Large model-based medical health consultation pushing method and system, and storage medium

Abstract

The invention provides a medical health consultation pushing method and system based on a large model and a storage medium, and the method comprises the steps: obtaining a group chat consultation text, inputting the group chat consultation text into a pre-trained large model, and carrying out the classification of user questions, and obtaining user consultation questions; performing semantic analysis on the user consultation questions to obtain user semantics, and determining consultation requirements and consultation diseases according to the user semantics; historical consultation question and answer information of the user corresponding to the user consultation question is acquired, and medical history information is determined according to the historical consultation question and answer information; and according to the consultation demand, the consultation disease and the medical history information, determining medical health push information, and according to the medical health push information, carrying out information push on a user corresponding to the user consultation question. According to the embodiment of the invention, the accuracy of the medical health pushing information determined based on the consultation demand, the consultation disease and the medical history information is relatively high, and the accuracy of medical health consultation pushing is improved.

Description

Technical Field

[0001] The present invention relates to the field of smart medical technology, and in particular to a medical health consultation push method, system and storage medium based on a large model. Background Art

[0002] With the development of computer and network technology, electronic medical and health information has gradually become the mainstream in the field of medical and health information. For example, hospitals or physical examination institutions establish electronic medical records for each patient user to facilitate doctors' management of patients; or, patients join group chats on terminal devices (such as smartphones, tablets, etc.) and conduct medical and health consultations online. In order to improve the user's experience of medical and health consultations, the issue of medical and health consultation push has received more and more attention.

[0003] In the existing medical and health consultation push process, the same health push information is generally pushed to users with different medical histories and the same problems, which leads to low accuracy of medical and health consultation push. Summary of the invention

[0004] The purpose of the embodiments of the present invention is to provide a method, system and storage medium for pushing medical and health consultation based on a large model, so as to solve the problem of low accuracy of pushing medical and health consultation in the prior art.

[0005] The embodiment of the present invention is implemented as follows: a medical health consultation push method based on a large model, the method comprising:

[0006] Obtaining a group chat consultation text, and inputting the group chat consultation text into a pre-trained large model to classify user questions, thereby obtaining user consultation questions;

[0007] Performing semantic analysis on the user's consultation questions to obtain user semantics, and determining consultation needs and consultation diseases based on the user semantics;

[0008] Obtain historical consultation question and answer information of the user corresponding to the user consultation question, and determine medical history information based on the historical consultation question and answer information;

[0009] Medical health push information is determined according to the consultation demand, the consultation disease and the medical history information, and information is pushed to the user corresponding to the user consultation question according to the medical health push information.

[0010] Preferably, determining the medical history information according to the historical consultation question and answer information includes:

[0011] Performing entity recognition on the historical consultation question and answer information to obtain a consultation entity, and determining a previous disease word, a symptom description word, a diagnosis description word, and a medication description word according to the entity type of the consultation entity;

[0012] Extract sentences based on the text position of the past disease words in the historical consultation question and answer information to obtain past history information, and extract sentences based on the text position of the symptom description words in the historical consultation question and answer information to obtain current medical history information;

[0013] Extracting sentences based on the text position of the diagnosis description word in the historical consultation question and answer information to obtain diagnosis information, and extracting sentences based on the text position of the medication description word in the historical consultation question and answer information to obtain medication information;

[0014] The medical history information, the medical history information, the diagnosis information and the medication information are combined to obtain the medical history information.

[0015] Preferably, after pushing information to the user corresponding to the user's consultation question according to the medical health push information, the method further includes:

[0016] Receiving user feedback text for the medical health push information, and performing sentiment analysis on the user feedback text to obtain user sentiment;

[0017] Determine question explanation information according to the user emotion and the medical health push information, and provide question feedback to the user feedback text according to the question explanation information.

[0018] Preferably, performing sentiment analysis on the user feedback text to obtain user sentiment includes:

[0019] Segmenting the user feedback text to obtain feedback segmented words, and removing stop words from the feedback segmented words;

[0020] Performing feature extraction on the feedback word segment after the stop word processing to obtain feedback features, and performing word embedding processing on the feedback features to obtain feedback word embedding features;

[0021] The feature similarity between the feedback word embedding feature and the emotion standard feature is calculated, and the emotion corresponding to the maximum feature similarity is set as the user emotion.

