CN118155817A - Department and expert recommendation method and system based on GPT model - Google Patents

Department and expert recommendation method and system based on GPT model Download PDF

Info

Publication number
CN118155817A
CN118155817A CN202410510142.6A CN202410510142A CN118155817A CN 118155817 A CN118155817 A CN 118155817A CN 202410510142 A CN202410510142 A CN 202410510142A CN 118155817 A CN118155817 A CN 118155817A
Authority
CN
China
Prior art keywords
department
model
gpt
gpt model
user
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
Application number
CN202410510142.6A
Other languages
Chinese (zh)
Inventor
康飞若
徐业贵
程达
蔡凌豪
曾焕洁
叶仁杰
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Xuhui Excellence Health Information Technology Co ltd
Original Assignee
Xuhui Excellence Health Information Technology Co ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Xuhui Excellence Health Information Technology Co ltd filed Critical Xuhui Excellence Health Information Technology Co ltd
Priority to CN202410510142.6A priority Critical patent/CN118155817A/en
Publication of CN118155817A publication Critical patent/CN118155817A/en
Pending legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H40/00—ICT specially adapted for the management or administration of healthcare resources or facilities; ICT specially adapted for the management or operation of medical equipment or devices
    • G16H40/20—ICT specially adapted for the management or administration of healthcare resources or facilities; ICT specially adapted for the management or operation of medical equipment or devices for the management or administration of healthcare resources or facilities, e.g. managing hospital staff or surgery rooms
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
    • G06F16/33—Querying
    • G06F16/332—Query formulation
    • G06F16/3329—Natural language query formulation or dialogue systems
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
    • G06F16/36—Creation of semantic tools, e.g. ontology or thesauri
    • G06F16/367—Ontology
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90—Details of database functions independent of the retrieved data types
    • G06F16/95—Retrieval from the web
    • G06F16/953—Querying, e.g. by the use of web search engines
    • G06F16/9535—Search customisation based on user profiles and personalisation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/22—Matching criteria, e.g. proximity measures
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/23—Clustering techniques
    • G06F18/232—Non-hierarchical techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/088—Non-supervised learning, e.g. competitive learning
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H10/00—ICT specially adapted for the handling or processing of patient-related medical or healthcare data
    • G16H10/60—ICT specially adapted for the handling or processing of patient-related medical or healthcare data for patient-specific data, e.g. for electronic patient records
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
    • G16H50/70—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for mining of medical data, e.g. analysing previous cases of other patients

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Health & Medical Sciences (AREA)
  • General Engineering & Computer Science (AREA)
  • Databases & Information Systems (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • General Health & Medical Sciences (AREA)
  • Computational Linguistics (AREA)
  • Mathematical Physics (AREA)
  • Biomedical Technology (AREA)
  • Public Health (AREA)
  • Evolutionary Computation (AREA)
  • Medical Informatics (AREA)
  • Primary Health Care (AREA)
  • Epidemiology (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Evolutionary Biology (AREA)
  • Software Systems (AREA)
  • General Business, Economics & Management (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Business, Economics & Management (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Biophysics (AREA)
  • Pathology (AREA)
  • Human Computer Interaction (AREA)
  • Animal Behavior & Ethology (AREA)
  • Medical Treatment And Welfare Office Work (AREA)

Abstract

The invention discloses a department and expert recommendation method and system based on a GPT model, and relates to the field of medical interactive systems, wherein the method comprises the following steps: s1, the system receives first description information input by a user, and symptom characteristics in the first description information are extracted; s2, performing first correlation calculation by combining the symptom characteristics and the medical knowledge graph of a GPT model in the system, and determining a plurality of preliminary correlation departments; and S3, the GPT model receives second description information input by the user according to the preliminary related department options, performs second correlation degree calculation by combining the knowledge graph, determines the experts under each department, and pushes the experts to the user interaction interface to display a plurality of expert options. According to the invention, the medical GPT model is trained to enable the patient to understand the symptom description input by the patient, the possible disease range and the corresponding department can be deduced according to the description, and the accurate diagnosis guide of the expert is realized by the interactive question-answering mode of the user and the GPT model.

Description

Department and expert recommendation method and system based on GPT model
Technical Field
The invention belongs to the field of medical interaction systems, and particularly relates to a department and expert recommendation method based on a GPT model. Meanwhile, the invention also relates to a department and expert recommendation system based on the GPT model.
Background
With the refinement of modern disease classification and lack of general knowledge of patient medicine, many patients come to hospitals, facing the refined department of diagnosis, and it is not very clear exactly which department he should register to visit. The patient may be negatively emotionally affected by the delayed illness and the back-and-forth running due to the wrong number.
The current guiding service is mainly divided into manual consultation guiding, graphic mark guiding, informationized platform guiding and intelligent robot guiding, wherein the informationized platform guiding and intelligent robot guiding have the problems of poor reliability of medical resource information, low guiding accuracy and the like, and meanwhile, the guiding and intelligent platform guiding and intelligent robot mainly serve for providing guiding, introducing, health ventilating and teaching, consultation guiding and the like functions, and the practical application effect needs to be verified in the aspects of coping with complex scenes and the like.
Disclosure of Invention
The invention aims to solve the defects in the prior art, and provides a department and expert recommendation method and system based on a GPT model, wherein the recommendation method is used for carrying out preliminary diagnosis by basic information input by a user, judging a possibly set department, and then accurately calculating an expert suitable for a user disease through an interactive question-answering mode.
In order to achieve the above purpose, the present invention provides the following technical solutions:
A department and expert recommendation method based on a GPT model comprises the following steps:
S1, the system receives first description information input by a user, and symptom characteristics in the first description information are extracted;
S2, performing first correlation calculation by combining the symptom characteristics and the medical knowledge graph of a GPT model in the system, determining a plurality of preliminary correlation departments, and pushing the preliminary correlation departments to a user interaction interface to display a plurality of preliminary correlation department options;
And S3, the GPT model in the system receives second description information input by the user according to the preliminary related department options, performs second correlation calculation by combining the knowledge graph, determines the experts under each department, and pushes the experts to the user interaction interface to display a plurality of expert options.
Preferably, in step S2, the data information in the medical knowledge graph includes: disease symptoms, diagnosis, treatment regimens, department functions, and doctor's direction of expertise;
The medical knowledge graph construction steps are as follows:
s11, entity identification is carried out on the integrated data information;
s12, simultaneously improving the precision of entity identification and relation extraction by utilizing a multitasking mode;
s13, carrying out normalization processing to link the entities expressing the same meaning.
Preferably, in step S2, the training step of the GPT model includes:
S21, pre-training the GPT model by using unlabeled large-scale medical text data;
S22, adjusting and optimizing the pre-training model according to the specific application scene;
S23, evaluating the performance of the model after adjustment and optimization;
s24, accumulating feedback in the interactive question-answering process of the user and the GPT model based on actual use, and continuously iterating the GPT model performance.
Preferably, in step S21, the GPT model pretraining adopts unsupervised pretraining, including:
Given no tag sequence ；
And carrying out maximum training on likelihood values, wherein the expression is as follows:
Where L 1 (u) represents likelihood values, k is the size of the sliding window, P is the conditional probability, Is a parameter of the model.
Preferably, in step S22, the performing adjustment optimization on the pre-training model includes:
a sequence of input tokens is included for each instance, and a label is included;
the input token sequence obtains the output of the last token of the last layer through a GPT model;
which is then predicted by the linear output layer of the parameters.
Preferably, the outputting the most probable several department options through calculation in step S3 includes:
s31, acquiring information data of a certain department in a medical knowledge graph, user input information and relationship types;
s32, calculating the correlation degree of each department according to the information of the department and the information data input by the user; wherein, the correlation calculation formula is:
D×R1×R2/>；
Wherein f is the correlation degree between the user and the corresponding department; r 1 represents information data of a certain department in a medical knowledge graph, R 2 represents user input information, and D represents a relationship type, wherein the relationship type is similarity;
S33, selecting a plurality of departments with the correlation degree exceeding a threshold value to generate recommended option output.
Preferably, the degree of relativity judgment principle of each department is as follows:
If f (D, R 1,R2) =1; the information data representing that the user input information and a certain department in the medical knowledge graph are successfully associated through the similarity, the similarity is higher than 50%, and the corresponding department in the medical knowledge graph is the inferred related department;
If f (D, R 1,R2) =0: the information data representing that the user input information is not successfully associated with a certain department in the medical knowledge graph through the similarity, and the corresponding department in the medical knowledge graph is the non-involved department.
Preferably, in step S4, the GPT model trains the GPT model by using a language fuzzy C-means clustering algorithm for expert pushing, and the data points are divided into different clustering centers by assigning fuzzy membership degrees to each data point;
the language fuzzy C-means clustering algorithm comprises the following steps:
S41, initializing and defining the number of clusters which are finally expected to be obtained;
S42, calculating a clustering center, wherein the clustering center is a weighted average of data points, and the weight is represented by membership;
S43, updating the membership degree according to the current cluster center value;
S44 repeatedly calculating the cluster center and updating the membership degree according to the current cluster center value, and performing iterative updating.
Preferably, in step S43, the membership degree is updated according to the current cluster center value, specifically, for each data point i and each cluster k, the updated membership degree U (i, k) has a formula as follows:
where X (i) is the eigenvector of data point i, C (k) is the eigenvector of cluster center k, j represents the index of all clusters, and m is the blurring factor.
A department and expert recommendation system based on a GPT model, where the system is configured to implement the department and expert recommendation method based on the GPT model, and the method includes:
The receiving unit is used for receiving first description information input by a user;
the computing unit is internally provided with a GPT model module and is used for realizing preliminary related department computation and expert pushing computation;
The human-computer interaction unit is used for realizing interaction between the system and the user, and is electrically connected with the receiving unit;
The display unit is arranged in the man-machine interaction unit;
The storage unit is used for storing medical knowledge maps and information data related to departments and doctors;
The information calling unit is used for calling and consulting the medical knowledge graph, the department and doctor related information data.
The invention has the technical effects and advantages that: according to the department and expert recommendation method based on the GPT model, the medical GPT model is trained to enable the patient to understand the symptom description of patient input, the possible disease range and the corresponding department can be deduced according to the description, and the expert accurate diagnosis is realized by performing interactive question-answering mode through the user and the GPT model;
secondly, in the interactive question-answering mode of the user and the GPT model, the GPT model is trained by adopting a language fuzzy C-means clustering algorithm, so that the interaction specificity is reduced to a certain extent, and the interaction of non-professional terms and dialects can be realized.
Drawings
FIG. 1 is a flow chart of a department and expert recommendation method based on a GPT model of the invention;
FIG. 2 is a flowchart of a method for training a medical GPT model in an embodiment of the invention;
fig. 3 is a flowchart of a training algorithm for expert screening by the GPT model in an embodiment of the present invention.
Detailed Description
The present invention will be described in further detail with reference to specific embodiments in order to make the objects, technical solutions and advantages of the present invention more apparent. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the invention. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
The GPT (GENERATIVE PRE-Trained) model is a large-scale natural language generation model based on a transducer architecture, the core is the transducer architecture, and the GPT model is a deep learning model for processing sequence data and is suitable for natural language processing tasks. Long-range dependencies in a text sequence can be captured through a self-attention mechanism. The GPT model takes advantage of this architecture to pretrain a large-scale language model and then fine tune on specific downstream tasks to improve its performance and applicability.
The invention provides a department and expert recommendation method based on a GPT model, which comprises the following steps as shown in figure 1:
S1, the system receives first description information input by a user, and symptom characteristics in the first description information are extracted;
