CN118116620A - Medical question answering method and device and electronic equipment - Google Patents
Medical question answering method and device and electronic equipment Download PDFInfo
- Publication number
- CN118116620A CN118116620A CN202410520623.5A CN202410520623A CN118116620A CN 118116620 A CN118116620 A CN 118116620A CN 202410520623 A CN202410520623 A CN 202410520623A CN 118116620 A CN118116620 A CN 118116620A
- Authority
- CN
- China
- Prior art keywords
- medical
- information
- consultation
- knowledge
- medical knowledge
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 61
- 238000012216 screening Methods 0.000 claims abstract description 32
- 238000004590 computer program Methods 0.000 claims description 8
- 238000000605 extraction Methods 0.000 claims description 7
- 230000002787 reinforcement Effects 0.000 claims description 7
- 238000012549 training Methods 0.000 claims description 6
- 239000000758 substrate Substances 0.000 claims 2
- 201000010099 disease Diseases 0.000 description 28
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 28
- 239000003814 drug Substances 0.000 description 26
- 208000024891 symptom Diseases 0.000 description 25
- 229940079593 drug Drugs 0.000 description 17
- 238000009223 counseling Methods 0.000 description 16
- 206010019233 Headaches Diseases 0.000 description 11
- 231100000869 headache Toxicity 0.000 description 11
- 230000008569 process Effects 0.000 description 10
- 206010037660 Pyrexia Diseases 0.000 description 7
- 238000010586 diagram Methods 0.000 description 7
- 238000012545 processing Methods 0.000 description 7
- 206010012601 diabetes mellitus Diseases 0.000 description 6
- 238000005516 engineering process Methods 0.000 description 6
- 206010039101 Rhinorrhoea Diseases 0.000 description 5
- 206010011224 Cough Diseases 0.000 description 4
- 208000036071 Rhinorrhea Diseases 0.000 description 4
- 201000009240 nasopharyngitis Diseases 0.000 description 4
- 230000004044 response Effects 0.000 description 4
- 206010008111 Cerebral haemorrhage Diseases 0.000 description 3
- 208000035977 Rare disease Diseases 0.000 description 3
- WDLWHQDACQUCJR-ZAMMOSSLSA-N (6r,7r)-7-[[(2r)-2-azaniumyl-2-(4-hydroxyphenyl)acetyl]amino]-8-oxo-3-[(e)-prop-1-enyl]-5-thia-1-azabicyclo[4.2.0]oct-2-ene-2-carboxylate Chemical compound C1([C@@H](N)C(=O)N[C@H]2[C@@H]3N(C2=O)C(=C(CS3)/C=C/C)C(O)=O)=CC=C(O)C=C1 WDLWHQDACQUCJR-ZAMMOSSLSA-N 0.000 description 2
- 208000035143 Bacterial infection Diseases 0.000 description 2
- 229930186147 Cephalosporin Natural products 0.000 description 2
- 206010057190 Respiratory tract infections Diseases 0.000 description 2
- 206010046306 Upper respiratory tract infection Diseases 0.000 description 2
- 208000036142 Viral infection Diseases 0.000 description 2
- 208000022362 bacterial infectious disease Diseases 0.000 description 2
- 230000006399 behavior Effects 0.000 description 2
- 230000005540 biological transmission Effects 0.000 description 2
- 229960002580 cefprozil Drugs 0.000 description 2
- 229940124587 cephalosporin Drugs 0.000 description 2
- 150000001780 cephalosporins Chemical class 0.000 description 2
- 229940124579 cold medicine Drugs 0.000 description 2
- 238000004891 communication Methods 0.000 description 2
- 238000003745 diagnosis Methods 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- OKTJSMMVPCPJKN-UHFFFAOYSA-N Carbon Chemical compound [C] OKTJSMMVPCPJKN-UHFFFAOYSA-N 0.000 description 1
- 206010009944 Colon cancer Diseases 0.000 description 1
- 208000001333 Colorectal Neoplasms Diseases 0.000 description 1
- 206010016059 Facial pain Diseases 0.000 description 1
- 208000002193 Pain Diseases 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- JFPVXVDWJQMJEE-IZRZKJBUSA-N cefuroxime Chemical compound N([C@@H]1C(N2C(=C(COC(N)=O)CS[C@@H]21)C(O)=O)=O)C(=O)\C(=N/OC)C1=CC=CO1 JFPVXVDWJQMJEE-IZRZKJBUSA-N 0.000 description 1
- 229960001668 cefuroxime Drugs 0.000 description 1
- 230000001413 cellular effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 229910021389 graphene Inorganic materials 0.000 description 1
- 208000010753 nasal discharge Diseases 0.000 description 1
- 238000003058 natural language processing Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 208000004371 toothache Diseases 0.000 description 1
- 230000009385 viral infection Effects 0.000 description 1
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
- Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation
Landscapes
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
One or more embodiments of the present specification provide a method, an apparatus, and an electronic device for answering a medical question, where the method includes: receiving a medical consultation problem of a user; extracting key medical information from the medical consultation problem; searching relevant medical knowledge from a medical knowledge graph based on the key medical information; according to the medical consultation problem, screening the medical knowledge searched from the medical knowledge graph to obtain the residual medical knowledge after screening; and inputting the medical knowledge and the medical consultation questions remained after screening into a dialogue model to obtain the question answers aiming at the medical consultation questions, which are generated by the dialogue model.
Description
Technical Field
One or more embodiments of the present disclosure relate to artificial intelligence technology, and more particularly, to a method, apparatus, and electronic device for medical question answering.
Background
In the field of medical technology that is rapidly developing today, intelligent dialog systems, which may be referred to as medical question-answering systems, have become a key tool for improving the quality and efficiency of medical services. The medical question answering system provides consultation services for patients by simulating human communication modes, and for example, different service contents such as disease diagnosis, treatment advice, medicine description and the like can be included.