[0022] Preferably, the method further comprises:

[0023] Obtain the historical consultation time of the group chat user, and determine the non-consultation time based on the historical consultation time and the current time;

[0024] If any of the non-consultation time periods is greater than the time threshold, the group chat user corresponding to the non-consultation time period is set as the target push user;

[0025] Obtaining the historical consultation diseases of the target push user, and determining health advice based on the historical consultation diseases;

[0026] The health advice is pushed to the target push user.

[0027] Preferably, before inputting the group chat consultation text into the pre-trained large model for user question classification, the method further includes:

[0028] Input the group chat consultation samples into the large model to classify user questions and obtain user classification results;

[0029] Input the consultation question samples into the big model for semantic analysis to obtain the semantics of the question samples, and perform demand prediction based on the semantics of the question samples to obtain predicted demand and predicted disease;

[0030] Obtaining historical consultation question and answer samples corresponding to the consultation question samples, and performing medical history prediction based on the historical consultation question and answer samples to obtain predicted medical history;

[0031] The sample push information is determined according to the predicted demand, the predicted disease and the predicted medical history, and the large model is trained according to the sample push information until the large model converges to obtain the pre-trained large model.

[0032] Preferably, performing model training on the large model according to the sample push information includes:

[0033] Calculating the classification similarity between the user classification result and the classification standard result, and determining the first model loss according to the classification similarity;

[0034] Calculating the semantic similarity between the semantics of the question sample and the semantics of the question standard, and determining the second model loss according to the semantic similarity;

[0035] Calculating the demand similarity between the predicted demand and the standard demand, and determining the third model loss according to the demand similarity;

[0036] Calculating the disease similarity between the predicted disease and the standard disease, and determining a fourth model loss according to the disease similarity;

[0037] Calculating the medical history similarity between the predicted medical history and the standard medical history, and determining the fifth model loss according to the medical history similarity;

[0038] Calculating information similarity between the sample push information and the standard push information, and determining a sixth model loss according to the information similarity;

[0039] The first model loss, the second model loss, the third model loss, the fourth model loss, the fifth model loss and the sixth model loss are weighted to obtain a total model loss, and parameters of the large model are updated according to the total model loss.

[0040] Another object of an embodiment of the present invention is to provide a medical health consultation push system based on a large model, the system comprising:

[0041] A user classification module is used to obtain group chat consultation texts and input the group chat consultation texts into a pre-trained large model to classify user questions to obtain user consultation questions;

[0042] A semantic analysis module, used to perform semantic analysis on the user's consultation questions, obtain user semantics, and determine consultation needs and consultation diseases based on the user semantics;

[0043] A medical history determination module is used to obtain historical consultation question and answer information of the user corresponding to the user consultation question, and determine the medical history information based on the historical consultation question and answer information;

[0044] The health push module is used to determine medical health push information according to the consultation needs, the consultation diseases and the medical history information, and push information to the user corresponding to the user's consultation questions according to the medical health push information.

[0045] Preferably, the medical history determination module is further used for:

[0046] Performing entity recognition on the historical consultation question and answer information to obtain a consultation entity, and determining a previous disease word, a symptom description word, a diagnosis description word, and a medication description word according to the entity type of the consultation entity;

[0047] Extract sentences based on the text position of the past disease words in the historical consultation question and answer information to obtain past history information, and extract sentences based on the text position of the symptom description words in the historical consultation question and answer information to obtain current medical history information;

[0048] Extracting sentences based on the text position of the diagnosis description word in the historical consultation question and answer information to obtain diagnosis information, and extracting sentences based on the text position of the medication description word in the historical consultation question and answer information to obtain medication information;

[0049] The medical history information, the medical history information, the diagnosis information and the medication information are combined to obtain the medical history information.

[0050] The embodiments of the present invention classify user questions in group chat consultation texts through a pre-trained big model, and can effectively identify user consultation questions sent by different users in the group chat consultation texts. At the same time, medical and health consultation push can be performed on user consultation questions sent by different users. By performing semantic analysis on user consultation questions through the pre-trained big model, the user semantics of user consultation questions can be effectively extracted. The medical and health push information determined based on consultation needs, consultation diseases and medical history information has high accuracy. For users with different medical histories and the same problems, the same medical and health push information is prevented from being pushed to users, thereby improving the accuracy of medical and health consultation push. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 is a flow chart of a medical health consultation push method based on a large model provided in a first embodiment of the present invention;

[0052] Figure 2 It is a structural diagram of a medical health consultation push system based on a large model provided by a second embodiment of the present invention;

[0053] Figure 3 It is a schematic diagram of the structure of a terminal device provided in the third embodiment of the present invention. DETAILED DESCRIPTION

[0054] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0055] In order to illustrate the technical solution of the present invention, a specific embodiment is provided below for illustration.