S2, performing first correlation calculation by combining the symptom characteristics and the medical knowledge graph of a GPT model in the system, determining a plurality of preliminary correlation departments, and pushing the preliminary correlation departments to a user interaction interface to display a plurality of preliminary correlation department options;
In step S2, the data information in the medical knowledge graph includes: disease symptoms, diagnosis, treatment regimens, department functions, and doctor's direction of expertise;
as shown in fig. 2, the medical knowledge graph construction steps are as follows:
s11, entity identification is carried out on the integrated data information;
s12, simultaneously improving the precision of entity identification and relation extraction by utilizing a multitasking mode;
s13, carrying out normalization processing to link the entities expressing the same meaning.
In step S2, the training step of the GPT model includes:
S21, pre-training the GPT model by using unlabeled large-scale medical text data;
in step S21, the GPT model pretraining adopts an unsupervised pretraining, specifically:
Given no tag sequence ；
Maximizing likelihood values and training, wherein the expression is as follows:
Where L 1 (u) represents likelihood values, k is the size of the sliding window, P is the conditional probability, Is a parameter of the model and is a parameter of the model,Representing a logarithmic function;
Illustratively, assuming the unlabeled sequence u= [1,2,3,4,5], the size of the sliding window k=2, the conditional probability P (x|y) represents the probability of the next element x given the last element y. The parameter θ of the model is a vector representing a logarithmic function.
The likelihood value L 1 (u) is used to evaluate the rationality of this sequence. Likelihood values are obtained by calculating the conditional probability product within each window. Specifically, for each window [ u i,ui +1] in the sequence u, we can calculate the conditional probability P (u i+1|ui) and then multiply the probabilities of all windows to get the likelihood value L 1 (u).
In accordance with the above-mentioned aspects of the present invention,
Windows [1,2]: p (2|1) =p (2) ×p (1|2) =p (2) ×exp (θ≡× [1,2 ])
Windows [2,3]: p (3|2) =p (3) ×p (2|3) =p (3) ×exp (θ≡× [2,3 ])
Windows [3,4]: p (4|3) =p (4) ×p (3|4) =p (4) ×exp (θ≡× [3,4 ])
Windows [4,5]: p (5|4) =p (5) ×p (4|5) =p (5) ×exp (θ≡× [4,5 ])
Multiplying the probabilities of all windows to obtain likelihood value L 1 (u):
L1(u)=P(2)×exp(θ^T×[1,2])×P(3)×exp(θ^T×[2,3])×P(4)×exp(θ^T×[3,4])×P(5)×exp(θ^T×[4,5])
Where P (x) is the probability distribution of x and θ is a parameter of the model. By adjusting the value of θ, the likelihood value L 1 (u) is maximized, thereby finding the best model parameters.
S22, adjusting and optimizing the pre-training model according to the specific application scene;
in step S22, performing adjustment optimization on the pre-training model includes:
S221, for each instance, a sequence x of input token is included 1 X m and a label y;
S222, inputting a token sequence to obtain the output of the last token of the last layer through a GPT model ；
S223, then predicting y through a linear output layer of a parameter W y;
The expression is: Where P is a conditional probability.
S23, evaluating the performance of the model after adjustment and optimization;
s24, accumulating feedback in the interactive question-answering process of the user and the GPT model based on actual use, and continuously iterating the GPT model performance.
S3, a GPT model in the system receives second description information input by a user according to preliminary related department options, performs second correlation calculation by combining a knowledge graph, determines experts under each department, and pushes the experts to a user interaction interface to display a plurality of expert options;
The recommending of the preliminary related department options in step S3 includes:
s31, acquiring information data of a certain department in a medical knowledge graph, user input information and relationship types;
s32, calculating the correlation degree of each department according to the information of the department and the information data input by the user; wherein, the correlation calculation formula is:
D×R1×R2/>；
Wherein f is the correlation degree between the user and the corresponding department; r 1 represents information data of a certain department in a medical knowledge graph, R 2 represents user input information, and D represents a relationship type, wherein the relationship type is similarity;
s33, selecting a plurality of departments with correlation degree exceeding a threshold value to generate recommended option output;
The judging principle of the relativity of each department is as follows:
If f (D, R 1,R2) =1; the information data representing that the user input information and a certain department in the medical knowledge graph are successfully associated through the similarity, the similarity is higher than 50%, and the corresponding department in the medical knowledge graph is the inferred related department;
If f (D, R 1,R2) =0: the information data representing that the user input information is not successfully associated with a certain department in the medical knowledge graph through the similarity, and the corresponding department in the medical knowledge graph is the non-involved department.
In step S3, the GPT model trains the GPT model by adopting a language fuzzy C-means clustering algorithm for expert pushing, and the data points are divided into different clustering centers by distributing fuzzy membership degree for each data point;
Referring to fig. 3, the language fuzzy C-means clustering algorithm specifically includes:
S41, initializing and defining the number of clusters which are finally expected to be obtained;
S42, calculating a clustering center, wherein the clustering center is a weighted average of data points, and the weight is represented by membership; the cluster center is a weighted average of the data points, where the weights are represented by membership; for each cluster k and each feature dimension d, the calculation formula of the cluster center C (k, d) is:
；
Wherein U (i, k) is the membership degree of the data point i belonging to the cluster k, X (i, d) is the value of the data point i in the characteristic dimension d, m is a fuzzy factor, and a real number larger than 1 is taken;
S43, updating the membership degree according to the current cluster center value; in step S43, the membership degree is updated according to the current cluster center value, specifically, for each data point i and each cluster k, the updated membership degree U (i, k) has a formula:
wherein X (i) is a feature vector of a data point i, C (k) is a feature vector of a cluster center k, j represents indexes of all clusters, and m is a fuzzy factor;
S44 repeatedly calculating the cluster center and updating the membership degree according to the current cluster center value, and performing iterative updating.
Note that the number of clusters k: defining the number of clusters that are ultimately desired, requires selection based on practical problems and experience.
Blur factor m: the fuzzy degree of the clustering is controlled, and a real number larger than 1 is taken. The larger the value, the more fuzzy the membership.
Specifically, training the GPT model by adopting a language fuzzy C-means clustering algorithm has the following advantages:
The possibility that the data points belong to a plurality of clusters is allowed, so that the GPT model after training is more flexible and adaptive under the condition of uncertainty.
Robustness to noise and outliers: the language fuzzy C-means clustering algorithm has certain robustness to noise and abnormal values. Since ambiguity is introduced, outliers do not have an excessive impact on the clustering results, but are partially assigned to clusters.
Interpretation of the clustering results: the language fuzzy C-means clustering algorithm provides the interpretation of the clustering result, and can quantitatively analyze whether each data point belongs to a certain cluster or not by outputting the membership degree of each data point to each cluster.
Algorithm flexibility: the language fuzzy C-means clustering algorithm can be customized and expanded according to application requirements. The value of the fuzzy factor m can be adjusted to control the fuzzy degree of the clusters, and the number of the clusters and other parameters can be adjusted to adapt to different data and problems.
The embodiment of the invention also provides a department and expert recommendation system based on the GPT model, which is used for realizing the department and expert recommendation method based on the GPT model, and comprises the following steps:
The receiving unit is used for receiving first description information input by a user;
the computing unit is internally provided with a GPT model module and is used for realizing preliminary related department computation and expert pushing computation;
The human-computer interaction unit is used for realizing interaction between the system and the user, and is electrically connected with the receiving unit;
The display unit is arranged in the man-machine interaction unit;
The storage unit is used for storing medical knowledge maps and information data related to departments and doctors;
The information calling unit is used for calling and consulting the medical knowledge graph, the department and doctor related information data.
In summary, the invention can understand the symptom description of patient input by training the medical GPT model, and can infer the possible disease range and the corresponding departments according to the description, and realize the accurate diagnosis guide of the expert by the interactive question-answering mode of the user and the GPT model.
Secondly, in the interactive question-answering mode of the user and the GPT model, the GPT model is trained by adopting a language fuzzy C-means clustering algorithm, so that the interaction specificity is reduced to a certain extent, and the interaction of non-professional terms and dialects can be realized.
Finally, it should be noted that: the foregoing description is only illustrative of the preferred embodiments of the present invention, and although the present invention has been described in detail with reference to the foregoing embodiments, it will be apparent to those skilled in the art that modifications may be made to the embodiments described, or equivalents may be substituted for elements thereof, and any modifications, equivalents, improvements or changes may be made without departing from the spirit and principles of the present invention.

Claims (10)

1. The department and expert recommendation method based on the GPT model is characterized by comprising the following steps:
S1, the system receives first description information input by a user, and symptom characteristics in the first description information are extracted;
S2, performing first correlation calculation by combining the symptom characteristics and the medical knowledge graph of a GPT model in the system, determining a plurality of preliminary correlation departments, and pushing the preliminary correlation departments to a user interaction interface to display a plurality of preliminary correlation department options;
And S3, the GPT model in the system receives second description information input by the user according to the preliminary related department options, performs second correlation calculation by combining the knowledge graph, determines the experts under each department, and pushes the experts to the user interaction interface to display a plurality of expert options.
2. The department and expert recommendation method based on the GPT model as claimed in claim 1, wherein in step S2, the data information in the medical knowledge graph includes: disease symptoms, diagnosis, treatment regimens, department functions, and doctor's direction of expertise;
The medical knowledge graph construction steps are as follows:
s11, entity identification is carried out on the integrated data information;
s12, simultaneously improving the precision of entity identification and relation extraction by utilizing a multitasking mode;
s13, carrying out normalization processing to link the entities expressing the same meaning.
3. The department and expert recommendation method based on the GPT model as claimed in claim 1, wherein in step S2, the step of training the GPT model comprises:
S21, pre-training the GPT model by using unlabeled large-scale medical text data;
S22, adjusting and optimizing the pre-training model according to the specific application scene;
S23, evaluating the performance of the model after adjustment and optimization;
s24, accumulating feedback in the interactive question-answering process of the user and the GPT model based on actual use, and continuously iterating the GPT model performance.
4. A department and expert recommendation method based on a GPT model according to claim 3, wherein in step S21, the GPT model pre-training is an unsupervised pre-training, comprising:
Given no tag sequence ；
And carrying out maximum training on likelihood values, wherein the expression is as follows:
Where L 1 (u) represents likelihood values, k is the size of the sliding window, P is the conditional probability, Is a parameter of the model.
5. The GPT model-based department and expert recommendation method of claim 4, wherein in step S22, said adjusting and optimizing the pre-training model comprises:
a sequence of input tokens is included for each instance, and a label is included;
the input token sequence obtains the output of the last token of the last layer through a GPT model;
which is then predicted by the linear output layer of the parameters.
6. The department and expert recommendation method based on the GPT model as claimed in claim 1, wherein in step S3, the preliminary related department option recommendation method comprises:
s31, acquiring information data of a certain department in a medical knowledge graph, user input information and relationship types;
s32, calculating the correlation degree of each department according to the information of the department and the information data input by the user; wherein, the correlation calculation formula is:
D×R1×R2/>；
Wherein f is the correlation degree between the user and the corresponding department; r 1 represents information data of a certain department in a medical knowledge graph, R 2 represents user input information, and D represents a relationship type, wherein the relationship type is similarity;
S33, selecting a plurality of departments with the correlation degree exceeding a threshold value to generate recommended option output.
7. The department and expert recommendation method based on the GPT model as claimed in claim 6, wherein the degree of correlation judgment principle of each department is:
If f (D, R 1,R2) =1; the information data representing that the user input information and a certain department in the medical knowledge graph are successfully associated through the similarity, the similarity is higher than 50%, and the corresponding department in the medical knowledge graph is the inferred related department;
If f (D, R 1,R2) =0: the information data representing that the user input information is not successfully associated with a certain department in the medical knowledge graph through the similarity, and the corresponding department in the medical knowledge graph is the non-involved department.
8. The department and expert recommendation method based on the GPT model according to claim 1, wherein in the step S3, the GPT model trains the GPT model by adopting a language fuzzy C-means clustering algorithm for expert pushing, and the data points are divided into different clustering centers by distributing fuzzy membership degree for each data point;
the language fuzzy C-means clustering algorithm comprises the following steps:
S41, initializing and defining the number of clusters which are finally expected to be obtained;
S42, calculating a clustering center, wherein the clustering center is a weighted average of data points, and the weight is represented by membership;
S43, updating the membership degree according to the current cluster center value;
S44 repeatedly calculating the cluster center and updating the membership degree according to the current cluster center value, and performing iterative updating.
9. The department and expert recommendation method based on the GPT model as claimed in claim 8, wherein in step S43, the membership degree is updated according to the current cluster center value, specifically, for each data point i and each cluster k, the updated membership degree U (i, k) has a formula:
where X (i) is the eigenvector of data point i, C (k) is the eigenvector of cluster center k, j represents the index of all clusters, and m is the blurring factor.
10. A department and expert recommendation system based on a GPT model, wherein the system is configured to implement the department and expert recommendation method based on a GPT model according to any one of claims 1 to 9, and comprises:
The receiving unit is used for receiving first description information input by a user;
the computing unit is internally provided with a GPT model module and is used for realizing preliminary related department computation and expert pushing computation;
The human-computer interaction unit is used for realizing interaction between the system and the user, and is electrically connected with the receiving unit;
The display unit is arranged in the man-machine interaction unit;
The storage unit is used for storing medical knowledge maps and information data related to departments and doctors;
The information calling unit is used for calling and consulting the medical knowledge graph, the department and doctor related information data.
CN202410510142.6A 2024-04-26 2024-04-26 Department and expert recommendation method and system based on GPT model Pending CN118155817A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202410510142.6A CN118155817A (en) 2024-04-26 2024-04-26 Department and expert recommendation method and system based on GPT model