In the related art, a medical question-answering system generally answers questions of a patient through a pre-trained dialogue model, but the answer effect of the dialogue model mainly depends on training data obtained in a pre-training stage. In practice, it has been found that the above-described manner of generating answers to questions of a patient via a dialogue model sometimes results in answers that are not sufficiently accurate and reliable.
Disclosure of Invention
In view of this, one or more embodiments of the present disclosure provide a medical question-answering method, apparatus, and electronic device to improve accuracy of answer answers to medical questions.
In order to achieve the above object, one or more embodiments of the present disclosure provide the following technical solutions:
According to a first aspect of one or more embodiments of the present specification, there is provided a method of medical question answering, the method comprising:
Receiving a medical consultation problem of a user;
Extracting key medical information from the medical consultation problem;
Searching relevant medical knowledge from a medical knowledge graph based on the key medical information;
according to the medical consultation problem, screening the medical knowledge searched from the medical knowledge graph to obtain the residual medical knowledge after screening;
And inputting the medical knowledge and the medical consultation questions remained after screening into a dialogue model to obtain the question answers aiming at the medical consultation questions, which are generated by the dialogue model.
According to a second aspect of one or more embodiments of the present specification, there is provided a medical question-answering system, comprising:
The problem receiving module is used for receiving medical consultation problems of users;
The information extraction module is used for extracting key medical information from the medical consultation problem;
The map searching module is used for searching relevant medical knowledge from medical knowledge maps based on the key medical information;
The information screening module is used for screening the medical knowledge searched from the medical knowledge graph according to the medical consultation problem to obtain the residual medical knowledge after screening;
And the answer generation module is used for inputting the medical knowledge and the medical consultation questions remained after screening into a dialogue model to obtain the question answers aiming at the medical consultation questions, which are generated by the dialogue model.
According to a third aspect of one or more embodiments of the present specification, there is provided an electronic device comprising:
A processor;
a memory for storing processor-executable instructions;
Wherein the processor implements the methods described in any of the embodiments of the present specification by executing the executable instructions.
According to a fourth aspect of one or more embodiments of the present description, there is provided a computer readable storage medium having stored thereon computer instructions which, when executed by a processor, implement a method as described in any of the embodiments of the present description.
According to a fifth aspect of one or more embodiments of the present description, there is provided a computer program product comprising a computer program which, when executed by a processor, implements the method of any of the embodiments of the present description.
According to the medical question answering method, the medical question answering device and the electronic equipment, related medical knowledge is searched from the medical knowledge graph based on the key medical information extracted from the medical consultation questions, and the medical knowledge and the medical consultation questions are combined to serve as input of a dialogue model, so that the dialogue model can acquire richer and more targeted information related to the consultation questions, and accordingly the answer of the questions generated by the model is more accurate; in addition, according to the method of the embodiment of the specification, further screening processing is performed on the searched result from the knowledge graph, so that the medical knowledge remained after screening is input into the dialogue model through the screening processing, invalid information input into the dialogue model can be reduced, and the accuracy of the question answer generated by the dialogue model is further improved.
Drawings
In order to more clearly illustrate the technical solutions of one or more embodiments of the present disclosure or related technologies, the following description will briefly describe the drawings that are required to be used in the embodiments or related technology descriptions, and it is apparent that the drawings in the following description are only some embodiments described in one or more embodiments of the present disclosure, and other drawings may be obtained according to these drawings without inventive effort to those of ordinary skill in the art.
FIG. 1 is a schematic diagram of some nodes and edges in a medical knowledge-graph, as provided by an exemplary embodiment.
Fig. 2 is a flow chart of a method of medical question answering provided in an exemplary embodiment.
Fig. 3 is a flow chart of a graph search provided by an exemplary embodiment.
Fig. 4 is a schematic diagram of a graph search provided by an exemplary embodiment.
Fig. 5 is a schematic diagram of a graph search provided by an exemplary embodiment.
Fig. 6 is a flow chart of another atlas search provided by an example embodiment.
Fig. 7 is a schematic diagram of an apparatus according to an exemplary embodiment.
Fig. 8 is a block diagram of a medical question-answering system provided by an exemplary embodiment.
Detailed Description
Reference will now be made in detail to exemplary embodiments, examples of which are illustrated in the accompanying drawings. When the following description refers to the accompanying drawings, the same numbers in different drawings refer to the same or similar elements, unless otherwise indicated. The implementations described in the following exemplary embodiments do not represent all implementations consistent with one or more embodiments of the present specification. Rather, they are merely examples of apparatus and methods consistent with aspects of one or more embodiments of the present description as detailed in the accompanying claims.
It should be noted that: in other embodiments, the steps of the corresponding method are not necessarily performed in the order shown and described in this specification. In some other embodiments, the method may include more or fewer steps than described in this specification. Furthermore, individual steps described in this specification, in other embodiments, may be described as being split into multiple steps; while various steps described in this specification may be combined into a single step in other embodiments.
The embodiment of the specification provides a medical question answering method which can be applied to simulating a human communication mode to provide various medical consultation services for patients so as to improve the quality and efficiency of the medical services.
In the following description of the medical question-answering method, a knowledge graph and a large language model will be used, and the following is briefly explained:
1) Knowledge graph:
Knowledge Graph (knowledgegraph) is a technique that organizes Knowledge into a network structure. The goal of the knowledge graph is to capture information and knowledge in the real world in a form that is easy for a computer to understand.
Wherein, the knowledge graph can comprise entities and edges. The entities exist in the form of nodes, and various relationships between the entities are represented by edges. In the knowledge graph of the medical field, the entity may be, for example: diseases, crowd, symptoms, medicines, etc. The relationship represented by the edges may be, for example: symptoms of the disease, medication of the disease, etc.
Referring to the example of fig. 1, fig. 1 illustrates some nodes and edges in a medical knowledge-graph.
As shown in fig. 1, some entities of the following types may be included:
{ disease } entity: "Cold";
{ drug } entity: "Ganmaoling" and "Cefprozil";
{ symptom } entity: cough, runny nose, fever;
{ disease cause } entity: "bacterial infection" and the like.
Also included are relationships between some entities of the following types:
For example, cold-cause of disease-viral infection;
Also for example, cold-symptoms of disease-fever.
The entities and edges described above may have their own properties. For example, the attribute of the "cold" entity may include "illness," i.e., meaning that the type of "cold" entity is illness.
Knowledge maps may be updated periodically to ensure timeliness and accuracy of knowledge, and the frequency of updating may depend on the specific field and requirements of knowledge map application. The knowledge graph of the embodiment can be continuously updated and perfected so as to ensure that the latest medical knowledge and research results are contained, and the accuracy of information in the knowledge graph is ensured as much as possible.
2) Large language model:
The large language model (Large Language Models, LLMs for short) is a complex algorithm built based on deep learning and natural language processing techniques that can understand, generate and translate natural language text.
In the present description embodiment, a pre-trained large language model may be utilized to extract information or generate dialogs.
3）SFT+RLHF：
Sft+ RLHF is an abbreviation that represents a machine learning method that combines supervised fine Tuning (Supervised Fine-Tuning, SFT) and human feedback-based reinforcement learning (Reinforcement Learning from Human Feedback, RLHF).
Among other things, supervised Fine Tuning (SFT) is a machine learning method in which models are trained under human provided labels or guidance. SFT typically involves fine tuning a pre-trained model to suit a particular task or domain. Reinforcement Learning (RLHF) based on human feedback is a reinforcement learning method in which a model adjusts its behavior according to human feedback. In RLHF, the human evaluates the model generated outputs and provides feedback telling the model which outputs are good and which are bad. These feedback are then used to train the model so that it is more likely to produce good output in the future.
In the present description embodiments, SFTs and RLHF may be used in combination, first by training the model through the SFT to follow a particular instruction or task, and then using RLHF to further adjust the behavior of the model, which may help the model better understand the intent of a human and generate a more natural, more human-intended response.
Fig. 2 is a flow chart of a method of medical question answering that may be applied to a medical question answering system, as provided by an exemplary embodiment. As shown in fig. 2, the method may include the following processes:
In step 200, a medical advice question for the user is received.
For example, a user may use a medical question-and-answer system to consult some medical questions, including, but not limited to: diagnosis of the condition, consultation of medication, symptoms of the disease, etc.
In this step, the medical question answering system may receive the medical consultation questions input by the user. The present embodiment is not limited to the input method of the medical consultation problem, and may be, for example, text input or voice input.
Illustratively, the medical consultation problem of the user may be as follows:
"what should me catch a cold? "
In step 202, critical medical information is extracted from the medical advice questions.
In this step, the medical question answering system can use information extraction techniques to extract critical medical information from medical consultation questions.
The key medical information may include: some information corresponding to "entities" in the knowledge-graph to be used later.
For example, in the medical knowledge graph to be used in the present embodiment, different types of entities including "diseases", "symptoms", and "medicines", etc., information about the diseases, symptoms, medicines, etc. may be extracted from the medical consultation problem as key medical information, so that it is convenient to search for related knowledge from the knowledge graph according to the key medical information in a subsequent step.
For example, for the medical consultation question mentioned in step 200, "what should i catch a cold? ", the following key medical information can be extracted: "disease: cold).
For another example, if the medical consultation problem of the user is "how do me get back after the neck is painful for a week? The key medical information from which it may be extracted may include: part: neck "," symptom: pain + last one week).
In addition, this step does not limit the way in which critical medical information is extracted from the medical advice problem. For example, extraction may be performed using a pre-trained large language model LLM.
By extracting key medical information from the medical consultation questions based on which the subsequent steps search for relevant knowledge from the knowledge graph based on the key medical information, interference of subsequent irrelevant information with the dialogue model used to generate the answer to the questions can be reduced.
In step 204, relevant medical knowledge is searched from a medical knowledge graph based on the key medical information.
The step can search relevant medical knowledge from the medical knowledge graph based on the extracted key medical information.
Two ways of searching for relevant medical knowledge are listed as follows:
In one example, referring to the flow illustrated in fig. 3, fig. 3 is a flowchart of a graph search provided in an exemplary embodiment, and may include the following processes:
in step 300, a counseling intent corresponding to the medical counseling problem is identified.
In this step, the counseling intention corresponding to the medical counseling problem of the user can be identified.
For example, if the medical consultation problem of the user is "what is i cold to eat? The consultation intention corresponding to the question may be: the user is consulting for medication.
As another example, if the medical consultation problem of the user is "what symptoms are caused by colorectal cancer? The consultation intention corresponding to the question may be: the user is consulting the symptoms of the disease.
I.e., different medical consultation questions, the consultation intentions of which may be different.
The embodiment is not limited to a specific way of identifying the counseling intents, for example, the same large language model LLM may be used, while identifying the counseling intents of the medical counseling problem, and extracting the key medical information. For another example, two models may be used, one for identifying the counseling intent of the medical counseling problem and the other for extracting the critical medical information of the medical counseling problem.
In step 302, medical knowledge associated with the critical medical information and the consultation intention is acquired from a medical knowledge graph.
In this step, when searching for relevant medical knowledge from the medical knowledge graph, the consultation intention obtained in the above step 300 and the key medical information obtained in the above step may be used to search for the information in combination of both aspects.
For example:
1) Medical consultation problem for user: "what should me catch a cold? ";
2) Key medical information extracted therefrom:
"disease: cold ";
3) Identifying the consultation intention corresponding to the obtained medical consultation problem:
"consultation medication".
And searching in the medical knowledge graph by combining the key medical information and the consultation intention. Please refer to the knowledge-graph illustrated in fig. 1:
First, entities related to critical medical information can be found: "Cold". Specifically, when searching, the type of the attribute can be found according to the attribute of each entity, namely "disease", and specifically the entity of cold.
Next, as can be seen from fig. 1, there are many sides of the physical "cold" connection, but not all knowledge of the side connection is required for the medical advice problem of the present example. At this time, according to the consultation intention, the user can take medicine in consultation, so that the side connected with the entity 'cold' can be searched for the side which represents the medicine taking of the disease, and the connected medicine entities 'cold medicine' and 'cephalosporin' can be correspondingly found. The medical knowledge that is finally found can be seen from the illustration of fig. 4.
In addition, in order to enrich the subsequently generated question replies, the answers may be finer when looking up medical knowledge. For example, as can also be seen in fig. 1, even if cold is present, the drugs for which different symptoms are applicable may be different. Based on this, in searching the medical knowledge graph, it is also possible to continue searching for symptom entities related to the drug entity connected to the entity "cold" to obtain information as shown in fig. 5, and it is understood that these symptoms are also symptoms related to the cold entity:
medicine: ganmaoling, cefuroxime;
symptoms: cough, runny nose, fever.
Relationship between drug and applicable symptoms: cough and nasal discharge, common cold, fever, and cephalos.
It should be noted that the entities, relationships, and the like as described above may not be realistic, but are merely schematic illustrations.
As described above, in the search flow of fig. 3, the consultation intention is recognized in advance, and then, when the knowledge graph is searched, only the medical knowledge related to the consultation intention is acquired from the knowledge graph in combination with the consultation intention.
In another example, it is also possible to use a method of searching for related medical knowledge from the medical knowledge graph based on the key medical information, i.e. not according to the intention of consultation when searching the graph. And then screening the search results by combining the consultation intention to obtain the finally required medical knowledge. Referring to the flow illustrated in fig. 6, fig. 6 is a flow chart of another graph searching provided in an exemplary embodiment, which may include the following processes:
in step 600, a counseling intent corresponding to the medical counseling problem is identified.
In this step, the counseling intention corresponding to the medical counseling problem of the user can be identified. The same medical consultation problem, the consultation intentions of which may be different.
The detailed processing may be combined with the foregoing embodiments, and will not be described in detail.
In step 602, an initial knowledge set associated with the critical medical information is obtained from a medical knowledge-graph.
In this step, when searching for relevant medical knowledge from the medical knowledge graph, the search may be performed only on the basis of the key medical information obtained in the previous step.
For example:
1) Medical consultation problem for user: "what should me catch a cold? ";
2) Key medical information extracted therefrom:
"disease: cold treating medicine "
In combination with the above-mentioned key medical information, the searching in the medical knowledge graph may be, for example, searching to obtain all medical knowledge associated with "cold" entities in fig. 1, including, for example:
Drug entity: "Ganmaoling" and "Cefprozil";
Symptomatic entity: cough, runny nose, fever;
The cause entity: "bacterial infection", "viral infection".
The above entities all have edge connection with the cold entity, namely have association relation. In addition to the above-mentioned entities, the relationships between the entities are also searched together, and will not be described in detail.
The set of information searched as described above may be the initial knowledge set.
In step 604, medical knowledge related to the consultation intent is selected from the initial knowledge set.
In this step, the medical knowledge related to the medical consultation intention can be selected by further screening from the initial knowledge set in combination with the consultation intention corresponding to the medical consultation problem identified in step 600.
For example, for a medical consultation problem of a user: "what should me catch a cold? ", the consultation intention corresponding to the identification is obtained by the identification: "consultation medication". Accordingly, the user can use medicine according to the consultation intention, and then the user can search the side which represents the relation of 'medicine for diseases' from the sides connected with the entity 'cold', correspondingly find the connected medicine entities 'cold medicine' and 'cephalosporin', and finally find medical knowledge as shown in the schematic diagram of fig. 4.
In other examples, the medical knowledge shown in fig. 5 may also be found to make the medical knowledge more abundant. The medical knowledge in fig. 5, although including "symptoms", is related to medicines, and different symptoms can take different medicines, so that the medical knowledge is also related to the consultation intention "consultation medication", and can be used as a basis for better answering the medical consultation questions of the user consultation medication.
As described above, in the search flow of fig. 6, the knowledge graph is searched based on the key medical information, and then the search result is screened in combination with the counseling intention, so as to finally obtain the required medical knowledge related to the counseling intention.
In step 206, according to the related information of the medical consultation problem, the medical knowledge searched from the medical knowledge graph is screened to obtain the rest medical knowledge after screening.
In still other examples, the search results obtained after the knowledge-graph search may include redundant information that, if entered into the dialogue model, would interfere with the generation of the response to the question. Therefore, before inputting the medical knowledge and the medical consultation questions into the dialogue model, the medical knowledge searched from the medical knowledge graph may be further filtered according to the related information of the medical consultation questions of the user, so as to input the medical knowledge remaining after the filtering into the dialogue model. After screening, the dialogue model is input, invalid information input for the dialogue model can be reduced, and the accuracy of the question answer generated by the dialogue model is further improved.
Wherein, the related information of the medical consultation problem refers to information related to the medical consultation problem, and for example, the related information may include but is not limited to: user information of the user who initiated the medical advice problem, key medical information extracted from the medical advice problem, and the like.
For example, it is assumed that the medical knowledge searched from the medical knowledge graph includes the graph contents of several tens of diseases, many of which are information of rare diseases, and the probability of belonging to the rare diseases is low in terms of the medical consultation problem of the user, and the information of the rare diseases can be filtered accordingly.
For another example, if the medical consultation problem of the user is "what the headache is," after searching the medical knowledge graph, it is found that the headache is present in the symptoms of the disease "cold" and the headache is also present in the symptoms of the disease "cerebral hemorrhage. But the probability of suffering from cerebral hemorrhage is relatively low according to the age of the user who initiates the medical consultation problem, so cerebral hemorrhage can be filtered out from the searching result of the atlas, and the subsequent dialogue model can be combined with the information related to cold to generate a problem answer.
As another example, suppose that "what would be a disease i have recently had a symptoms? By querying the medical knowledge graph, the symptom can be obtained for a plurality of diseases. At this time, a case database including data statistics of the incidence of various types of diseases and their corresponding symptoms over a recent period of time (e.g., 3 months) may be combined. In combination with the case database, it is possible to calculate which disease has the highest or higher incidence among the diseases having the symptom. Some unlikely diseases can then be screened accordingly and the remaining information sent to the dialogue model to generate a question answer.
In step 208, the medical knowledge and the medical consultation questions remaining after the screening are input into a dialogue model, and a question answer for the medical consultation questions generated by the dialogue model is obtained.
In this embodiment, the question answer for the medical consultation question may be generated by a dialogue model.
For example, the dialog model may be a pre-trained large language model LLM. The medical consultation questions of the user and the medical knowledge searched from the medical knowledge graph can be input into the large language model LLM, and the answers of the questions output by the large language model and aiming at the medical consultation questions can be obtained.
For example, the manner of inputting the medical knowledge and the medical consultation problem into the dialogue model may be as follows:
{ please refer to the following information answer dialog: symptoms of the disease common cold: headache, fever and runny nose; symptoms of upper respiratory tract infection are: headache, fever and sneeze;
medical consultation problem for patients: some headache }
That is, the medical consultation questions and the medical knowledge obtained by searching can be spliced together and then input into the dialogue model.
For example, the question answer generated by the dialogue model for the medical consultation question may be "you may be a common cold, or an upper respiratory tract infection, you may make further examination tests".
Further, when the medical knowledge and the medical consultation questions are input into the dialogue model, the medical consultation questions therein may include: the method comprises the steps of carrying out a conversation on a historical medical consultation problem before the medical consultation problem, wherein the historical medical consultation problem and the medical consultation problem belong to the same conversation.
For example, the patient user continues to have a certain period of time from the beginning of the use of the medical question-answering system to the end of the use of the medical question-answering system for a total of 3 conversations, i.e., a total of 3 medical consultation questions are consulted. It is assumed that when a user initiates a 3 rd medical consultation problem, key medical information may be extracted from the medical consultation problem according to the medical question answering method according to the embodiment of the present specification, and related medical knowledge may be searched from a medical knowledge graph based on the key medical information. When the medical knowledge obtained by searching is input into the dialogue model together with the medical consultation questions, the 3 rd medical consultation questions and the two previous medical consultation questions can be input together, so that the dialogue model can better combine the context to generate the question answers.
As described above, the medical knowledge searched from the medical knowledge graph and the medical consultation questions of the user are combined to form a rich context input, so that more detailed background information is provided for the dialogue model, the dialogue model can combine the medical consultation questions and generate the question answers for the medical consultation questions in combination with the medical knowledge searched from the medical knowledge graph, and the question answers generated by the dialogue model are more accurate.
In addition, the medical question answering method in the embodiment of the present disclosure dynamically integrates information in a knowledge graph during a dialogue process, and searches relevant medical knowledge from the medical knowledge graph according to key medical information extracted from medical consultation questions during each dialogue process, and then generates a question answer in combination with the medical knowledge. In this way, on one hand, the searched medical knowledge is related to the medical consultation questions of the dialogue, the pertinence is provided, the medical consultation questions are different, and the corresponding medical knowledge is also different, so that the answer of the questions generated by the dialogue model is more pertinence and more accurate; on the other hand, the information in the medical knowledge graph is rich, and the medical knowledge graph can be updated timely, so that the medical knowledge searched according to the information is newer and richer, and a real-time, accurate and professional medical knowledge background can be provided for the dialogue model, thereby being beneficial to more accurate question replies generated by the dialogue model.
In some examples, to make the generated question replies more natural and more specialized, the dialog model may employ LLM trained on sft+ RLHF through medical information.
For example, the LLM can be fine-tuned using the preprocessed annotated medical dialogue data, training the model to generate accurate, compliant medical advice and answers. And may invite a medical professional to evaluate the model generated output and provide feedback, train the model using expert feedback as a reward signal, and adjust parameters of the model through a reinforcement learning algorithm. Through the process, the LLM can gradually adapt to the professional requirements of medical dialogue, and the professionality and medical applicability of the question reply are improved.
Further, when searching for relevant medical knowledge from the medical knowledge graph based on the key medical information, it may include: searching from a medical knowledge graph based on the key medical information to obtain a plurality of first node information; selecting target node information corresponding to the key medical information from the plurality of first node information according to node similarity; and searching relevant medical knowledge from the medical knowledge graph according to the target node information.
Specifically, embedding of the extracted critical medical information may be compared with embedding of each node in the medical knowledge graph to obtain some candidates similar to the critical medical information, and these candidates are referred to as first node information. Then, for these first node information, further selection is made according to the node similarity, and the first node information corresponding to the key medical information satisfying the similarity condition (for example, may be the first node information having the highest similarity probability) is selected as the target node information. And finally, searching relevant medical knowledge from the medical knowledge graph according to the target node information.
As a specific example: assuming that the key medical information is "headache", the plurality of first node information searched from the medical knowledge graph includes "headache", "facial pain", "toothache", and the like through the matching comparison of embedding. These entities may be referred to as first node information. Then, the "headache" with the highest similarity can be selected as the target node information most similar to the key medical information "headache" according to the similarity. And finally, searching relevant medical knowledge in the medical knowledge graph according to the target node information, for example, searching some entity information with connecting edges with headache.
Through the processing, more accurate information can be searched from the medical knowledge graph based on the key medical information, redundant information is reduced, noise input to the dialogue model in the subsequent steps is less, and the dialogue model is more accurate in generating the question response.