[0056] Embodiment 1

[0057] See also Figure 1 , is a flow chart of a medical health consultation push method based on a big model provided by a first embodiment of the present invention. The medical health consultation push method based on a big model can be applied to any device or system. The medical health consultation push method based on a big model includes the following steps:

[0058] Step S10, obtaining a group chat consultation text, and inputting the group chat consultation text into a pre-trained large model to classify user questions, thereby obtaining user consultation questions;

[0059] Among them, by using the pre-trained large model to classify user questions in the group chat consultation text, it is possible to effectively identify the user consultation questions sent by different users in the group chat consultation text, and at the same time, medical and health consultations can be pushed to user consultation questions sent by different users, so as to achieve the effect of multi-threaded processing of user consultation questions. In this step, the user name of the user question in the group chat consultation text can be obtained, and the user question classification can be performed based on the user name.

[0060] Step S20, performing semantic analysis on the user's consultation question to obtain user semantics, and determining consultation needs and consultation diseases according to the user semantics;

[0061] Among them, semantic analysis of user consultation questions is performed through a pre-trained large model, which can effectively extract the user semantics of the user consultation questions. Based on the user semantics, the corresponding user's consultation needs and consultation diseases can be effectively determined. In this step, a named entity recognition algorithm can be used to perform semantic analysis of user consultation questions.

[0062] Step S30, obtaining historical consultation question and answer information of the user corresponding to the user consultation question, and determining medical history information according to the historical consultation question and answer information;

[0063] Among them, the historical group chat information of the group chat corresponding to the group chat consultation text is obtained, the name of the user corresponding to the user consultation question is matched with the historical group chat information, the historical consultation question and answer information is determined based on the matching result, and the medical history information of the corresponding user can be automatically determined based on the historical consultation question and answer information.

[0064] Optionally, determining the medical history information based on the historical consultation question and answer information includes:

[0065] Performing entity recognition on the historical consultation question and answer information to obtain a consultation entity, and determining the previous disease words, symptom description words, diagnosis description words, and medication description words according to the entity type of the consultation entity; wherein, by matching the entity type of the consultation entity with a preset entity type, determining the previous disease words, symptom description words, diagnosis description words, and medication description words in the consultation entity based on the matching result of the entity type;

[0066] Sentence extraction is performed based on the text position of the past disease words in the historical consultation question and answer information to obtain past history information, and sentence extraction is performed based on the text position of the symptom description words in the historical consultation question and answer information to obtain current medical history information; wherein, the context text of the sentence corresponding to the past disease word in the historical consultation question and answer information is obtained to obtain the past history information, and the context text of the sentence corresponding to the symptom description word in the historical consultation question and answer information is obtained to obtain the current medical history information;

[0067] Sentence extraction is performed based on the text position of the diagnosis descriptor in the historical consultation question and answer information to obtain diagnosis information, and sentence extraction is performed based on the text position of the medication description in the historical consultation question and answer information to obtain medication information; wherein, the context text of the sentence corresponding to the diagnosis descriptor in the historical consultation question and answer information is obtained to obtain diagnosis information, and the context text of the sentence corresponding to the medication description in the historical consultation question and answer information is obtained to obtain medication information;

[0068] The medical history information, the medical history information, the diagnosis information and the medication information are combined to obtain the medical history information.

[0069] Step S40, determining medical health push information according to the consultation demand, the consultation disease and the medical history information, and pushing information to the user corresponding to the user consultation question according to the medical health push information;

[0070] Among them, the consultation needs, consultation diseases and medical history information are mapped respectively to obtain the demand vector, disease vector and medical history vector, the demand vector, disease vector and medical history vector are combined to obtain the combination vector, the combination vector is matched with the medical health push database to obtain medical health push information, and the medical health push database stores the correspondence between different combination vectors and corresponding medical health push information.

[0071] Optionally, after information is pushed to the user corresponding to the user's consultation question according to the medical health push information, the method further includes:

[0072] Receiving user feedback text for the medical health push information, and performing sentiment analysis on the user feedback text to obtain user sentiment; wherein the user's current emotional state is obtained by performing sentiment analysis on the user feedback text;

[0073] Determine problem explanation information based on the user emotion and the medical health push information, and provide problem feedback to the user feedback text based on the problem explanation information; wherein, when the user emotion is a preset emotion, combine the user emotion and the medical health push information to obtain combined information, match the combined information with a feedback database to obtain problem explanation information, and the preset emotion can be set according to needs, for example, the preset emotion can be set to anger or excitement, etc.

[0074] Further, sentiment analysis is performed on the user feedback text to obtain user sentiment, including:

[0075] Segmenting the user feedback text to obtain feedback segmentation, and performing stop word removal processing on the feedback segmentation; wherein, by performing stop word removal processing on the feedback segmentation, invalid characters in the feedback segmentation can be effectively removed, thereby improving the quality of the feedback segmentation;

[0076] Performing feature extraction on the feedback word segment after the stop word removal process to obtain feedback features, and performing word embedding process on the feedback features to obtain feedback word embedding features; wherein, by performing word embedding process on the feedback features, the feedback features can be effectively characterized and embedded;

[0077] The feature similarity between the feedback word embedding feature and the emotion standard feature is calculated, and the emotion corresponding to the maximum feature similarity is set as the user emotion; wherein the emotion standard feature can be set according to the demand.

[0078] Furthermore, the method further comprises:

[0079] Obtain the historical consultation time of the group chat user, and determine the non-consultation time according to the historical consultation time and the current time; wherein, obtain the sending time of the last consultation question of the group chat user to obtain the historical consultation time;

[0080] If any of the non-consultation time is greater than the time threshold, the group chat user corresponding to the non-consultation time is set as the target push user; wherein the time threshold can be set according to demand, and when the non-consultation time is greater than the time threshold, it is determined that the corresponding group chat user has not had medical health consultation for a long time;

[0081] Obtaining the historical consultation diseases of the target push user, and determining health advice based on the historical consultation diseases;

[0082] The health advice is pushed to the target push user; wherein, by pushing the health advice to the target push user, the effect of automatically pushing information to users who have not received medical health consultation for a long time is achieved.

[0083] Optionally, before inputting the group chat consultation text into the pre-trained large model for user question classification, the method further includes:

[0084] Input the group chat consultation sample into the large model to classify the user questions and obtain the user classification result; wherein the user question classification is used to classify the questions in the group chat consultation sample by users;

[0085] Input the consultation question samples into the big model for semantic analysis to obtain the semantics of the question samples, and perform demand prediction based on the semantics of the question samples to obtain predicted demand and predicted disease; wherein, by inputting the consultation question samples into the big model for semantic analysis, the semantics of the question samples of the consultation question samples can be effectively extracted, and based on the semantics of the question samples, the predicted demand and predicted disease can be effectively determined;

[0086] Obtaining historical consultation question and answer samples corresponding to the consultation question samples, and performing medical history prediction based on the historical consultation question and answer samples to obtain predicted medical history;

[0087] The sample push information is determined according to the predicted demand, the predicted disease and the predicted medical history, and the large model is trained according to the sample push information until the large model converges to obtain the pre-trained large model.

[0088] Further, the large model is trained according to the sample push information, including:

[0089] Calculate the classification similarity between the user classification result and the classification standard result, and determine the first model loss according to the classification similarity, calculate the semantic similarity between the problem sample semantics and the problem standard semantics, and determine the second model loss according to the semantic similarity, calculate the demand similarity between the predicted demand and the standard demand, and determine the third model loss according to the demand similarity, calculate the disease similarity between the predicted disease and the standard disease, and determine the fourth model loss according to the disease similarity, calculate the medical history similarity between the predicted medical history and the standard medical history, and determine the fifth model loss according to the medical history similarity, calculate the information similarity between the sample push information and the standard push information, and determine the sixth model loss according to the information similarity; wherein, the classification standard result, problem standard semantics, standard demand, standard disease, standard medical history and standard push information can all be set according to demand;

[0090] The first model loss, the second model loss, the third model loss, the fourth model loss, the fifth model loss and the sixth model loss are weighted to obtain a total model loss, and the parameters of the large model are updated according to the total model loss; wherein the weighting coefficients of the first model loss, the second model loss, the third model loss, the fourth model loss, the fifth model loss and the sixth model loss in the weighted calculation process can be set according to demand.

[0091] Multi-user, multi-round intelligent group push mechanism: This embodiment is based on a large language model, which intelligently classifies the questions of different users in the group and maintains contextual understanding of different users in multiple rounds of conversations to achieve accurate push of user needs. Through the context memory ability of the model, multi-round, multi-threaded intelligent push can be performed for different patients in the same group to ensure that the needs of each patient can be fully responded to.

[0092] For example, in a group, multiple patients are consulting about different diseases at the same time. The big model can identify the keywords, disease names, and medical history information in the patients' questions, and intelligently assign each patient's questions to push medical knowledge or suggestions related to their needs.

[0093] Through the contextual memory and multi-threaded processing capabilities of large-scale language models, it can effectively cope with complex scenarios where multiple users in a group are consulting at the same time, ensuring that the needs of each user can be accurately understood and responded to, solving the limitations of existing technologies in multi-round, multi-threaded scenarios.

[0094] Professional medical knowledge push based on large language models: Using large language models, with the support of professional knowledge bases in the field of medical health, highly professional advice is provided for patients' symptoms, medication consultation and other issues. The large model can not only understand complex medical terms, but also recommend health consultation content that better meets the patient's needs based on the patient's specific situation, such as pushing personalized diet advice and treatment plans based on different medical histories and symptoms.

[0095] For example, when a patient asks a question about "hypertension", the big model can understand the patient's medical history and current medication status through contextual information, and push the most relevant content to his or her condition, such as drug side effects, medication precautions, etc., while also pushing relevant health education information.

[0096] By combining with professional medical knowledge bases, the big model can provide more accurate and professional medical and health advice, no longer relying on simple keyword matching and templated responses, and addressing the shortcomings of existing chatbots in dealing with medical professional issues.

[0097] Context tracking and memory of multiple rounds of conversations: The large model has the ability to remember the context of the conversation, and can track and maintain the memory of the patient's personal information and medical history in multiple rounds of doctor-patient interaction. For example, if a patient mentioned that he had high blood pressure in a previous round of conversation, the model can remember the patient's condition in the next conversation and provide more accurate suggestions and answers based on previous information. The context tracking capability eliminates the need for patients to repeatedly describe their condition, and can provide more accurate health push based on historical conversations, ensuring that the information is coherent and efficient.

[0098] Multi-threaded parallel processing and precise guidance: This embodiment can handle the different needs of multiple users in the group at the same time through multi-threaded parallel processing. In actual applications, when multiple patients are consulting on different issues at the same time, the big model can perform multi-threaded processing on different consultation contents, accurately identify the content of the user's questions, and push medical and health information. At the same time, through the guidance ability of the big model, it can intervene in the user's discussion at the appropriate time to provide question answers, emotional regulation or opinion guidance.

[0099] For example, if a patient shows misunderstanding of the treatment plan or emotional fluctuations during the consultation process, the big model can be used to perform sentiment analysis on the patient's sentences to detect the emotional state, and provide emotional guidance or problem explanations in a timely manner, thereby avoiding tension in the doctor-patient relationship and communication breakdown.

[0100] Personalized push notifications for group user needs: Large language models can dynamically adjust according to the individual needs of each patient and provide personalized push notifications for medical and health information. For example, for users with diabetes, the large model can push health advice specifically for diabetic patients on diet, exercise, and drug management; for users with a history of cardiovascular disease, the large model can push relevant medical advice on preventing cardiovascular disease attacks.

[0101] In addition, the big model can continuously optimize the push content based on continuous learning of each patient, thereby ensuring the personalization and professionalism of medical advice. The big model can also dynamically adjust according to the individual needs of patients to provide more personalized health information push, further improving the application effect and user experience. At the same time, combined with multiple rounds of context tracking, the model can continuously optimize the push content and achieve long-term accurate feedback on user needs.

[0102] In this embodiment, user questions in group chat consultation texts are classified through a pre-trained big model, which can effectively identify user consultation questions sent by different users in the group chat consultation texts, and medical and health consultation push can be performed on user consultation questions sent by different users at the same time. User consultation questions are semantically analyzed through a pre-trained big model, which can effectively extract user semantics of user consultation questions. The medical and health push information determined based on consultation needs, consultation diseases and medical history information has high accuracy. For users with different medical histories and the same problems, the same medical and health push information is prevented from being pushed to users, thereby improving the accuracy of medical and health consultation push.

[0103] Embodiment 2

[0104] See also Figure 2 , is a schematic diagram of the structure of a medical health consultation push system 100 based on a large model provided in a second embodiment of the present invention, including:

[0105] The user classification module 10 is used to obtain the group chat consultation text, and input the group chat consultation text into the pre-trained large model to classify user questions to obtain user consultation questions.

[0106] Optionally, the user classification module 10 is further used to: input the group chat consultation sample into the large model to classify the user questions and obtain the user classification result;

[0107] Input the consultation question samples into the big model for semantic analysis to obtain the semantics of the question samples, and perform demand prediction based on the semantics of the question samples to obtain predicted demand and predicted disease;

[0108] Obtaining historical consultation question and answer samples corresponding to the consultation question samples, and performing medical history prediction based on the historical consultation question and answer samples to obtain predicted medical history;

[0109] The sample push information is determined according to the predicted demand, the predicted disease and the predicted medical history, and the large model is trained according to the sample push information until the large model converges to obtain the pre-trained large model.

[0110] Furthermore, the user classification module 10 is further used to: calculate the classification similarity between the user classification result and the classification standard result, and determine the first model loss according to the classification similarity;

[0111] Calculating the semantic similarity between the semantics of the question sample and the semantics of the question standard, and determining the second model loss according to the semantic similarity;

[0112] Calculating the demand similarity between the predicted demand and the standard demand, and determining the third model loss according to the demand similarity;

[0113] Calculating the disease similarity between the predicted disease and the standard disease, and determining a fourth model loss according to the disease similarity;

[0114] Calculating the medical history similarity between the predicted medical history and the standard medical history, and determining the fifth model loss according to the medical history similarity;

[0115] Calculating information similarity between the sample push information and the standard push information, and determining a sixth model loss according to the information similarity;

[0116] The first model loss, the second model loss, the third model loss, the fourth model loss, the fifth model loss and the sixth model loss are weighted to obtain a total model loss, and parameters of the large model are updated according to the total model loss.

[0117] The semantic analysis module 11 is used to perform semantic analysis on the user's consultation questions to obtain user semantics, and determine the consultation needs and consultation diseases based on the user semantics.

[0118] The medical history determination module 12 is used to obtain historical consultation question and answer information of the user corresponding to the user consultation question, and determine the medical history information based on the historical consultation question and answer information.

[0119] Optionally, the medical history determination module 12 is further used to: perform entity recognition on the historical consultation question and answer information to obtain a consultation entity, and determine a previous disease word, a symptom description word, a diagnosis description word, and a medication description word according to the entity type of the consultation entity;

[0120] Extract sentences based on the text position of the past disease words in the historical consultation question and answer information to obtain past history information, and extract sentences based on the text position of the symptom description words in the historical consultation question and answer information to obtain current medical history information;

[0121] Extracting sentences based on the text position of the diagnosis description word in the historical consultation question and answer information to obtain diagnosis information, and extracting sentences based on the text position of the medication description word in the historical consultation question and answer information to obtain medication information;

[0122] The medical history information, the medical history information, the diagnosis information and the medication information are combined to obtain the medical history information.

[0123] The health push module 13 is used to determine medical health push information according to the consultation demand, the consultation disease and the medical history information, and push information to the user corresponding to the user's consultation question according to the medical health push information.

[0124] Optionally, the health push module 13 is further used to: receive user feedback text for the medical health push information, and perform sentiment analysis on the user feedback text to obtain user sentiment;

[0125] Determine question explanation information according to the user emotion and the medical health push information, and provide question feedback to the user feedback text according to the question explanation information.

[0126] Furthermore, the health push module 13 is also used to: segment the user feedback text to obtain feedback segmentation, and remove stop words from the feedback segmentation;

[0127] Performing feature extraction on the feedback word segment after the stop word processing to obtain feedback features, and performing word embedding processing on the feedback features to obtain feedback word embedding features;

[0128] The feature similarity between the feedback word embedding feature and the emotion standard feature is calculated, and the emotion corresponding to the maximum feature similarity is set as the user emotion.

[0129] Furthermore, the health push module 13 is also used to: obtain the historical consultation time of the group chat user, and determine the non-consultation time according to the historical consultation time and the current time;

[0130] If any of the non-consultation time periods is greater than the time threshold, the group chat user corresponding to the non-consultation time period is set as the target push user;

[0131] Obtaining the historical consultation diseases of the target push user, and determining health advice based on the historical consultation diseases;

[0132] The health advice is pushed to the target push user.

[0133] In this embodiment, user questions are classified in group chat consultation texts through a pre-trained big model, and user consultation questions sent by different users in the group chat consultation texts can be effectively identified. Medical and health consultation push can be performed on user consultation questions sent by different users at the same time. Semantic analysis of user consultation questions is performed through a pre-trained big model, and user semantics of user consultation questions can be effectively extracted. The medical and health push information determined based on consultation needs, consultation diseases and medical history information has high accuracy. For users with different medical histories and the same problems, the same medical and health push information is prevented from being pushed to users, thereby improving the accuracy of medical and health consultation push.

[0134] Embodiment 3

[0135] Figure 3 2 is a block diagram of a terminal device 2 provided in the third embodiment of the present application. Figure 3 As shown, the terminal device 2 of this embodiment includes: a processor 20, a memory 21, and a computer program 22 stored in the memory 21 and executable on the processor 20, such as a program of a medical health consultation push method based on a large model. When the processor 20 executes the computer program 22, the steps in each embodiment of the medical health consultation push method based on a large model are implemented.

[0136] Exemplarily, the computer program 22 may be divided into one or more modules, which are stored in the memory 21 and executed by the processor 20 to complete the present application. The one or more modules may be a series of computer program instruction segments capable of completing specific functions, which are used to describe the execution process of the computer program 22 in the terminal device 2. The terminal device may include, but is not limited to, a processor 20 and a memory 21.

[0137] The processor 20 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc.

[0138] The memory 21 may be an internal storage unit of the terminal device 2, such as a hard disk or memory of the terminal device 2. The memory 21 may also be an external storage device of the terminal device 2, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the terminal device 2. Further, the memory 21 may also include both an internal storage unit and an external storage device of the terminal device 2. The memory 21 is used to store the computer program and other programs and data required by the terminal device. The memory 21 may also be used to temporarily store data that has been output or is to be output.

[0139] In addition, each functional module in each embodiment of the present application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The above integrated unit can be implemented in the form of hardware or in the form of software functional unit.

[0140] If the integrated module is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Among them, the computer-readable storage medium can be non-volatile or volatile. Based on this understanding, the present application implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable storage medium may include: any entity or device capable of carrying computer program code, recording medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium, etc. It should be noted that the content contained in computer-readable storage media can be appropriately increased or decreased according to the requirements of legislation and patent practices in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practices, computer-readable storage media do not include electrical carrier signals and telecommunication signals.

[0141] The embodiments described above are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, a person skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application, and should all be included in the protection scope of the present application.

Claims

1. A method for pushing medical and health consultation based on a large model, characterized in that: The method comprises: Obtaining a group chat consultation text, and inputting the group chat consultation text into a pre-trained large model to classify user questions, thereby obtaining user consultation questions; Performing semantic analysis on the user's consultation questions to obtain user semantics, and determining consultation needs and consultation diseases based on the user semantics; Obtain historical consultation question and answer information of the user corresponding to the user consultation question, and determine medical history information based on the historical consultation question and answer information; Medical health push information is determined according to the consultation demand, the consultation disease and the medical history information, and information is pushed to the user corresponding to the user consultation question according to the medical health push information.

2. The method for pushing medical and health consultation based on a large model as claimed in claim 1, characterized in that: Determine the medical history information based on the historical consultation question and answer information, including: Performing entity recognition on the historical consultation question and answer information to obtain a consultation entity, and determining a previous disease word, a symptom description word, a diagnosis description word, and a medication description word according to the entity type of the consultation entity; Extract sentences based on the text position of the past disease words in the historical consultation question and answer information to obtain past history information, and extract sentences based on the text position of the symptom description words in the historical consultation question and answer information to obtain current medical history information; Extracting sentences based on the text position of the diagnosis description word in the historical consultation question and answer information to obtain diagnosis information, and extracting sentences based on the text position of the medication description word in the historical consultation question and answer information to obtain medication information; The medical history information, the medical history information, the diagnosis information and the medication information are combined to obtain the medical history information.

3. The method for pushing medical and health consultation based on a large model as claimed in claim 1, characterized in that: After pushing information to the user corresponding to the user's consultation question according to the medical health push information, the method further includes: Receiving user feedback text for the medical health push information, and performing sentiment analysis on the user feedback text to obtain user sentiment; Determine question explanation information according to the user emotion and the medical health push information, and provide question feedback to the user feedback text according to the question explanation information.

4. The method for pushing medical and health consultation based on a large model as claimed in claim 3, characterized in that: Performing sentiment analysis on the user feedback text to obtain user sentiment includes: Segmenting the user feedback text to obtain feedback segmented words, and removing stop words from the feedback segmented words; Performing feature extraction on the feedback word segment after the stop word processing to obtain feedback features, and performing word embedding processing on the feedback features to obtain feedback word embedding features; The feature similarity between the feedback word embedding feature and the emotion standard feature is calculated, and the emotion corresponding to the maximum feature similarity is set as the user emotion.

5. The method for pushing medical and health consultation based on a large model as claimed in claim 1, characterized in that: The method further comprises: Obtain the historical consultation time of the group chat user, and determine the non-consultation time based on the historical consultation time and the current time; If any of the non-consultation time periods is greater than the time threshold, the group chat user corresponding to the non-consultation time period is set as the target push user; Obtaining the historical consultation diseases of the target push user, and determining health advice based on the historical consultation diseases; The health advice is pushed to the target push user.

6. The method for pushing medical and health consultation based on a large model as claimed in claim 1, characterized in that: Before inputting the group chat consultation text into the pre-trained large model for user question classification, the method further includes: Input the group chat consultation samples into the large model to classify user questions and obtain user classification results; Input the consultation question samples into the big model for semantic analysis to obtain the semantics of the question samples, and perform demand prediction based on the semantics of the question samples to obtain predicted demand and predicted disease; Obtaining historical consultation question and answer samples corresponding to the consultation question samples, and performing medical history prediction based on the historical consultation question and answer samples to obtain predicted medical history; The sample push information is determined according to the predicted demand, the predicted disease and the predicted medical history, and the large model is trained according to the sample push information until the large model converges to obtain the pre-trained large model.

7. The method for pushing medical and health consultation based on a large model as claimed in claim 6, characterized in that: Performing model training on the large model according to the sample push information includes: Calculating the classification similarity between the user classification result and the classification standard result, and determining the first model loss according to the classification similarity; Calculating the semantic similarity between the semantics of the question sample and the semantics of the question standard, and determining the second model loss according to the semantic similarity; Calculating the demand similarity between the predicted demand and the standard demand, and determining the third model loss according to the demand similarity; Calculating the disease similarity between the predicted disease and the standard disease, and determining a fourth model loss according to the disease similarity; Calculating the medical history similarity between the predicted medical history and the standard medical history, and determining the fifth model loss according to the medical history similarity; Calculating information similarity between the sample push information and the standard push information, and determining a sixth model loss according to the information similarity; The first model loss, the second model loss, the third model loss, the fourth model loss, the fifth model loss and the sixth model loss are weighted to obtain a total model loss, and parameters of the large model are updated according to the total model loss.

8. A medical health consultation push system based on a large model, characterized in that: The system comprises: A user classification module is used to obtain group chat consultation texts and input the group chat consultation texts into a pre-trained large model to classify user questions to obtain user consultation questions; A semantic analysis module, used to perform semantic analysis on the user's consultation questions, obtain user semantics, and determine consultation needs and consultation diseases based on the user semantics; A medical history determination module is used to obtain historical consultation question and answer information of the user corresponding to the user consultation question, and determine the medical history information based on the historical consultation question and answer information; The health push module is used to determine medical health push information according to the consultation needs, the consultation diseases and the medical history information, and push information to the user corresponding to the user's consultation questions according to the medical health push information.

9. The medical health consultation push system based on a large model as claimed in claim 8, characterized in that: The medical history determination module is also used to: Performing entity recognition on the historical consultation question and answer information to obtain a consultation entity, and determining a previous disease word, a symptom description word, a diagnosis description word, and a medication description word according to the entity type of the consultation entity; Extract sentences based on the text position of the past disease words in the historical consultation question and answer information to obtain past history information, and extract sentences based on the text position of the symptom description words in the historical consultation question and answer information to obtain current medical history information; Extracting sentences based on the text position of the diagnosis description word in the historical consultation question and answer information to obtain diagnosis information, and extracting sentences based on the text position of the medication description word in the historical consultation question and answer information to obtain medication information; The medical history information, the medical history information, the diagnosis information and the medication information are combined to obtain the medical history information.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images