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202410510142.6A CN118155817A (en) 2024-04-26 2024-04-26 Department and expert recommendation method and system based on GPT model

Publications (1)

Publication Number Publication Date
CN118155817A true CN118155817A (en) 2024-06-07

Family

ID=91285244

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202410510142.6A Pending CN118155817A (en) 2024-04-26 2024-04-26 Department and expert recommendation method and system based on GPT model

Country Status (1)

Country Link
CN (1) CN118155817A (en)

Citations (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20170344711A1 (en) * 2016-05-31 2017-11-30 Baidu Usa Llc System and method for processing medical queries using automatic question and answering diagnosis system
CN110489566A (en) * 2019-08-22 2019-11-22 上海软中信息系统咨询有限公司 A kind of hospital guide's method of intelligence hospital guide's service robot
CN111897967A (en) * 2020-07-06 2020-11-06 北京大学 Medical inquiry recommendation method based on knowledge graph and social media
CN113889249A (en) * 2021-10-26 2022-01-04 东北大学秦皇岛分校 GPT-2 model-based intelligent triage and medical question and answer method and device
CN114220528A (en) * 2021-12-28 2022-03-22 深圳科卫机器人科技有限公司 Hospital department recommendation method and device, computer equipment and storage medium
CN114611735A (en) * 2022-03-23 2022-06-10 康键信息技术（深圳）有限公司 Internet registration method, device, equipment and storage medium for hospitalizing
CN116483976A (en) * 2023-04-28 2023-07-25 平安科技（深圳）有限公司 Registration department recommendation method, device, equipment and storage medium
CN116741358A (en) * 2023-06-14 2023-09-12 平安科技（深圳）有限公司 Inquiry registration recommendation method, inquiry registration recommendation device, inquiry registration recommendation equipment and storage medium
CN116936058A (en) * 2023-09-14 2023-10-24 北京健康有益科技有限公司 Intelligent diagnosis guiding method and system based on deep learning and knowledge graph
CN117079798A (en) * 2023-07-05 2023-11-17 上海深至信息科技有限公司 Intelligent pre-consultation and rapid diagnosis and treatment system and method based on large language model

Patent Citations (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20170344711A1 (en) * 2016-05-31 2017-11-30 Baidu Usa Llc System and method for processing medical queries using automatic question and answering diagnosis system
CN110489566A (en) * 2019-08-22 2019-11-22 上海软中信息系统咨询有限公司 A kind of hospital guide's method of intelligence hospital guide's service robot
CN111897967A (en) * 2020-07-06 2020-11-06 北京大学 Medical inquiry recommendation method based on knowledge graph and social media
CN113889249A (en) * 2021-10-26 2022-01-04 东北大学秦皇岛分校 GPT-2 model-based intelligent triage and medical question and answer method and device
CN114220528A (en) * 2021-12-28 2022-03-22 深圳科卫机器人科技有限公司 Hospital department recommendation method and device, computer equipment and storage medium
CN114611735A (en) * 2022-03-23 2022-06-10 康键信息技术（深圳）有限公司 Internet registration method, device, equipment and storage medium for hospitalizing
CN116483976A (en) * 2023-04-28 2023-07-25 平安科技（深圳）有限公司 Registration department recommendation method, device, equipment and storage medium
CN116741358A (en) * 2023-06-14 2023-09-12 平安科技（深圳）有限公司 Inquiry registration recommendation method, inquiry registration recommendation device, inquiry registration recommendation equipment and storage medium
CN117079798A (en) * 2023-07-05 2023-11-17 上海深至信息科技有限公司 Intelligent pre-consultation and rapid diagnosis and treatment system and method based on large language model
CN116936058A (en) * 2023-09-14 2023-10-24 北京健康有益科技有限公司 Intelligent diagnosis guiding method and system based on deep learning and knowledge graph

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
宋培彦: "《术语计算与知识组织研究》", 30 June 2018, 科学技术文献出版社, pages: 105 *
戴亚平 等: "《决策支持系统中的经典与智能化方法》", 31 May 2022, 北京理工大学出版社, pages: 172 - 173 *
朱晨光: "《机器阅读理解：算法与实践》", 30 April 2020, 机械工业出版社, pages: 129 - 131 *

Similar Documents

Publication Publication Date Title
Lin et al. Diffusion models for time-series applications: a survey
CN114048331A (en) Knowledge graph recommendation method and system based on improved KGAT model
CN112257449B (en) Named entity recognition method and device, computer equipment and storage medium
CN108563624A (en) A kind of spatial term method based on deep learning
US20060129034A1 (en) Medical decision support systems utilizing gene expression and clinical information and method for use
CN112256866B (en) Text fine-grained emotion analysis algorithm based on deep learning
CN117198466B (en) Diet management method and system for kidney disease patients
CN112052318A (en) Semantic recognition method and device, computer equipment and storage medium
CN112668543B (en) Isolated word sign language recognition method based on hand model perception
Estevez-Velarde et al. AutoML strategy based on grammatical evolution: A case study about knowledge discovery from text
CN115579141A (en) Interpretable disease risk prediction model construction method and disease risk prediction device
CN116822579A (en) Disease classification ICD automatic coding method and device based on contrast learning
CN114783601A (en) Physiological data analysis method and device, electronic equipment and storage medium
CN116913459B (en) Medicine recommendation method and system based on deep convolution network control gate model
US11948387B2 (en) Optimized policy-based active learning for content detection
Ravikumar et al. Deep learning fundamentals
CN115171870A (en) Diagnosis guiding and prompting method and system based on m-BERT pre-training model
Al-Ars et al. A web/mobile decision support system to improve medical diagnosis using a combination of K-mean and fuzzy logic
CN114694841B (en) Adverse event risk prediction method based on patient electronic health record
CN116401364A (en) Language model training method, electronic device, storage medium and product
CN118155817A (en) Department and expert recommendation method and system based on GPT model
CN115964475A (en) Dialogue abstract generation method for medical inquiry
CN113780394B (en) Training method, device and equipment for strong classifier model
US20220172064A1 (en) Machine learning method and machine learning device for eliminating spurious correlation
CN115312158A (en) Metadata and image feature double-perception personalized federal learning method

Legal Events

Date Code Title Description
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination