keyword adaptation coefficients between the symptom description information and the plurality of candidate drug information, respectively, according to the keywords of the symptom description information and the keywords respectively corresponding to the plurality of candidate drug information;

[0041] According to the pharmacological adaptation coefficients and keyword adaptation coefficients respectively corresponding to the multiple candidate drug information, the candidate drug information matching the medication-related features corresponding to the symptom description information is selected from the multiple candidate drug information as the recommended drug information.

[0042] In the embodiment of the present invention, illustratively, taking the patient Xiao Li mentioned above as an example, his symptom description information is "Patient Xiao Li has a fever (body temperature 38.5 degrees) and is accompanied by coughing (more frequent), a slight sore throat, fatigue and other symptoms". The server has screened out multiple candidate drug information that matches the low-dimensional vector representation of Xiao Li's symptom description information, such as "Amoxicillin", "Azithromycin", "Ibuprofen", etc. Now the pharmacological fitness coefficient is determined based on their respective text description features. For "Amoxicillin", its text description features contain a lot of important information, such as it is a broad-spectrum semi-synthetic penicillin antibiotic, which mainly exerts its antibacterial effect by inhibiting the synthesis of bacterial cell walls, and has good antibacterial effects on a variety of Gram-positive and Gram-negative bacteria, especially suitable for respiratory tract infections and other diseases caused by bacterial infections. The server conducts a detailed comparison and analysis of Xiao Li's symptom description information and the text description features of "Amoxicillin". Xiao Li's fever symptoms may be caused by bacterial infection, and the antibacterial effect of "Amoxicillin" happens to target such causes. From this perspective, the two have a certain degree of pharmacological compatibility. In order to quantify this compatibility, the server calculates based on a pre-set pharmacological compatibility evaluation algorithm. This algorithm takes into account factors such as the drug's mechanism of action, the range of applicable diseases, and the possible causes of the patient's symptoms. For example, the algorithm may determine the pharmacological compatibility coefficient between Xiao Li's symptom description information and "Amoxicillin" after a series of complex calculation steps based on the targeting of "Amoxicillin" to bacterial infections, its applicability to respiratory tract infections, and the evaluation of the possibility of bacterial infection in Xiao Li's symptoms. Assume that after precise calculation, this pharmacological compatibility coefficient is 0.7. Looking at "Azithromycin", its text description features indicate that it is a macrolide antibiotic with a different antibacterial spectrum from "Amoxicillin". It is mainly targeted at some special types of bacterial infections. Although it can also be used for respiratory infections, it is relatively weak in targeting Xiao Li's fever, which may be caused by common bacterial infections. The server also compares and analyzes Xiao Li's symptom description information with the text description features of "azithromycin" according to the pharmacological adaptation evaluation algorithm. Taking into account the antibacterial characteristics of "azithromycin" and the specific situation of Xiao Li's symptoms, after calculation, it is concluded that the pharmacological adaptation coefficient between Xiao Li's symptom description information and "azithromycin" may be 0.5. For "ibuprofen", its text description features show that it mainly plays the role of antipyretic, analgesic, and anti-inflammatory, and can effectively relieve symptoms such as fever and physical fatigue, but from an antibacterial perspective, it cannot directly target the possible bacterial infection causes of Xiao Li. Based on the pharmacological adaptation evaluation algorithm, the server matches and analyzes Xiao Li's symptom description information with the text description features of "ibuprofen". After calculation, it is determined that the pharmacological adaptation coefficient between Xiao Li's symptom description information and "ibuprofen" is assumed to be 0.6.Through such a detailed comparison of the text description features of each candidate drug information and Xiao Li's symptom description information and calculation based on a specific algorithm, the server obtained the pharmacological adaptation coefficients between Xiao Li's symptom description information and candidate drug information such as "amoxicillin", "azithromycin" and "ibuprofen", laying the foundation for further screening and recommending drug information. In addition to determining the pharmacological adaptation coefficient based on the text description features, the server also analyzes the adaptation situation based on keywords to more comprehensively and accurately evaluate the matching degree between the candidate drug information and the symptom description information. The keywords in Xiao Li's symptom description information include "fever", "cough", "sore throat" and "fatigue". For "amoxicillin", the corresponding keywords may include "antibacterial", "anti-inflammatory", "bacterial infection" and "respiratory tract infection". The server first determines the semantic adaptation between the keywords of the symptom description information and the keywords corresponding to "amoxicillin". For example, "fever" may imply the presence of bacterial infection, and the "antibacterial" keyword of "amoxicillin" has a certain semantic association with it. Through a specific semantic analysis algorithm, the semantic adaptation between the two is calculated. Assume that after calculation, this semantic fitness is 0.6. Next, the server determines the first keyword fitness coefficient between the two based on the core medical terms in the keywords of the symptom description information (such as "fever" and "cough") and the core medical terms in the keywords corresponding to "amoxicillin" (such as "antibacterial"). By considering factors such as the relevance of these core medical terms in the medical knowledge system and their importance in the current situation, after calculation by a special algorithm, it is assumed that the first keyword fitness coefficient is 0.7. Then, the server determines the second keyword fitness coefficient between the two based on the non-core medical terms in the keywords of the symptom description information (such as "more frequent" and "slightly") and the non-core medical terms in the keywords corresponding to "amoxicillin" (there may be fewer corresponding non-core medical terms in the keywords of "amoxicillin", which are assumed to be null values ​​here). According to the corresponding calculation rules, it is assumed that the second keyword fitness coefficient is 0.3. Finally, the server performs a cumulative operation on the semantic fitness, the first keyword fitness coefficient, and the second keyword fitness coefficient corresponding to "Amoxicillin", that is, 0.6+0.7+0.3=1.6, and obtains the keyword fitness coefficient between Xiao Li's symptom description information and "Amoxicillin". Similarly, for "azithromycin", its keywords may include "antibacterial spectrum", "special bacterial infection", "macrolides", etc. According to the above steps, the server first determines the semantic fitness between "azithromycin" and Xiao Li's symptom description information, assuming it is 0.5; then calculates the first keyword fitness coefficient, assuming it is 0.6; finally calculates the second keyword fitness coefficient, assuming it is 0.2. Adding these three coefficients, 0.5+0.6+0.2=1.3, we get the keyword fitness coefficient between Xiao Li's symptom description information and "azithromycin". For "ibuprofen", its keywords may include "antipyretic", "analgesic", "anti-inflammatory", etc.The server performs calculations in sequence and determines that the semantic adaptation is assumed to be 0.6, the first keyword adaptation coefficient is assumed to be 0.5, and the second keyword adaptation coefficient is assumed to be 0.4. After accumulation, 0.6+0.5+0.4=1.5, and the keyword adaptation coefficient between Xiao Li's symptom description information and "ibuprofen" is obtained. Through such detailed analysis and calculation of the keywords of each candidate drug information and Xiao Li's symptom description information, the server obtains the keyword adaptation coefficients between Xiao Li's symptom description information and candidate drug information such as "amoxicillin", "azithromycin", and "ibuprofen", and further improves the evaluation of the matching degree between candidate drug information and symptom description information. After the server obtains the pharmacological adaptation coefficient and keyword adaptation coefficient between Xiao Li's symptom description information and candidate drug information such as "amoxicillin", "azithromycin", and "ibuprofen", it will combine these coefficients to select the final recommended drug information. For "amoxicillin", its pharmacological adaptation coefficient is 0.7 and the keyword adaptation coefficient is 1.6. The server will comprehensively consider these two coefficients based on the pre-set comprehensive evaluation algorithm. This comprehensive evaluation algorithm may be designed based on factors such as the hospital's diagnosis and treatment experience, the importance of the drug, and the relative weight of different coefficients in decision-making. Assume that in this comprehensive evaluation algorithm, the weight of the pharmacological fitness coefficient is 0.4 and the weight of the keyword fitness coefficient is 0.6. Then, the server calculates the comprehensive fitness of "amoxicillin" as: 0.7×0.4+1.6×0.6=1.24. Similarly, for "azithromycin", the pharmacological fitness coefficient is 0.5 and the keyword fitness coefficient is 1.3. According to the above comprehensive evaluation algorithm, assuming that the weight of the pharmacological fitness coefficient is still 0.4 and the weight of the keyword fitness coefficient is still 0.6, the comprehensive fitness is calculated as: 0.5×0.4+1.3×0.6=0.98. For "ibuprofen", the pharmacological fitness coefficient is 0.6 and the keyword fitness coefficient is 1.5. According to the comprehensive evaluation algorithm, its comprehensive fitness is calculated as: 0.6×0.6+1.5×0.6=1.26. The server will set a recommended fitness threshold, which is also determined based on the hospital's diagnosis and treatment experience, the safety and effectiveness of the drug, and other factors. Assume that the recommended fitness threshold is 1.2. By comparing the comprehensive fitness of each candidate drug information with the recommended fitness threshold, it is found that the comprehensive fitness of "amoxicillin" and "ibuprofen" (1.24 and 1.26) exceeds the recommended fitness threshold, while the comprehensive fitness of "azithromycin" (0.98) does not exceed the threshold. Therefore, the server will select "amoxicillin" and "ibuprofen" as the recommended drug information that matches the medication-related features corresponding to Xiao Li's symptom description information. These two drugs will be used as the main reference in the subsequent generation of medication analysis reports and other links, providing a more accurate and scientific basis for doctors' diagnosis and treatment decisions and patients' medication.In actual medical data processing, the server will continuously repeat similar analysis processes to accurately screen out the most suitable recommended drug information based on the symptom description information of different patients and the information of numerous candidate drugs, so as to improve the quality and effectiveness of medical services.

[0043] In an embodiment of the present invention, determining the keyword adaptation coefficients between the symptom description information and the multiple candidate drug information respectively based on the keywords of the symptom description information and the keywords respectively corresponding to the multiple candidate drug information can be implemented through the following examples.

[0044] Determining the semantic compatibility between the target candidate drug information and the symptom description information according to the keywords of the symptom description information and the keywords corresponding to the target candidate drug information;

[0045] Determining a first keyword adaptation coefficient between the target candidate drug information and the symptom description information based on the core medical terms in the keywords of the symptom description information and the core medical terms in the keywords corresponding to the target candidate drug information;

[0046] Determining a second keyword adaptation coefficient between the target candidate drug information and the symptom description information based on non-core medical terms in the keywords of the symptom description information and non-core medical terms in the keywords corresponding to the target candidate drug information;

[0047] An accumulation operation is performed on the semantic fitness corresponding to the target candidate drug information, the first keyword fitness coefficient and the second keyword fitness coefficient to obtain the keyword fitness coefficients between the symptom description information and the multiple candidate drug information respectively.

[0048] In the embodiment of the present invention, for example, taking the previously concerned patient Xiao Li as an example, the symptom description information is "Patient Xiao Li has a fever (body temperature 38.5 degrees) and is accompanied by coughing (more frequent), a slight sore throat, physical fatigue and other symptoms". After the preliminary processing, the server has screened out a series of candidate drug information, such as "amoxicillin", "azithromycin", "ibuprofen", etc. Now, "amoxicillin" is used as the target candidate drug information to explain this process in detail. First, the server will extract the keywords in Xiao Li's symptom description information, which accurately summarize Xiao Li's symptoms, namely "fever", "cough", "sore throat", "physical fatigue", etc. For "amoxicillin", as a commonly used antibacterial drug, its corresponding keywords may include "antibacterial", "anti-inflammatory", "bacterial infection", "respiratory tract infection", etc. after analysis and sorting. Next, the server will determine the semantic fit between "amoxicillin" and Xiao Li's symptom description information. This process involves complex semantic analysis technology, and the server will make judgments based on a large amount of medical field knowledge and a pre-set semantic analysis algorithm. For example, the keyword "fever" appears in Xiao Li's symptom description. Usually, fever may be caused by many reasons, but in many common respiratory infections, bacterial infection is an important factor causing fever. The keywords of "amoxicillin" include "antibacterial" and "bacterial infection", which shows that "amoxicillin" is semantically related to Xiao Li's "fever" symptom, because it may play a therapeutic role against the bacterial infection factors that cause fever. Similarly, "cough" is also a key symptom in Xiao Li's symptom description information, and the "respiratory infection" keyword of "amoxicillin" is also related to it, because cough is often one of the common symptoms of respiratory tract infection, and "amoxicillin" can be used to treat respiratory tract infection-related diseases, so there is also semantic adaptability. For symptoms such as "sore throat" and "fatigue", although there is no direct corresponding expression in the keywords of "amoxicillin", considering the series of symptoms that may be accompanied by respiratory tract infection and the overall therapeutic effect of "amoxicillin" on respiratory tract infection, it can also reflect the semantic association to a certain extent. By comprehensively considering the various semantic associations between these keywords and performing detailed calculations based on the semantic analysis algorithm, the server finally determined the semantic fit between "Amoxicillin" and Xiao Li's symptom description information. Assume that after precise calculation, the value of this semantic fit is 0.6. In the same way, the server will analyze and calculate the above semantic fit with Xiao Li's symptom description information for other candidate drug information, such as "Azithromycin" and "Ibuprofen". For "Azithromycin", it is assumed that after analysis and calculation, the semantic fit between it and Xiao Li's symptom description information is 0.5; for "Ibuprofen", it is assumed that its semantic fit is 0.6.After completing the determination of semantic adaptation, the server will then further analyze the adaptation situation based on the core medical terms in the symptom description information and the target candidate drug information. The core medical terms in Xiao Li's symptom description information are undoubtedly the most critical and essential descriptive words for his symptoms, mainly including "fever" and "cough". For "amoxicillin", the corresponding core medical term related keywords are "antibacterial" as mentioned above. The server will determine the first keyword adaptation coefficient between "amoxicillin" and Xiao Li's symptom description information based on the pre-set core medical term adaptation algorithm. This algorithm will deeply consider factors such as the importance of core medical terms in the medical knowledge system, their direct correlation, and their role in the current specific disease context. Take "fever" and "antibacterial" as an example. In the knowledge system of the medical field, fever may be caused by bacterial infection, and the "antibacterial" effect of "amoxicillin" is aimed at bacterial infection, so there is a close logical connection between the two core medical terms "fever" and "antibacterial". Similarly, there is a certain correlation between "cough" and "respiratory infection" (the keyword corresponding to "amoxicillin"), because cough is often one of the common symptoms of respiratory infection, and "amoxicillin" can be used to treat respiratory infection symptoms. By comprehensively considering the various associations between these core medical terms and performing detailed calculations based on the core medical term adaptation algorithm, the server finally determined the first keyword adaptation coefficient between "amoxicillin" and Xiao Li's symptom description information. Assume that after precise calculation, the value of this first keyword adaptation coefficient is 0.7. In the same way, the server will analyze and calculate the above first keyword adaptation coefficients for other candidate drug information, such as "azithromycin" and "ibuprofen", respectively, and Xiao Li's symptom description information. For "azithromycin", it is assumed that after analysis and calculation, the first keyword adaptation coefficient between it and Xiao Li's symptom description information is 0.6; for "ibuprofen", it is assumed that its first keyword adaptation coefficient is 0.5. In addition to core medical terms, there are some non-core medical terms in the symptom description information and target candidate drug information. Although these words do not have a direct and critical impact on the symptoms and drug effects like core medical terms, they also reflect the specific manifestations of the symptoms and the relevant characteristics of the drugs to a certain extent, so they also need to be analyzed to more comprehensively evaluate the adaptation situation. The non-core medical terms in Xiao Li's symptom description information include "more frequent" (used to describe the frequency of coughing), "slightly" (used to describe the degree of throat pain), etc. For "amoxicillin", there may be relatively few words in its corresponding keywords that directly correspond to these non-core medical terms. Here, it is assumed that the part of the "amoxicillin" keyword corresponding to the non-core medical terms is null (the actual situation may vary depending on the specific drug and symptom analysis). The server will determine the second keyword adaptation coefficient between "amoxicillin" and Xiao Li's symptom description information based on the pre-set non-core medical term adaptation algorithm.This algorithm considers factors such as the role of non-core medical terms in describing the specific manifestations of symptoms and drug-related characteristics, the indirect correlation between them, and the impact in the current specific symptom context. Since the part of the keyword "Amoxicillin" that corresponds to the non-core medical terms in Xiao Li's symptom description information is null, in this case, according to the non-core medical term adaptation algorithm, it is assumed that the second keyword adaptation coefficient between "Amoxicillin" and Xiao Li's symptom description information is 0.3. In the same way, the server will analyze and calculate the above-mentioned second keyword adaptation coefficient for other candidate drug information, such as "azithromycin" and "ibuprofen", and Xiao Li's symptom description information. For "azithromycin", it is assumed that some of its non-core medical terms have a certain correlation with the non-core medical terms in Xiao Li's symptom description information. After analysis and calculation, the second keyword adaptation coefficient between it and Xiao Li's symptom description information is 0.2; for "ibuprofen", it is assumed that some of its non-core medical terms have a certain correlation with the non-core medical terms in Xiao Li's symptom description information. After analysis and calculation, the second keyword adaptation coefficient between it and Xiao Li's symptom description information is 0.4. After the server determines the semantic adaptation degree (0.6), the first keyword adaptation coefficient (0.7) and the second keyword adaptation coefficient (0.3) between "amoxicillin" and Xiao Li's symptom description information, the server will then perform cumulative calculations on these three coefficients to obtain the keyword adaptation coefficient between Xiao Li's symptom description information and "amoxicillin". The specific calculation process is: 0.6+0.7+0.3=1.6, so the keyword adaptation coefficient between Xiao Li's symptom description information and "amoxicillin" is 1.6. Following the same method, the server will perform the above-mentioned cumulative operation on other candidate drug information, such as "azithromycin" and "ibuprofen", and Xiao Li's symptom description information in turn to obtain the keyword adaptation coefficient between them and Xiao Li's symptom description information. For "azithromycin", its semantic adaptation is 0.5, the first keyword adaptation coefficient is 0.6, and the second keyword adaptation coefficient is 0.2. The cumulative operation is: 0.5+0.6+0.2=1.3, so the keyword adaptation coefficient between Xiao Li's symptom description information and "azithromycin" is 1.3. For "ibuprofen", its semantic adaptation is 0.6, the first keyword adaptation coefficient is 0.5, and the second keyword adaptation coefficient is 0.4. The cumulative operation is: 0.6+0.5+0.4=1.5, so the keyword adaptation coefficient between Xiao Li's symptom description information and "ibuprofen" is 1.5.Through such detailed analysis and calculation of each candidate drug information and Xiao Li's symptom description information, the server obtained the keyword adaptation coefficients between Xiao Li's symptom description information and candidate drug information such as "amoxicillin", "azithromycin" and "ibuprofen". These coefficients, together with the pharmacological adaptation coefficient, will provide important basis for the subsequent selection of the most suitable recommended drug information for Xiao Li from a large number of candidate drug information, thereby providing more accurate support for doctors' diagnosis and treatment decisions and patients' medication.

[0049] In an embodiment of the present invention, the medication-related features corresponding to the candidate drug information further include a storage time node of the candidate drug information;

[0050] According to the pharmacological adaptation coefficients and keyword adaptation coefficients respectively corresponding to the multiple candidate drug information, candidate drug information matching the medication-related features corresponding to the symptom description information is selected from the multiple candidate drug information as recommended drug information, which can be implemented through the following examples.

[0051] Determine the effectiveness coefficients respectively corresponding to the plurality of candidate drug information according to the differences between the storage time nodes and the execution time nodes respectively corresponding to the plurality of candidate drug information;

[0052] Performing a cumulative operation on the effectiveness coefficient, the pharmacological adaptation coefficient, and the keyword adaptation coefficient corresponding to the target candidate drug information to obtain a first recommended adaptation degree between the target candidate drug information and the symptom description information;

[0053] The candidate drug information whose first recommended fitness degree exceeds the preset first fitness degree among the multiple candidate drug information is used as the recommended drug information matching the medication-related feature corresponding to the symptom description information.

[0054] In an embodiment of the present invention, illustratively, continuing to take patient Xiao Li as an example, the server has previously screened out a series of candidate drug information for Xiao Li's symptom description information (such as fever, cough, sore throat, physical fatigue, etc.), such as "amoxicillin", "azithromycin", "ibuprofen", etc. Now we need to consider the factor of the storage time node of these candidate drug information to further determine their effectiveness coefficients. The hospital's drug management database will record the storage time node of each candidate drug information in detail. This time node is accurate to a specific date and time, and it represents the time when the drug was first entered into the hospital drug management system. The execution time node is the time when the server currently performs analysis and processing to determine the recommended drug information, which is also accurate to a specific date and time. Taking "amoxicillin" as an example, assuming that its storage time node is 10:00 on May 10, 2023, and the current execution time node of the server is 14:00 on August 20, 2023. The server will first calculate the difference between the two time nodes, and calculate the number of days and hours between 10:00 on May 10, 2023 and 14:00 on August 20, 2023 through the date and time calculation algorithm. After calculation, it is found that the number of days between them is 102 days, and the details such as the number of hours are not repeated for the time being. Then, the server determines the effectiveness coefficient corresponding to "amoxicillin" according to the pre-set effectiveness coefficient calculation rules. This calculation rule is based on the hospital's long-term medical practice and understanding of drug research and development, updates, etc. Generally speaking, drugs with a relatively recent storage time may be more in line with current medical standards and research progress in terms of efficacy and safety, because with the continuous deepening of medical research, the relevant characteristics of the drug may be further optimized or new scopes of application may be discovered. Assuming that according to this calculation rule, for drugs with an interval of 100-150 days, the calculation method of its effectiveness coefficient is: effectiveness coefficient = 1-(interval days-100) / 50. Then for "Amoxicillin", the interval days are 102 days, and the formula is substituted to obtain: effectiveness coefficient = 1-(102-100) / 50=1-2 / 50=0.96. Similarly, for "Azithromycin", assuming that its entry time node is 8:00 on March 15, 2023, the interval days between it and the current execution time node is 160 days after calculation. According to the above calculation rules, the effectiveness coefficient is 1-(160-100) / 50=1-60 / 50=-0.2 (the negative numbers here are just example calculation results. In actual situations, negative numbers may be specially handled, such as set to the minimum value of 0, etc., to ensure that the coefficient is within a reasonable range). Let's look at "Ibuprofen" again. Assuming that its entry time node is 12:00 on June 5, 2023, the calculated interval days between it and the execution time node are 76 days.Substituting into the calculation rules, we can get the effectiveness coefficient = 1-(76-100) / 50=1+24 / 50=1.48 (Similarly, there may be a situation where it exceeds 1, and reasonable processing will be carried out in practice, such as limiting it to 1, etc.). Through such a calculation process, the server determines the effectiveness coefficients corresponding to the candidate drug information such as "amoxicillin", "azithromycin" and "ibuprofen", which will serve as one of the important bases for subsequent comprehensive evaluation. After the server determines the effectiveness coefficients of each candidate drug information, it will then add up the three coefficients corresponding to the target candidate drug information to obtain the first recommended fit between the target candidate drug information and Xiao Li's symptom description information. First, "amoxicillin" is used as the target candidate drug information to explain in detail. It has been calculated that the effectiveness coefficient of "amoxicillin" is 0.96, and its pharmacological adaptation coefficient is assumed to be 0.7 (previously calculated based on the pharmacological effects of the drug and the fit of Xiao Li's symptoms), and the keyword adaptation coefficient is assumed to be 1.6 (obtained through analysis and calculation of keywords). The server performs cumulative calculations on these three coefficients according to the set calculation process: the first recommended fitness = effectiveness coefficient + pharmacological fitness coefficient + keyword fitness coefficient = 0.96 + 0.7 + 1.6 = 3.26 (further processing of the results is not considered here for the time being, and the results are obtained by normal calculation first). Similarly, for "azithromycin", its effectiveness coefficient is assumed to be 0 after processing (as mentioned above, the negative number can be set to the lowest value 0), the pharmacological fitness coefficient is 0.5, and the keyword fitness coefficient is 1.3. Cumulative calculations yield: the first recommended fitness = 0 + 0.5 + 1.3 = 1.8. For "ibuprofen", its effectiveness coefficient is assumed to be 1 after processing (as mentioned above, the case of exceeding 1 can be limited to 1), the pharmacological fitness coefficient is 0.6, and the keyword fitness coefficient is 1.5. Cumulative calculations yield: the first recommended fitness = 1 + 0.6 + 1.5 = 3.1. By performing such cumulative operations on each target candidate drug information, the server obtains the first recommended fitness between candidate drug information such as "Amoxicillin", "Azithromycin", "Ibuprofen" and Xiao Li's symptom description information, and these results will be used for subsequent screening and comparison of recommended drug information. In the hospital's information management system, in order to ensure that the drug information recommended to patients not only meets the needs of the disease but also takes into account multiple factors such as the effectiveness of the drug, the server will pre-set a first fitness preset value. This preset value is determined based on the hospital's clinical experience, drug treatment effect evaluation, and comprehensive consideration of patient safety medication. Assume that the first fitness preset value set by the hospital is 2.5. The server will compare the first recommended fitness between each candidate drug information calculated previously and Xiao Li's symptom description information one by one. For "Amoxicillin", its first recommended fitness is 3.26, which is greater than the preset value of 2.5, so "Amoxicillin" meets the conditions for being a recommended drug information.For "azithromycin", its first recommended fitness is 1.8, which is less than the preset value of 2.5, so "azithromycin" does not meet the conditions for being recommended drug information. For "ibuprofen", its first recommended fitness is 3.1, which is greater than the preset value of 2.5, so "ibuprofen" also meets the conditions for being recommended drug information. Through such a comparison and screening process, the server uses "amoxicillin" and "ibuprofen", two candidate drug information whose first recommended fitness exceeds the preset first fitness, as recommended drug information that matches the medication-related features corresponding to Xiao Li's symptom description information. These recommended drug information will be used as the main reference in the subsequent generation of medication analysis reports and other links, providing a more accurate and scientific basis for doctors' diagnosis and treatment decisions and patients' medication. At the same time, when faced with symptom description information of different patients and numerous candidate drug information, the server will continue to repeat similar analysis processes to ensure that the most suitable recommended drug information for patients can be accurately screened out every time, thereby improving the quality and effectiveness of medical services.

[0055] In the embodiment of the present invention, the selection of the drug suitable for the symptom from the plurality of candidate drug information according to the pharmacological fitness coefficients and the keyword fitness coefficients respectively corresponding to the plurality of candidate drug information can be implemented through the following examples.

[0056] Performing a cumulative operation on the pharmacological fitness coefficient and the keyword fitness coefficient corresponding to the target candidate drug information to obtain a second recommended fitness between the target candidate drug information and the symptom description information;

[0057] The candidate drug information whose second recommended fitness degree exceeds the preset second fitness degree among the multiple candidate drug information is used as the recommended drug information matching the medication-related feature corresponding to the symptom description information.

[0058] In the embodiment of the present invention, illustratively, continue to take patient Xiao Li as an example, and his symptom description information is "Patient Xiao Li has a fever (body temperature 38.5 degrees) and is accompanied by coughing (more frequent), a slight sore throat, physical fatigue and other symptoms". Previously, the server has screened out a number of candidate drug information through a series of complex processing flows, such as "amoxicillin", "azithromycin", "ibuprofen", etc. Now, the server will calculate the second recommended fitness between each target candidate drug information and Xiao Li's symptom description information based on the pharmacological fitness coefficients and keyword fitness coefficients corresponding to these candidate drug information. First, "amoxicillin" is used as the target candidate drug information for explanation. Previously, the server has calculated its pharmacological fitness coefficient based on the pharmacological properties of "amoxicillin" and its adaptation to Xiao Li's symptom description information. Assume that after detailed analysis and calculation of a specific algorithm, the pharmacological fitness coefficient of "amoxicillin" is 0.7. This coefficient takes into account the pharmacological adaptability of "amoxicillin" as a broad-spectrum semi-synthetic penicillin antibiotic, which mainly exerts its antibacterial effect by inhibiting the synthesis of bacterial cell walls, to Xiao Li's fever and other symptoms that may be caused by bacterial infection. At the same time, the server has also determined its keyword adaptation coefficient based on the keyword analysis of "amoxicillin" and Xiao Li's symptom description information. Assume that after a series of operations such as semantic analysis of keywords, matching of core and non-core medical terms, the keyword adaptation coefficient of "amoxicillin" is 1.6. Next, the server performs cumulative operations on the pharmacological adaptation coefficient and keyword adaptation coefficient corresponding to "amoxicillin" according to the established calculation rules to obtain the second recommended adaptation degree between "amoxicillin" and Xiao Li's symptom description information. The specific calculation process is as follows: Second recommended adaptation degree = pharmacological adaptation coefficient + keyword adaptation coefficient = 0.7 + 1.6 = 2.3. Similarly, for "azithromycin", the server first determines its pharmacological adaptation coefficient. "Azithromycin" is a macrolide antibiotic, and its antibacterial spectrum is different from that of "Amoxicillin". For Xiao Li's symptoms, its pharmacological fitness coefficient is 0.5 after analysis and calculation. According to the keyword analysis of "Azithromycin" and Xiao Li's symptom description information, its keyword fitness coefficient is determined to be 1.3. Then perform cumulative calculation: the second recommended fitness = pharmacological fitness coefficient + keyword fitness coefficient = 0.5 + 1.3 = 1.8. For "ibuprofen", its pharmacological fitness coefficient is based on its main antipyretic, analgesic and anti-inflammatory effects, which can relieve Xiao Li's fever, physical fatigue and other symptoms, but from an antibacterial perspective, it is not as good as "Amoxicillin" in directly targeting the cause of the disease. After calculation, it is assumed that its pharmacological fitness coefficient is 0.6. Through keyword analysis, its keyword fitness coefficient is determined to be 1.5. Then perform cumulative calculation: the second recommended fitness = pharmacological fitness coefficient + keyword fitness coefficient = 0.6 + 1.5 = 2.1.By performing such cumulative operations on each target candidate drug information, the server obtains the second recommended fitness between candidate drug information such as "Amoxicillin", "Azithromycin", "Ibuprofen" and Xiao Li's symptom description information, and these results will serve as an important basis for subsequent screening of recommended drug information. In the hospital's information management system, in order to ensure that the drug information finally recommended to the patient is the most suitable for the patient's symptoms and medication-related characteristics, the server will pre-set a second fitness preset value. This preset value is determined based on the hospital's rich clinical experience, long-term observation of the therapeutic effects of various types of drugs, and comprehensive consideration of the safety and effectiveness of patient medication. Assume that the second fitness preset value set by the hospital is 2.0. The server will compare the second recommended fitness between each candidate drug information calculated previously and Xiao Li's symptom description information one by one. For "Amoxicillin", its second recommended fitness is 2.3, which is greater than the preset value of 2.0, so "Amoxicillin" meets the conditions for being a recommended drug information. For "azithromycin", its second recommended fitness is 1.8, which is less than the preset value of 2.0, so "azithromycin" does not meet the conditions for being recommended drug information. For "ibuprofen", its second recommended fitness is 2.1, which is greater than the preset value of 2.0, so "ibuprofen" also meets the conditions for being recommended drug information. Through such a comparison and screening process, the server uses "amoxicillin" and "ibuprofen", two candidate drug information whose second recommended fitness exceeds the preset second fitness, as recommended drug information that matches the medication-related features corresponding to Xiao Li's symptom description information. These recommended drug information will be used as the main reference in the subsequent generation of medication analysis reports and other links, providing a more accurate and scientific basis for doctors' diagnosis and treatment decisions and patients' medication. At the same time, when faced with symptom description information of different patients and numerous candidate drug information, the server will continue to repeat similar analysis processes to ensure that the most suitable recommended drug information for patients can be accurately screened out every time, thereby improving the quality and effectiveness of medical services. For example, when generating a medication analysis report later, for patient Xiao Li, the server will list in detail key information such as drug name, dosage, frequency of use, expected duration of use, and adverse reactions based on the relevant information of "Amoxicillin" and "Ibuprofen". In addition, reference links related to these drugs will be added to the medication analysis report, pointing to the corresponding medical literature and drug instructions, so that doctors and patients can further understand the details of the drugs. Then, through a preset push method, such as network transmission or message queue push, the medication analysis report is pushed to the preset user interaction interface for doctors to view during the diagnosis and treatment process and for patients to refer to when understanding their own medication situation. In actual medical scenarios, different patients have different symptoms and a variety of candidate drug information. Through such a rigorous and scientific analysis process, the server continuously and accurately screens out appropriate recommended drug information for each patient, ensuring the rationality and effectiveness of medical medication and promoting the high-quality development of medical services.

[0059] In the embodiments of the present invention, the following implementation modes are also provided.

[0060] Acquire medical record data to be processed, and a disease association type corresponding to the medical record data; the disease association type corresponds to a symptom category;

[0061] Extracting and identifying the medical record data according to the disease association type to obtain multiple candidate drug information, and obtaining low-dimensional vector representations corresponding to the multiple candidate drug information respectively;

[0062] The low-dimensional vector representations corresponding to the plurality of candidate drug information are stored in a drug database corresponding to the disease association type.

[0063] In the embodiment of the present invention, for example, taking the cardiology department as an example, a patient named Mr. Zhang comes for treatment. The doctor conducts a detailed interview, physical examination and a series of related auxiliary examinations on Mr. Zhang, such as electrocardiogram, cardiac ultrasound, etc. In this process, the doctor enters all relevant information of Mr. Zhang into the electronic medical record system to form a complete medical record data. This medical record data contains Mr. Zhang's basic situation, such as age, gender, past medical history, etc., as well as his symptoms during this visit, such as palpitations, chest tightness, occasional chest pain, etc., and the specific results of various examinations. At the same time, the server will determine the symptom association type corresponding to this medical record data according to the pre-set symptom classification rules. For Mr. Zhang's situation, after analyzing his palpitations, chest tightness, chest pain and other symptoms, the server determines that his symptom association type is "cardiovascular system disease category", and this symptom association type corresponds to "palpitations, chest tightness, chest pain and other related symptom categories". In this way, the server successfully obtains the medical record data of Mr. Zhang to be processed, as well as its corresponding symptom association type, and is ready for further analysis and processing. After the server has determined that the disease association type corresponding to Mr. Zhang's medical record data is "cardiovascular system disease", it begins to extract and identify the medical record data in a targeted manner based on this disease association type. First, the server will determine the key medical terms and phrases related to it based on the disease association type of "cardiovascular system disease". For this type of disease, possible key medical terms include "palpitation", "chest tightness", "chest pain", "myocardial infarction", "arrhythmia", "coronary heart disease", etc., as well as some related phrases such as "insufficient blood supply to the heart" and "myocardial ischemia". Then, the server carefully searches for text fragments containing these key medical terms and phrases in Mr. Zhang's medical record data. For example, in the doctor's description of the symptoms, the sentence "the patient reported palpitation, chest tightness aggravated after activity, and occasionally accompanied by chest pain" will be searched because it contains the key medical terms "palpitation", "chest tightness", and "chest pain". Next, the searched text fragments are cleaned. This may include removing some irrelevant punctuation marks, extra spaces, etc., to make the text more standardized and easier to process. For example, "The patient reported palpitations, chest tightness aggravated after activities, and occasionally accompanied by chest pain" is cleaned up to "The patient reported palpitations, chest tightness aggravated after activities, and occasionally accompanied by chest pain". After that, the cleaned text fragments are processed by word segmentation to obtain multiple vocabulary units. For example, after word segmentation, the above cleaned text may obtain vocabulary units such as "patient", "self-reported", "palpitations", "after activities", "chest tightness", "aggravated", "occasionally", "accompanied by", and "chest pain". These vocabulary units are then generated according to the preset vector generation algorithm to generate low-dimensional vectors corresponding to these vocabulary units. This vector generation algorithm is trained based on a large amount of medical text data. It can accurately convert vocabulary units into low-dimensional vectors based on their semantics and associations in the medical field.For example, after the word unit "heart palpitation" is processed by the vector generation algorithm, a specific low-dimensional vector representation will be obtained. Finally, the generated low-dimensional vectors are combined and optimized to obtain low-dimensional vector representations corresponding to multiple candidate drug information. Assume that in Mr. Zhang's medical record data, through the previous processing, some potential drug mentions related to the treatment of cardiovascular diseases, such as "aspirin", "nitroglycerin", "betaloc", etc., are found. Then, for "aspirin", after a series of previous processing of related text fragments, a low-dimensional vector representation that can represent "aspirin" in the context of this medical record data is finally obtained. Similarly, for drugs such as "nitroglycerin" and "betaloc", their corresponding low-dimensional vector representations are also obtained. After the server obtains the low-dimensional vector representations corresponding to multiple candidate drug information such as "aspirin", "nitroglycerin", "betaloc", etc., it will store these low-dimensional vector representations in the drug database corresponding to the disease association type of "cardiovascular disease class". This drug database is specially established to store drug information related to various disease association types and their low-dimensional vector representations. In the database, the data has been classified and stored according to the disease association type, so that it can be matched and queried quickly and accurately when analyzing patient symptoms and finding suitable drugs. For the low-dimensional vector representation of "aspirin", the server will accurately store it in the corresponding position of the drug database corresponding to the disease association type of "cardiovascular system disease", and put it together with other low-dimensional vector representations of drugs related to the treatment of cardiovascular system diseases that have been stored. Similarly, for the low-dimensional vector representations of drugs such as "nitroglycerin" and "betaloc", they will also be stored in the appropriate position of the drug database according to the prescribed storage method. In this way, when a patient subsequently develops symptoms similar to cardiovascular system diseases, the server can quickly obtain the low-dimensional vector representation of the relevant drugs from the drug database corresponding to this disease association type when performing drug matching analysis, and then provide patients with more efficient services such as accurate medication recommendations. Through such a process, the server continuously enriches and improves the drug database corresponding to each disease association type, improves the efficiency and accuracy of medical data processing, and plays an important supporting role in improving the quality of medical services in hospitals.

[0064] In the embodiment of the present invention, the medical record data is extracted and identified according to the disease association type to obtain multiple candidate drug information, and low-dimensional vector representations corresponding to the multiple candidate drug information are obtained, which can be implemented through the following examples.

[0065] According to the type of association of the disease, determine the key medical words and phrases related to it;

[0066] searching the medical record data for text segments containing the key medical words and phrases;

[0067] Performing text cleaning on the searched text fragments, and performing word segmentation on the cleaned text fragments to obtain multiple vocabulary units;

[0068] Generate low-dimensional vectors corresponding to the plurality of vocabulary units according to a preset vector generation algorithm;

[0069] The generated low-dimensional vectors are combined and optimized to obtain low-dimensional vector representations corresponding to the multiple candidate drug information.

[0070] In the embodiment of the present invention, for example, taking the medical record data of Mr. Zhang, a cardiology patient, as an example, the server has determined that the disease association type is "cardiovascular system disease". Based on this disease association type, the server needs to determine the key medical terms and phrases closely related to it. For cardiovascular system disease symptoms, the server sorts out a series of key medical terms and phrases based on the pre-built medical knowledge graph and a large amount of clinical data accumulation. For example, words such as "palpitation", "chest tightness", "chest pain", "palpitations", "myocardial infarction", "arrhythmia", "coronary heart disease", "cardiac insufficiency", "myocardial ischemia", "hypertension", "hyperlipidemia", and phrases such as "coronary atherosclerosis", "heart valve disease", and "myocardial hypertrophy". These key medical terms and phrases are jointly screened and confirmed by professional medical personnel and data analysts. They are of great significance in describing the symptoms, causes, pathophysiological processes, etc. of cardiovascular system diseases, and have potential associations with the therapeutic drugs that may be used in the future, which is an important basis for further extracting candidate drug information. After the server identified the key medical terms and phrases related to the "cardiovascular system disease" symptom association type, it began to conduct a comprehensive search in Mr. Zhang's medical record data to find text fragments containing these key elements. Mr. Zhang's medical record data is rich and diverse, covering basic information, symptom descriptions, test results, diagnosis and treatment processes, and other aspects. The server will scan and search these data line by line and paragraph by paragraph. For example, in the symptom description section, the doctor recorded that "the patient reported palpitations, chest tightness worsened after activities, occasionally accompanied by chest pain, and palpitations have been felt more frequently recently." This text contains multiple key medical terms such as "palpitations," "chest tightness," "chest pain," and "palpitations," so it will be accurately searched by the server as a qualified text fragment. For another example, in the examination results section, it is mentioned that "the electrocardiogram shows arrhythmia and suspected myocardial ischemia." Here, "arrhythmia" and "myocardial ischemia" are also key medical terms, so this part of the text will also be selected. Even in the records of the diagnosis and treatment process, if the doctor writes "considering that the patient may have coronary heart disease, the corresponding drug treatment has been given, and the current blood pressure control situation remains to be observed", the words "coronary heart disease" and "blood pressure" also make this part of the text a text fragment containing key medical words and phrases. Through such a meticulous search process, the server can filter out many text fragments related to cardiovascular diseases from Mr. Zhang's entire medical record data, providing rich materials for subsequent processing. After searching for text fragments containing key medical words and phrases, the server will further process these text fragments to make them easier to analyze and generate vector representations. First, text cleaning is performed. In actual medical record data, text fragments may contain some unnecessary punctuation, extra spaces, irregular capitalization, etc.For example, the searched text segment "The patient reported palpitations, chest tightness aggravated after activities, and occasionally accompanied by chest pain." The commas and periods here may interfere with subsequent processing, and the use of spaces is not standardized. The server will remove these redundant punctuation marks according to the preset text cleaning rules, normalize the spaces, and clean the text segment into "The patient reported palpitations, chest tightness aggravated after activities, and occasionally accompanied by chest pain." After completing the text cleaning, the server immediately performs word segmentation on the cleaned text segment. Word segmentation is to split the continuous text into independent vocabulary units according to certain rules. For the cleaned text "The patient reported palpitations, chest tightness aggravated after activities, and occasionally accompanied by chest pain", the server will segment it into multiple vocabulary units such as "patient", "self-reported", "palpitations", "after activities", "chest tightness", "aggravated", "occasionally", "accompanied by", "chest pain" based on the pre-trained word segmentation model. In this way, through text cleaning and word segmentation, the server converts the originally complex and less standardized text segments into clear vocabulary units, laying the foundation for the subsequent generation of vector representations. After the server obtains multiple vocabulary units that have been cleaned and segmented, it will convert these vocabulary units into low-dimensional vectors according to the preset vector generation algorithm. This vector generation algorithm is trained based on a large amount of medical text data. It can accurately capture the semantic information of each vocabulary unit in the medical field and convert it into a suitable low-dimensional vector form. Taking the vocabulary unit "heart palpitation" as an example, the vector generation algorithm will comprehensively consider the meaning of "heart palpitation" in the context of cardiovascular system diseases, the association with other symptom words, and the frequency of occurrence in different medical records. Through a series of complex mathematical calculations and model processing, "heart palpitation" is converted into a low-dimensional vector of a specific dimension, such as a three-dimensional vector [0.2, 0.3, 0.5] (this is just an example, the actual dimension and value will depend on the specific algorithm). Similarly, for other vocabulary units such as "chest tightness", "chest pain", "patient", "self-report", etc., the vector generation algorithm will also convert them into corresponding low-dimensional vectors according to their respective semantic characteristics and positions in the medical field, which may be vector forms of different dimensions and different values. In this way, the server converts each vocabulary unit into a low-dimensional vector with a specific semantic representation, so that these vocabulary units can be subsequently combined and analyzed in the vector space. After the server converts each vocabulary unit into a low-dimensional vector, it will then combine and optimize these low-dimensional vectors to obtain low-dimensional vector representations corresponding to multiple candidate drug information. In the previous processing of Mr. Zhang’s medical record data, through a series of operations on text fragments containing key medical vocabulary and phrases, low-dimensional vectors corresponding to vocabulary units such as "palpitation", "chest tightness", and "chest pain" have been obtained. Suppose some potential therapeutic drugs are also mentioned in these text fragments, such as "aspirin", "nitroglycerin", "Betaloc", etc.For "aspirin", the server will comprehensively consider the low-dimensional vectors of vocabulary units related to "aspirin", such as symptom vocabulary units related to the treatment of cardiovascular diseases (such as "palpitation", "chest tightness", "chest pain", etc.) and low-dimensional vectors of vocabulary units describing drug effects, scope of application, etc. (which may be mentioned in other parts of the medical record data). The server will reasonably combine and adjust these low-dimensional vectors related to "aspirin" through specific combination and optimization algorithms to form a low-dimensional vector representation that can fully represent "aspirin" in the context of this medical record data. Similarly, for drugs such as "nitroglycerin" and "betaloc", the server will also combine and optimize the low-dimensional vectors of their respective related vocabulary units to obtain their respective corresponding low-dimensional vector representations. Through such combination and optimization processing, the server can accurately extract the low-dimensional vector representations corresponding to multiple candidate drug information from the medical record data, and these low-dimensional vector representations will be stored in the corresponding drug database, so that they can quickly and accurately perform matching queries when analyzing patient symptoms and finding suitable drugs in the future, thereby providing patients with more accurate medication recommendations and diagnosis and treatment services.

[0071] In the embodiment of the present invention, obtaining the guide words corresponding to the large language model based on the recommended drug information and the symptom description information can be implemented through the following examples.

[0072] Get the guide word framework corresponding to the large language model;

[0073] Storing the symptom description information in a storage location corresponding to the symptom identifier in the guide word framework;

[0074] The recommended drug information is placed in a storage location corresponding to the drug identification in the guide word framework to obtain the guide word corresponding to the large language model.

[0075] In an embodiment of the present invention, exemplarily, in an advanced information management system of a hospital, the server, as a core data processing unit, works in collaboration with a variety of intelligent tools, including a large language model, to achieve more accurate medical data analysis and decision support. In order to enable the large language model to give accurate medication advice information according to the specific patient situation, it is first necessary to obtain the guide word framework corresponding to the large language model. This guide word framework is carefully designed and configured in advance, and it has a specific structure and format, which is intended to prepare for the subsequent filling of relevant information to generate effective guide words. For example, in a hospital's medical intelligent assistance system, this guide word framework may be a template-like text structure, which is stored in a specific database or storage area accessible to the server. The guide word framework is roughly as follows: "For patients with [symptom identification], it is recommended to use [drug identification] for treatment. What is the [specific inquiry point] about this treatment plan?" In this framework, "symptom identification", "drug identification" and "specific inquiry point" are all reserved placeholders for subsequent filling of specific content, thereby forming a complete guide word, so as to accurately convey the questions that need to be analyzed and answered to the large language model. Through the pre-set program and data access path, the server can quickly and accurately obtain the guide word framework corresponding to this large language model from the storage location, laying the foundation for the next steps. After the server obtains the guide word framework, it will then accurately fill the specific patient symptom description information into the corresponding position in the framework. Taking the patient Xiao Li mentioned earlier as an example, his symptom description information is "Patient Xiao Li has a fever (body temperature 38.5 degrees) and is accompanied by coughing (more frequent), a slight sore throat, physical fatigue and other symptoms." The server will identify the storage location corresponding to the "symptom identifier" in the guide word framework, and then fill Xiao Li's symptom description information completely into the location. During the specific operation, the server will adapt the format and content of the symptom description information to ensure that it can be perfectly integrated with the guide word framework. For example, some unnecessary punctuation marks may be removed or some expressions may be fine-tuned to make it read more smoothly and naturally after filling in the framework. After processing, the part corresponding to the "symptom identifier" in the guide word frame will be updated to: "Patient Xiao Li has a fever (body temperature 38.5 degrees) accompanied by coughing (more frequent), slight sore throat, physical fatigue and other symptoms." In this way, by filling accurate and detailed symptom description information into the corresponding position in the guide word frame, the large language model is provided with clear information about the patient's specific condition so that it can conduct targeted medication recommendation analysis based on these symptoms. After completing the filling of the symptom description information, the server must also fill the previously determined recommended drug information into the storage location corresponding to the "drug identifier" in the guide word frame. Suppose that after a series of rigorous analysis and screening processes, it is determined that the drug information recommended to patient Xiao Li is "Amoxicillin."The server will find the storage location corresponding to the "drug identification" in the guide word framework, and then accurately place "amoxicillin" at that location. At this point, the guide word framework will be updated to: "For patients with fever (body temperature 38.5 degrees) accompanied by coughing (more frequent), slight sore throat, and fatigue, amoxicillin is recommended for treatment. What are the [specific inquiry points] about this treatment plan?" The "specific inquiry points" here may be further refined according to actual needs, such as "drug dosage, frequency of use, expected duration of use, and possible adverse reactions". By accurately filling in the recommended drug information and symptom description information into the corresponding positions in the guide word framework, the server successfully obtains the guide words corresponding to the large language model. For example, a complete guide word might be: "For the patient Xiao Li who has a fever (body temperature 38.5 degrees) accompanied by coughing (more frequent), a slight sore throat, and fatigue, amoxicillin is recommended for treatment. What are the dosage, frequency, expected duration, and possible adverse reactions of this treatment plan?" This guide word can accurately convey to the large language model the specific situation that needs to be analyzed, that is, the specific symptoms of Xiao Li and the details of the medication when using amoxicillin for treatment. After receiving this guide word, the large language model can give a detailed answer to the medication recommendation information for Xiao Li based on its powerful language understanding and generation capabilities, combined with its own extensive medical knowledge and medication experience, thereby providing an important reference for doctors' diagnosis and treatment decisions and patients' medication.

[0076] In the embodiment of the present invention, the medical record information is subjected to denoising and standardization processing respectively to obtain pre-processed medical record information, which can be implemented through the following examples.

[0077] Performing denoising on the medical record information, wherein the denoising adopts a text filtering algorithm to remove duplicate information, irrelevant formatting marks, and erroneous statements that do not comply with medical record specifications in the medical record information;

[0078] The denoised medical record information is standardized, and the standardization includes unifying the date format into a preset standard format, restoring the abbreviations of medical terms into complete forms, and converting measurement data under different unit systems into a unified standard unit system to obtain the pre-processed medical record information.

[0079] In an embodiment of the present invention, illustratively, in the hospital's information management system, the server receives the medical record information of the patient Xiao Wang. This record contains Xiao Wang's multiple visits to the doctor, and there is a lot of repeated information, such as the same temperature value recorded by the nurse at different times. Through the text filtering algorithm, the server can accurately identify and remove these repeated contents. At the same time, there are some irrelevant formatting marks in the medical record, such as extra spaces and special characters that appear for typesetting, which will also be cleared by the server according to the algorithm. In addition, when the doctor enters, there may be erroneous statements that do not conform to the medical record specifications, such as "abdominal pain" is mistakenly written as "abdominal pain". The server corrects such erroneous statements by comparing with the standard medical terminology library to complete the denoising process. The denoised Xiao Wang medical record information needs to be further standardized. The server will unify the dates in different formats, such as "2023.5.10" and "10 / 05 / 2023", into the preset standard format "YYYY-MM-DD". For medical term abbreviations, such as "WBC", it will be restored to the full form "White Blood Cell". If the records contain measurement data in different units, such as blood pressure, some are in "mmHg" and some are in "kPa", the server will convert them all into a unified standard unit, such as "mmHg", and finally obtain the pre-processed medical record information for subsequent accurate analysis.

[0080] In an embodiment of the present invention, calling a preset NLP model to parse the preprocessed medical record information to obtain the symptom description information of the target patient can be implemented through the following example.

[0081] Segmenting the preprocessed medical record information according to a preset text segmentation rule to obtain a plurality of text paragraphs;

[0082] Calling a preset NLP model to perform part-of-speech tagging on the multiple text paragraphs in sequence to determine the part-of-speech of each word in each paragraph;

[0083] Based on the part-of-speech tagging results, each paragraph is subjected to syntactic analysis by the NLP model to determine the grammatical structure of the sentence;

[0084] According to the grammatical structure and part-of-speech tagging information, the NLP model extracts words and phrases related to the symptoms;

[0085] Perform semantic analysis on the extracted symptom-related words and phrases, and determine the exact meanings and relationships between the symptom-related words and phrases in combination with medical field knowledge;

[0086] The symptom-related words and phrases, the accurate meanings, and the interrelationships are integrated to generate symptom description information of the target patient.

[0087] In an embodiment of the present invention, illustratively, after the server obtains the medical record information of the patient Xiao Zhang after preprocessing, it works according to the preset text segmentation rules. For example, the basic information part (name, age, gender, etc.) is divided into one section, the medical experience and the doctor's preliminary diagnosis part is divided into one section, the various test results part is divided into one section, the symptom description part is divided into one section, etc., so that multiple text paragraphs are obtained, which is convenient for subsequent one-by-one analysis. Then, the server calls the preset NLP model to perform part-of-speech tagging on these multiple text paragraphs in turn. Taking the symptom description paragraph "Patient Xiao Zhang feels headache, accompanied by slight dizziness, body temperature 37.8 degrees, and throat is a little itchy" as an example, the NLP model will accurately mark "patient" as a noun, "feel" as a verb, "headache" as a noun, "accompanied by" as a verb, "slight" as an adjective, "dizziness" as a noun, "body temperature" as a noun, "37.8 degrees" as a quantifier, "throat" as a noun, "some" as an adverb, "itching" as a verb, etc., to determine the part of speech of each word in each paragraph. Based on the results of part-of-speech tagging, the NLP model further performs syntactic analysis on each paragraph. Taking the above symptom description paragraph as an example, the grammatical structure of the sentence can be determined through analysis, such as "patient Xiao Zhang" is the subject, "feeling headache", "accompanied by slight dizziness", "body temperature 37.8 degrees", "throat a little itchy", etc. are the predicate parts, and the grammatical relationship of each part in the sentence is clarified. According to the grammatical structure and part-of-speech tagging information, the NLP model extracts words and phrases related to the symptoms, such as "headache", "dizziness", "body temperature 37.8 degrees", "throat itchy", etc. in this example. Then, the extracted words and phrases are semantically analyzed, and their exact meanings and relationships are determined in combination with medical field knowledge. For example, "headache" may be a feeling of head pain caused by various reasons, and "dizziness" may be related to factors such as brain blood supply, and they may have some correlation with symptoms such as "body temperature 37.8 degrees", such as a series of manifestations of slight physical discomfort. Finally, the server integrates these symptom-related words and phrases, their exact meanings, and their interrelationships to generate symptom description information for the patient Xiao Zhang, namely, "Patient Xiao Zhang has headache and dizziness (mild) symptoms, a body temperature of 37.8 degrees, an itchy throat, and may have mild physical discomfort."

[0088] In an embodiment of the present invention, generating a medication analysis report for the target patient based on the medication recommendation information and pushing the medication analysis report to a preset user interaction interface can be implemented through the following examples.

[0089] Extract key information based on the medication recommendation information, wherein the key information includes the name of the drug, dosage, frequency of medication, expected duration of medication, and adverse reactions;

[0090] According to the preset report template, the key information is typeset and sorted to generate a medication analysis report containing the title, basic patient information, symptom description, medication recommendation details, and medication precautions;

[0091] Adding a reference link related to the medication recommendation information in the medication analysis report, wherein the reference link points to corresponding medical literature and drug instructions;

[0092] The medication analysis report is pushed to a preset user interaction interface through a preset push method, wherein the push method includes at least one of network transmission and message queue push.

[0093] In an embodiment of the present invention, exemplarily, in the hospital's information management system, after completing a series of medical data analysis and processing of the target patient (such as the patient Xiao Li mentioned above), the server obtains medication recommendation information for Xiao Li. Assume that the medication recommendation information is "For patient Xiao Li, amoxicillin is recommended for treatment. The dosage is 0.5 grams each time, three times a day, and the expected duration of medication is five days. Mild gastrointestinal discomfort and other adverse reactions may occur during medication." The first thing the server needs to do is to extract key information from this medication recommendation information. For the drug name, it is obviously "Amoxicillin", and the server accurately determines it through the recognition and extraction rules of the text content. In terms of medication dosage, the server identifies the key data "0.5 grams each time", which represents the specific amount of amoxicillin taken each time. The frequency of medication is "three times a day", and the server extracts this important information based on the analysis rules of the set time unit and frequency keywords (such as "daily" and "each time", etc.). The estimated duration of medication is "five days". Similarly, by identifying the combination of words and numbers that represent the length of time, the server extracts it as one of the key information. The adverse reaction is "mild gastrointestinal discomfort may occur". The server accurately captures this key content based on the detection of words related to describing negative effects (such as "discomfort" and "adverse reaction", etc.). Through such a meticulous extraction process, the server extracts the key information in the medication recommendation information completely and accurately, preparing for the subsequent generation of the medication analysis report. After extracting the key information, the server generates the medication analysis report based on the preset report template. The first is the title of the report. According to the template setting, it may be "Medication Analysis Report of Patient Xiao Li". The server accurately fills the patient's name in the title position to make it clearly targeted. In the patient's basic information part, the server obtains Xiao Li's relevant basic information from the hospital's patient information database, such as name, age, gender, contact information, etc., and typesets it according to the format required by the template, and neatly lists it in the corresponding position of the report. In the symptom description section, the server will call up the previously generated symptom description information of patient Xiao Li, such as "Patient Xiao Li has a fever (body temperature 38.5 degrees) and is accompanied by coughing (more frequent), slight sore throat, physical fatigue and other symptoms", and fill it in the symptom description area of ​​the report completely. The details of the medication recommendations list the key information just extracted in detail. The name of the drug "Amoxicillin" will be clearly written in the corresponding position, and the dosage of "0.5 grams each time", the frequency of use of "three times a day", and the estimated duration of use of "five days" will also be accurately filled in their respective columns in turn, so that people reading the report can understand the specific medication arrangements at a glance.In terms of medication precautions, based on the extracted adverse reaction "mild gastrointestinal discomfort may occur", the server will add corresponding precautions in the report, such as "pay attention to a light diet during medication, avoid spicy, greasy, and irritating foods, and seek medical attention in time if gastrointestinal discomfort worsens or other abnormal conditions occur". By carefully formatting and arranging key information according to the template, the server successfully generated a complete medication analysis report containing the title, basic patient information, symptom description, medication recommendation details, and medication precautions, providing doctors and patients with comprehensive and clear medication guidance. After the server generates the medication analysis report, in order to enable doctors and patients to have a deeper understanding of medication-related knowledge, it will add reference links related to medication recommendation information to the report. For the recommended drug "Amoxicillin", the server will search for the corresponding drug instructions and related medical literature in the hospital's drug information database and medical literature library. For example, the drug instructions for amoxicillin are found, which details the pharmacological effects, scope of application, contraindications, adverse reactions and other comprehensive information of amoxicillin; at the same time, some medical literature on amoxicillin in the treatment of symptoms similar to those of Xiao Li (such as fever and cough caused by respiratory tract infection) is also found. These literature may contain clinical research results, summary of medication experience and other content. The server will accurately add the links to these found drug instructions and medical literature to the corresponding positions of the medication analysis report. A "reference" column may be set up in the report, and each link will be accompanied by a brief description, such as "amoxicillin drug instructions link" and "medical literature link on amoxicillin for the treatment of respiratory tract infection", so that people who read the report can quickly locate and click to view the details. By adding these reference links, doctors and patients can further understand the details of the drug as needed when viewing the medication analysis report, so as to better follow the medication recommendations and make reasonable diagnosis and treatment decisions. After the server completes the generation of the medication analysis report and adds the reference link, it will push the report to the preset user interaction interface so that doctors and patients can view it in time. Assume that the preset push method adopted by the hospital is a combination of network transmission and message queue push. For network transmission, the server will send the generated medication analysis report in the form of an electronic file to the designated user interaction interface through the hospital's internal LAN. For example, the doctor has a software interface dedicated to viewing the patient's medication analysis report on the hospital's diagnosis and treatment workstation. The server will send the report to the designated receiving folder of this interface. The doctor can see the newly generated report in time as long as he opens the software. For message queue push, the server will encapsulate the medication analysis report according to the set message format and then put it into the message queue. The message queue is a first-in-first-out message delivery mechanism that pushes reports to the corresponding recipients in order. For example, patients can receive medication analysis reports pushed by the message queue through the official hospital APP installed on their mobile phones.When a new report is generated, the APP will receive a push notification, and the patient can click the notification to view the detailed medication analysis report. Through these two preset push methods, the server ensures that the medication analysis report can reach the preset user interaction interface in a timely and accurate manner. Whether it is the doctor in the diagnosis and treatment process or the patient when understanding their own medication situation, they can easily and quickly obtain this important medication analysis report, thereby better guiding medical and medication behavior.

[0094] The embodiment of the present invention provides a computer device 100, which includes a processor and a non-volatile memory storing computer instructions. When the computer instructions are executed by the processor, the computer device 100 executes the aforementioned medical data analysis auxiliary method based on artificial intelligence. Figure 2 As shown, Figure 2 The block diagram of the computer device 100 provided in the embodiment of the present invention. The computer device 100 includes a memory 111, a processor 112 and a communication unit 113. To achieve data transmission or interaction, the memory 111, the processor 112 and the communication unit 113 are directly or indirectly electrically connected to each other. For example, these components can be electrically connected to each other through one or more communication buses or signal lines.

[0095] For illustrative purposes, the foregoing description is made with reference to specific embodiments. However, the above illustrative discussion is not intended to be exhaustive or to limit the present disclosure to the precise form disclosed. Numerous modifications and variations are possible in accordance with the above teachings. These embodiments are selected and described in order to best illustrate the principles of the present disclosure and its practical application, so that those skilled in the art can best utilize the present disclosure and utilize various embodiments with different modifications to suit the intended specific application.

Claims

1. An artificial intelligence-based medical data analysis auxiliary method, characterized in that: include: Obtain medical record information of target patients; Performing denoising and standardization processing on the medical record information respectively to obtain pre-processed medical record information; Calling a preset NLP model to parse the preprocessed medical record information to obtain symptom description information of the target patient; Processing the symptom description information based on a preset large language model to obtain medication recommendation information for the symptom description information; Generate a medication analysis report for the target patient based on the medication recommendation information, and push the medication analysis report to a preset user interaction interface; The symptom description information is processed based on a preset large language model to obtain medication recommendation information of the symptom description information, including: Acquire symptom description information, and determine multiple candidate drug information from the drug database based on cosine similarity; Determining pharmacological compatibility coefficients between the symptom description information and the plurality of candidate drug information, respectively, according to the text description features of the symptom description information and the text description features respectively corresponding to the plurality of candidate drug information; Determining the semantic compatibility between the target candidate drug information and the symptom description information according to the keywords of the symptom description information and the keywords corresponding to the target candidate drug information; determining a first keyword adaptation coefficient between the target candidate drug information and the symptom description information based on the core medical terms in the keywords of the symptom description information and the core medical terms in the keywords corresponding to the target candidate drug information; Determining a second keyword adaptation coefficient between the target candidate drug information and the symptom description information based on non-core medical terms in the keywords of the symptom description information and non-core medical terms in the keywords corresponding to the target candidate drug information; Performing a cumulative operation on the semantic fitness corresponding to the target candidate drug information, the first keyword fitness coefficient, and the second keyword fitness coefficient to obtain the keyword fitness coefficients between the symptom description information and the plurality of candidate drug information respectively; According to the pharmacological adaptation coefficients and the keyword adaptation coefficients respectively corresponding to the plurality of candidate drug information, selecting the candidate drug information matching the medication-related features corresponding to the symptom description information from the plurality of candidate drug information as the recommended drug information; A guide word framework corresponding to the large language model is obtained, and the recommended drug information is placed in the guide word framework to obtain the medication recommendation information.

2. The method according to claim 1, characterized in that The processing of the symptom description information based on a preset large language model to obtain medication recommendation information of the symptom description information also includes: Obtaining symptom description information and the symptom category corresponding to the symptom description information; Determine a drug database matching the symptom category from a plurality of drug databases; the drug database includes a low-dimensional vector representation corresponding to candidate drug information matching the symptom category; Determining the cosine similarity between the low-dimensional vector representation of the symptom description information and the low-dimensional vector representation in the drug database; Determining, from the drug database, a plurality of low-dimensional vector representations whose cosine similarity exceeds a preset cosine similarity threshold; Determining, according to the feature conversion relationship between the low-dimensional vector representation and the candidate drug information, the candidate drug information having feature conversion relationships with the multiple low-dimensional vector representations respectively; The candidate drug information having feature conversion relationships with the multiple low-dimensional vector representations is used as multiple candidate drug information matched with the low-dimensional vector representation of the symptom description information; Get the guide word framework corresponding to the large language model; Storing the symptom description information in a storage location corresponding to the symptom identifier in the guide word framework; The recommended drug information is placed in the storage location corresponding to the drug identification in the guide word framework, and the guide word corresponding to the large language model is obtained. Through the large language model, the medication recommendation information of the symptom description information is obtained according to the guide word.

3. The method according to claim 2, characterized in that The medication-related features corresponding to the candidate drug information also include the storage time node of the candidate drug information; The selecting, from the plurality of candidate drug information, candidate drug information matching the medication-related feature corresponding to the symptom description information as the recommended drug information according to the pharmacological adaptation coefficients and the keyword adaptation coefficients respectively corresponding to the plurality of candidate drug information, includes: Determine the effectiveness coefficients respectively corresponding to the plurality of candidate drug information according to the differences between the storage time nodes and the execution time nodes respectively corresponding to the plurality of candidate drug information; Performing a cumulative operation on the effectiveness coefficient, the pharmacological adaptation coefficient, and the keyword adaptation coefficient corresponding to the target candidate drug information to obtain a first recommended adaptation degree between the target candidate drug information and the symptom description information; The candidate drug information whose first recommended fitness degree exceeds the preset first fitness degree among the multiple candidate drug information is used as the recommended drug information matching the medication-related feature corresponding to the symptom description information.

4. The method according to claim 2, characterized in that: The selecting, from the plurality of candidate drug information, candidate drug information matching the medication-related feature corresponding to the symptom description information as the recommended drug information according to the pharmacological adaptation coefficients and the keyword adaptation coefficients respectively corresponding to the plurality of candidate drug information, includes: Performing a cumulative operation on the pharmacological fitness coefficient and the keyword fitness coefficient corresponding to the target candidate drug information to obtain a second recommended fitness between the target candidate drug information and the symptom description information; The candidate drug information whose second recommended fitness degree exceeds the preset second fitness degree among the multiple candidate drug information is used as the recommended drug information matching the medication-related feature corresponding to the symptom description information.

5. The method according to claim 2, characterized in that: The method further comprises: Acquire medical record data to be processed, and a disease association type corresponding to the medical record data; the disease association type corresponds to a symptom category; Extracting and identifying the medical record data according to the disease association type to obtain multiple candidate drug information, and obtaining low-dimensional vector representations corresponding to the multiple candidate drug information respectively; The low-dimensional vector representations corresponding to the plurality of candidate drug information are stored in a drug database corresponding to the disease association type.

6. The method according to claim 5, characterized in that The extracting and identifying the medical record data according to the disease association type to obtain multiple candidate drug information and obtaining low-dimensional vector representations corresponding to the multiple candidate drug information respectively include: According to the type of association of the disease, determine the key medical words and phrases related to it; searching the medical record data for text segments containing the key medical words and phrases; Performing text cleaning on the searched text fragments, and performing word segmentation on the cleaned text fragments to obtain multiple vocabulary units; Generate low-dimensional vectors corresponding to the plurality of vocabulary units according to a preset vector generation algorithm; The generated low-dimensional vectors are combined and optimized to obtain low-dimensional vector representations corresponding to the multiple candidate drug information.

7. The method according to claim 1, characterized in that The medical record information is subjected to denoising and standardization processing respectively to obtain pre-processed medical record information, including: Performing denoising on the medical record information, wherein the denoising adopts a text filtering algorithm to remove duplicate information, irrelevant formatting marks, and erroneous statements that do not comply with medical record specifications in the medical record information; The denoised medical record information is standardized, and the standardization includes unifying the date format into a preset standard format, restoring the abbreviations of medical terms into complete forms, and converting measurement data under different unit systems into a unified standard unit system to obtain the pre-processed medical record information.

8. The method according to claim 1, characterized in that The calling of a preset NLP model to parse the preprocessed medical record information to obtain symptom description information of the target patient includes: Segmenting the preprocessed medical record information according to a preset text segmentation rule to obtain a plurality of text paragraphs; Calling a preset NLP model to perform part-of-speech tagging on the multiple text paragraphs in sequence to determine the part-of-speech of each word in each paragraph; Based on the part-of-speech tagging results, each paragraph is subjected to syntactic analysis by the NLP model to determine the grammatical structure of the sentence; According to the grammatical structure and part-of-speech tagging information, the NLP model extracts words and phrases related to the symptoms; Perform semantic analysis on the extracted symptom-related words and phrases, and determine the exact meanings and relationships between the symptom-related words and phrases in combination with medical field knowledge; The symptom-related words and phrases, the accurate meanings, and the interrelationships are integrated to generate symptom description information of the target patient.

9. The method according to claim 1, characterized in that: The generating a medication analysis report for the target patient based on the medication recommendation information, and pushing the medication analysis report to a preset user interaction interface, includes: Extract key information based on the medication recommendation information, wherein the key information includes the name of the drug, dosage, frequency of medication, expected duration of medication, and adverse reactions; According to the preset report template, the key information is typeset and sorted to generate a medication analysis report containing the title, basic patient information, symptom description, medication recommendation details, and medication precautions; Adding a reference link related to the medication recommendation information in the medication analysis report, wherein the reference link points to corresponding medical literature and drug instructions; The medication analysis report is pushed to a preset user interaction interface through a preset push method, wherein the push method includes at least one of network transmission and message queue push.

10. A server system, characterized in that: The method comprises a server, wherein the server is used to execute the method described in any one of claims 1 to 9.

Images