In addition, the method of the embodiment of the present specification can be easily extended to a new medical field as long as a knowledge graph corresponding to the medical field is used. Specifically, according to the medical consultation problem of the user, the problem field corresponding to the medical consultation problem can be identified, and the medical knowledge graph corresponding to the problem field can be obtained. And searching relevant medical knowledge from the medical knowledge graph corresponding to the problem field based on the key medical information extracted from the medical consultation problem.
For example, the medical knowledge graph may be a generic graph, i.e. include generic medical knowledge; or a knowledge graph for a certain subdivided medical field, for example, a medical knowledge graph for diabetes, wherein the knowledge graph includes treatment knowledge related to diabetes. Then, if the user consults about a diabetes related problem, it can be recognized that the problem area corresponding to the medical consultation problem is a diabetes area. Further, the medical knowledge graph of diabetes can be obtained, and then the related medical knowledge can be searched from the medical knowledge graph based on the key medical information extracted from the medical consultation problem. And inputting the medical knowledge and the medical consultation questions into a dialogue model to obtain the diabetes question answers generated by the dialogue model.
According to the medical question and answer method, relevant medical knowledge is searched from the medical knowledge graph based on the key medical information extracted from the medical consultation questions, and the medical knowledge and the medical consultation questions are combined to serve as inputs of a dialogue model, so that the dialogue model can acquire richer and more targeted information related to the consultation questions, and the answer of the questions generated by the model is more accurate.
Fig. 7 is a schematic block diagram of an apparatus according to an exemplary embodiment. Referring to fig. 7, at the hardware level, the device includes a processor 702, an internal bus 704, a network interface 706, a memory 708, and a non-volatile storage 710, although other hardware required by the service is possible. One or more embodiments of the present description may be implemented in a software-based manner, such as by the processor 702 reading a corresponding computer program from the non-volatile storage 710 into the memory 708 and then running. Of course, in addition to software implementation, one or more embodiments of the present disclosure do not exclude other implementation manners, such as a logic device or a combination of software and hardware, etc., that is, the execution subject of the following processing flow is not limited to each logic unit, but may also be hardware or a logic device.
Referring to fig. 8, the medical question-answering system may be applied to the apparatus shown in fig. 7 to implement the technical solution of the present specification. Wherein, the medical question-answering system can include: a question receiving module 81, an information extracting module 82, a graph searching module 83, an information screening module 84, and an answer generating module 85.
The problem receiving module 81 is configured to receive a medical consultation problem of a user.
An information extraction module 82 is configured to extract critical medical information from the medical consultation questions.
The map searching module 83 is configured to search for relevant medical knowledge from the medical knowledge maps based on the key medical information.
The information screening module 84 is configured to perform screening processing on medical knowledge obtained by searching from the medical knowledge graph according to the medical consultation problem, so as to obtain residual medical knowledge after screening;
And the answer generation module 85 is configured to input the medical knowledge and the medical consultation questions remaining after the screening into a dialogue model, and obtain the question answers for the medical consultation questions generated by the dialogue model.
In one example, the information extraction module 82 is further configured to: and identifying the consultation intention corresponding to the medical consultation problem. The map search module 83, when used for searching for relevant medical knowledge from medical knowledge maps based on key medical information, includes: medical knowledge associated with the key medical information and the consultation intention is acquired from a medical knowledge graph.
In one example, the information extraction module 82 is further configured to: and identifying the consultation intention corresponding to the medical consultation problem. The map search module 83, when used for searching for relevant medical knowledge from medical knowledge maps based on key medical information, includes: acquiring an initial knowledge set associated with the key medical information from a medical knowledge graph; medical knowledge is selected from the initial knowledge set that is relevant to the consultation intent.
In one example, the answer generation module 85, when configured to input the medical knowledge and the medical advice question into a dialogue model to obtain a question answer for the medical advice question generated by the dialogue model, includes: inputting the medical knowledge and the medical consultation questions into a large language model serving as a dialogue model to obtain the answer to the medical consultation questions output by the large language model; the large language model is obtained by performing supervised micro-scale and reinforcement learning training based on human feedback through medical information.
In one example, the map searching module 83, when configured to search for relevant medical knowledge from a medical knowledge map based on the key medical information, further includes: searching from a medical knowledge graph based on the key medical information to obtain a plurality of first node information; selecting target node information corresponding to the key medical information from the plurality of first node information according to node similarity; and searching relevant medical knowledge from the medical knowledge graph according to the target node information.
In one example, the graph search module 83 is further configured to: identifying a problem field corresponding to the medical consultation problem according to the medical consultation problem; and acquiring a medical knowledge graph corresponding to the problem field.
The system, apparatus, module or unit set forth in the above embodiments may be implemented in particular by a computer chip or entity, or by a product having a certain function. A typical implementation device is a computer, which may be in the form of a personal computer, laptop computer, cellular telephone, camera phone, smart phone, personal digital assistant, media player, navigation device, email device, game console, tablet computer, wearable device, or a combination of any of these devices.
In a typical configuration, a computer includes one or more processors (CPUs), input/output interfaces, network interfaces, and memory.
The memory may include volatile memory in a computer-readable medium, random Access Memory (RAM) and/or nonvolatile memory, such as Read Only Memory (ROM) or flash memory (flash RAM). Memory is an example of computer-readable media.
Embodiments of the present description provide a computer-readable storage medium having stored thereon computer instructions which, when executed by a processor, implement a method as described in any of the embodiments of the present description.
Computer readable media, including both non-transitory and non-transitory, removable and non-removable media, may implement information storage by any method or technology. The information may be computer readable instructions, data structures, modules of a program, or other data. Examples of storage media for a computer include, but are not limited to, phase change memory (PRAM), static Random Access Memory (SRAM), dynamic Random Access Memory (DRAM), other types of Random Access Memory (RAM), read Only Memory (ROM), electrically Erasable Programmable Read Only Memory (EEPROM), flash memory or other memory technology, read only compact disc read only memory (CD-ROM), digital Versatile Discs (DVD) or other optical storage, magnetic cassettes, magnetic disk storage, quantum memory, graphene-based storage or other magnetic storage devices, or any other non-transmission medium, which can be used to store information that can be accessed by the computing device. Computer-readable media, as defined herein, does not include transitory computer-readable media (transmission media), such as modulated data signals and carrier waves.
Embodiments of the present specification also provide a computer program product comprising a computer program which, when executed by a processor, implements the method of any of the embodiments of the present specification.
It should also be noted that the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one … …" does not exclude the presence of other like elements in a process, method, article or apparatus that comprises the element.
The foregoing describes specific embodiments of the present disclosure. Other embodiments are within the scope of the following claims. In some cases, the actions or steps recited in the claims can be performed in a different order than in the embodiments and still achieve desirable results. In addition, the processes depicted in the accompanying figures do not necessarily require the particular order shown, or sequential order, to achieve desirable results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.
The terminology used in the one or more embodiments of the specification is for the purpose of describing particular embodiments only and is not intended to be limiting of the one or more embodiments of the specification. As used in this specification, one or more embodiments and the appended claims, the singular forms "a," "an," and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It should also be understood that the term "and/or" as used herein refers to and encompasses any or all possible combinations of one or more of the associated listed items.
It should be understood that although the terms first, second, third, etc. may be used in one or more embodiments of the present description to describe various information, these information should not be limited to these terms. These terms are only used to distinguish one type of information from another. For example, first information may also be referred to as second information, and similarly, second information may also be referred to as first information, without departing from the scope of one or more embodiments of the present description. The word "if" as used herein may be interpreted as "at … …" or "at … …" or "in response to a determination" depending on the context.
The foregoing description of the preferred embodiment(s) is (are) merely intended to illustrate the embodiment(s) of the present invention, and it is not intended to limit the embodiment(s) of the present invention to the particular embodiment(s) described.
Claims (10)
1. A method of medical question answering, the method comprising:
Receiving a medical consultation problem of a user;
Extracting key medical information from the medical consultation problem;
Searching relevant medical knowledge from a medical knowledge graph based on the key medical information;
according to the related information of the medical consultation problem, screening the medical knowledge searched from the medical knowledge graph to obtain the residual medical knowledge after screening;
And inputting the medical knowledge and the medical consultation questions remained after screening into a dialogue model to obtain the question answers aiming at the medical consultation questions, which are generated by the dialogue model.
2. The method of claim 1, wherein the step of determining the position of the substrate comprises,
The method further comprises the steps of: identifying the consultation intention corresponding to the medical consultation problem;
The searching for relevant medical knowledge from the medical knowledge graph based on the key medical information comprises the following steps:
medical knowledge associated with the key medical information and the consultation intention is acquired from a medical knowledge graph.
3. The method of claim 1, wherein the step of determining the position of the substrate comprises,
The method further comprises the steps of: identifying the consultation intention corresponding to the medical consultation problem;
The searching for relevant medical knowledge from the medical knowledge graph based on the key medical information comprises the following steps:
Acquiring an initial knowledge set associated with the key medical information from a medical knowledge graph;
medical knowledge is selected from the initial knowledge set that is relevant to the consultation intent.
4. The method of claim 1, wherein the inputting the medical knowledge remaining after the screening and the medical advice questions into a dialogue model to obtain the question replies to the medical advice questions generated by the dialogue model comprises:
Inputting the medical knowledge and the medical consultation questions remained after screening into a large language model serving as a dialogue model to obtain a question answer aiming at the medical consultation questions output by the large language model;
the large language model is obtained by performing supervised micro-scale and reinforcement learning training based on human feedback through medical information.
5. The method of claim 1, wherein searching for relevant medical knowledge from a medical knowledge graph based on the critical medical information comprises:
searching from a medical knowledge graph based on the key medical information to obtain a plurality of first node information;
selecting target node information corresponding to the key medical information from the plurality of first node information according to node similarity;
And searching relevant medical knowledge from the medical knowledge graph according to the target node information.
6. The method of claim 1, wherein prior to searching for relevant medical knowledge from a medical knowledge graph based on the critical medical information, the method further comprises:
Identifying a problem field corresponding to the medical consultation problem according to the medical consultation problem;
And acquiring a medical knowledge graph corresponding to the problem field.
7. A medical question-answering system, the system comprising:
The problem receiving module is used for receiving medical consultation problems of users;
The information extraction module is used for extracting key medical information from the medical consultation problem;
The map searching module is used for searching relevant medical knowledge from medical knowledge maps based on the key medical information;
The information screening module is used for screening the medical knowledge obtained by searching from the medical knowledge graph according to the related information of the medical consultation problem to obtain the residual medical knowledge after screening;
And the answer generation module is used for inputting the medical knowledge and the medical consultation questions remained after screening into a dialogue model to obtain the question answers aiming at the medical consultation questions, which are generated by the dialogue model.
8. An electronic device, comprising:
A processor;
a memory for storing processor-executable instructions;
Wherein the processor is configured to implement the method of any of claims 1-6 by executing the executable instructions.
9. A computer readable storage medium having stored thereon computer instructions which, when executed by a processor, implement the steps of the method according to any of claims 1-6.
10. A computer program product comprising a computer program, characterized in that the computer program, when executed by a processor, implements the steps of the method according to any one of claims 1 to 6.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410520623.5A CN118116620A (en) | 2024-04-28 | 2024-04-28 | Medical question answering method and device and electronic equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410520623.5A CN118116620A (en) | 2024-04-28 | 2024-04-28 | Medical question answering method and device and electronic equipment |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN118116620A true CN118116620A (en) | 2024-05-31 |
Family
ID=91217382
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202410520623.5A Pending CN118116620A (en) | 2024-04-28 | 2024-04-28 | Medical question answering method and device and electronic equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN118116620A (en) |
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111309877A (en) * | 2018-12-12 | 2020-06-19 | 北京文因互联科技有限公司 | Intelligent question-answering method and system based on knowledge graph |
| CN111949758A (en) * | 2019-05-16 | 2020-11-17 | 北大医疗信息技术有限公司 | Medical question and answer recommendation method, recommendation system and computer readable storage medium |
| CN112836058A (en) * | 2019-11-25 | 2021-05-25 | 北京搜狗科技发展有限公司 | Medical knowledge map establishing method and device and medical knowledge map inquiring method and device |
| CN113764112A (en) * | 2021-09-16 | 2021-12-07 | 山东大学第二医院 | Online medical question and answer method |
| WO2022041727A1 (en) * | 2020-08-28 | 2022-03-03 | 康键信息技术（深圳）有限公司 | Question and answer management method, apparatus, and device for medical inquiry system, and storage medium |
| CN114925681A (en) * | 2022-06-08 | 2022-08-19 | 工银科技有限公司 | Knowledge map question-answer entity linking method, device, equipment and medium |
| CN116030944A (en) * | 2022-12-26 | 2023-04-28 | 郑州中业科技股份有限公司 | Medical consultation information recommendation method and system |
| CN117077786A (en) * | 2023-07-07 | 2023-11-17 | 南京邮电大学 | Knowledge graph-based data knowledge dual-drive intelligent medical dialogue system and method |
| CN117473057A (en) * | 2023-10-23 | 2024-01-30 | 讯飞医疗科技股份有限公司 | Question-answering processing method, system, equipment and storage medium |
| CN117932022A (en) * | 2023-12-29 | 2024-04-26 | 绿盟科技集团股份有限公司 | Intelligent question-answering method and device, electronic equipment and storage medium |
-
2024
- 2024-04-28 CN CN202410520623.5A patent/CN118116620A/en active Pending
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111309877A (en) * | 2018-12-12 | 2020-06-19 | 北京文因互联科技有限公司 | Intelligent question-answering method and system based on knowledge graph |
| CN111949758A (en) * | 2019-05-16 | 2020-11-17 | 北大医疗信息技术有限公司 | Medical question and answer recommendation method, recommendation system and computer readable storage medium |
| CN112836058A (en) * | 2019-11-25 | 2021-05-25 | 北京搜狗科技发展有限公司 | Medical knowledge map establishing method and device and medical knowledge map inquiring method and device |
| WO2022041727A1 (en) * | 2020-08-28 | 2022-03-03 | 康键信息技术（深圳）有限公司 | Question and answer management method, apparatus, and device for medical inquiry system, and storage medium |
| CN113764112A (en) * | 2021-09-16 | 2021-12-07 | 山东大学第二医院 | Online medical question and answer method |
| CN114925681A (en) * | 2022-06-08 | 2022-08-19 | 工银科技有限公司 | Knowledge map question-answer entity linking method, device, equipment and medium |
| CN116030944A (en) * | 2022-12-26 | 2023-04-28 | 郑州中业科技股份有限公司 | Medical consultation information recommendation method and system |
| CN117077786A (en) * | 2023-07-07 | 2023-11-17 | 南京邮电大学 | Knowledge graph-based data knowledge dual-drive intelligent medical dialogue system and method |
| CN117473057A (en) * | 2023-10-23 | 2024-01-30 | 讯飞医疗科技股份有限公司 | Question-answering processing method, system, equipment and storage medium |
| CN117932022A (en) * | 2023-12-29 | 2024-04-26 | 绿盟科技集团股份有限公司 | Intelligent question-answering method and device, electronic equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Luo et al. | A critical review of state‐of‐the‐art chatbot designs and applications | |
| US20230105969A1 (en) | Computer-based system for applying machine learning models to healthcare related data | |
| US10847265B2 (en) | Systems and methods for responding to healthcare inquiries | |
| US12001964B2 (en) | Artificial intelligence advisory systems and methods for behavioral pattern matching and language generation | |
| WO2021208444A1 (en) | Method and apparatus for automatically generating electronic cases, a device, and a storage medium | |
| CN113488200B (en) | Intelligent inquiry method, intelligent inquiry device, computer equipment and storage medium | |
| Qian et al. | Computer audition for fighting the SARS-CoV-2 corona crisis—Introducing the multitask speech corpus for COVID-19 | |
| CN112052318A (en) | Semantic recognition method and device, computer equipment and storage medium | |
| US20200258600A1 (en) | Methods and systems for recommeding filters to apply to clinical trial search results using machine learning techniques | |
| CN112837772A (en) | Pre-inquiry case history generation method and device | |
| CN117216209A (en) | Ultrasonic examination report reading system based on large language model | |
| CN114999676A (en) | Method, system, apparatus and medium for automatically replying to medical consultation | |
| CN113722507B (en) | Hospitalization cost prediction method and device based on knowledge graph and computer equipment | |
| CN114925216A (en) | Health management method and device based on knowledge graph | |
| Liao et al. | Medical data inquiry using a question answering model | |
| CN111783473B (en) | Method and device for identifying best answer in medical question and answer and computer equipment | |
| CN117877660A (en) | Medical report acquisition method and system based on voice recognition | |
| CN116028613B (en) | General knowledge question answering method, system, computer device and storage medium | |
| van Bree | A critical perspective on neural mechanisms in cognitive neuroscience: towards unification | |
| CN117253629A (en) | Medical guide information pushing method, device, equipment, medium and computer program product | |
| CN111797626B (en) | Named entity recognition method and device | |
| Sarrouti et al. | Multi-task transfer learning with data augmentation for recognizing question entailment in the medical domain | |
| CN109599176B (en) | Method and device for recommending inquiry skills and online auxiliary diagnosis system | |
| CN118116620A (en) | Medical question answering method and device and electronic equipment | |
| Konam et al. | Abridge: A Mission Driven Approach to Machine Learning for Healthcare Conversation. |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination |