WO2023029512A1 - Knowledge graph-based medical question answering method and apparatus, device and medium - Google Patents

Abstract

The present application relates to technology of artificial intelligence, and discloses a knowledge graph-based medical question answering method, comprising: performing entity relationship extraction on an original user query statement by using a preset entity relationship joint extraction model so as to obtain a relationship between entities; inputting the original user query statement into a preset intention recognition model for intention recognition to obtain an intention recognition result; performing scene classification on the original user query statement according to the relationship between entities and the intention recognition result to obtain a scene category corresponding to the original user query statement; and according to the scene category, indexing an answer corresponding to the original user query statement from a pre-constructed medical knowledge graph. In addition, the present application further relates to blockchain technology. The intention recognition result can be stored in a node of a blockchain. The present application also provides a knowledge graph-based medical question answering apparatus, an electronic device, and a storage medium. The present application can improve the accuracy of medical question answering.

Description

Method, device, equipment and medium for answering medical questions based on knowledge graph

This application claims the priority of the Chinese patent application submitted to the China Patent Office on August 30, 2021, with the application number 202111004877.4, and the title of the invention is "Medical Question Answering Method, Device, Equipment and Medium Based on Knowledge Graph", the entire content of which Incorporated in this application by reference.

technical field

The present application relates to the technical field of artificial intelligence, and in particular to a method, device, electronic device and computer-readable storage medium for answering medical questions based on knowledge graphs.

Background technique

The rapid development of the Internet has greatly reduced the threshold for users to access the Internet, and a large number of medical-related search needs have emerged. In response to these search needs, medical intelligent question-and-answer services have been derived. Medical intelligent question and answer refers to the automatic search, processing, and processing of the user's medical questions to obtain answers that can answer the user's questions.

The inventor realized that the existing methods for answering medical questions are usually based on the information of a single entity searched by the user, but cannot provide effective and accurate retrieval results for multiple entities or search queries with specific intentions.

Contents of the invention

This application provides a method for answering medical questions based on knowledge graphs, including:

Obtain the original user query statement, and use the preset entity-relationship joint extraction model to perform entity-relationship extraction on the original user query statement to obtain entities and relationships between entities;

inputting the original user query sentence into a preset intent recognition model to perform intent recognition, and obtain an intent recognition result;

performing scene classification on the original user query statement according to the entity, the relationship between the entities, and the intent recognition result, to obtain the scene category corresponding to the original user query statement;

A pre-built medical knowledge map is obtained, and an answer corresponding to the original user query statement is indexed in the medical knowledge map according to the scene category.

The present application also provides a medical question answering device based on a knowledge graph, the device comprising:

The entity relationship extraction module is used to obtain the original user query statement, and use the preset entity relationship joint extraction model to perform entity relationship extraction on the original user query statement to obtain the entity and the relationship between the entities;

An intent recognition module, configured to input the original user query sentence into a preset intent recognition model for intent recognition, and obtain an intent recognition result;

A scene classification module, configured to classify the scene of the original user query statement according to the entity, the relationship between the entities and the intention recognition result, and obtain the scene category corresponding to the original user query statement;

An answer indexing module, configured to obtain a pre-built medical knowledge map, and index the answer corresponding to the original user query statement in the medical knowledge map according to the scene category.

The present application also provides an electronic device, the electronic device comprising:

at least one processor; and,

a memory communicatively coupled to the at least one processor; wherein,

The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor, so that the at least one processor can perform the following knowledge map-based medical treatment Question answer method:

Obtain the original user query statement, utilize the preset entity relationship joint extraction model to carry out entity relationship extraction to the original user query statement, and obtain the relationship between entities and entities;

inputting the original user query sentence into a preset intent recognition model to perform intent recognition, and obtain an intent recognition result;

performing scene classification on the original user query statement according to the entity, the relationship between the entities, and the intent recognition result, to obtain the scene category corresponding to the original user query statement;

A pre-built medical knowledge map is obtained, and an answer corresponding to the original user query statement is indexed in the medical knowledge map according to the scene category.

The present application also provides a computer-readable storage medium, at least one computer program is stored in the computer-readable storage medium, and the at least one computer program is executed by a processor in an electronic device to realize the following knowledge graph-based Methods of answering medical questions:

Obtain the original user query statement, and use the preset entity-relationship joint extraction model to perform entity-relationship extraction on the original user query statement to obtain entities and relationships between entities;

inputting the original user query sentence into a preset intent recognition model to perform intent recognition, and obtain an intent recognition result;

performing scene classification on the original user query statement according to the entity, the relationship between the entities, and the intent recognition result, to obtain the scene category corresponding to the original user query statement;

A pre-built medical knowledge map is obtained, and an answer corresponding to the original user query statement is indexed in the medical knowledge map according to the scene category.

Description of drawings

FIG. 1 is a schematic flow diagram of a method for answering medical questions based on knowledge graphs provided by an embodiment of the present application;

FIG. 2 is a functional block diagram of a medical question answering device based on a knowledge map provided by an embodiment of the present application;

FIG. 3 is a schematic structural diagram of an electronic device implementing the knowledge graph-based medical question answering method provided by an embodiment of the present application.

The realization, functional features and advantages of the present application will be further described in conjunction with the embodiments and with reference to the accompanying drawings.

Detailed ways

It should be understood that the specific embodiments described here are only used to explain the present application, and are not intended to limit the present application.

An embodiment of the present application provides a method for answering medical questions based on a knowledge graph. The execution subject of the knowledge graph-based medical question answering method includes but is not limited to at least one of electronic devices such as a server and a terminal that can be configured to execute the method provided by the embodiment of the present application. In other words, the knowledge graph-based medical question answering method can be executed by software or hardware installed on a terminal device or a server device, and the software can be a block chain platform. The server includes, but is not limited to: a single server, a server cluster, a cloud server or a cloud server cluster, and the like. The server may be an independent server, or may provide cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery network (ContentDeliveryNetwork, CDN), and cloud servers for basic cloud computing services such as big data and artificial intelligence platforms.

Referring to FIG. 1 , it is a schematic flowchart of a method for answering medical questions based on a knowledge map provided by an embodiment of the present application. In this embodiment, the medical question answering method based on the knowledge map includes:

S1. Obtain an original user query statement, and use a preset entity-relationship joint extraction model to perform entity relationship extraction on the original user query statement to obtain entities and relationships between entities.

In the embodiment of the present application, the original user query sentence is a query sentence that the patient wants to inquire about medical problems. For example, the original user query sentence is: "upper respiratory tract infection", "difference between upper respiratory tract infection and cold" or " Can I take roxithromycin for upper respiratory tract infection?"

Specifically, using the preset entity-relationship joint extraction model to perform entity-relationship extraction on the original user query statement to obtain entities and relationships between entities, including:

encoding the original user query statement by using the shared encoding layer in the joint entity-relationship extraction model to obtain original encoded data;

Inputting the original coded data into the entity recognition module in the entity-relationship joint extraction model for entity recognition to obtain one or more entities;

The multiple entities are input into the relationship extraction module in the entity-relationship joint extraction model to obtain the relationship between the entities.

In detail, the entity-relationship joint extraction model includes a shared coding layer, an entity recognition module and a relationship extraction module, wherein the shared coding layer is an Embedding layer, and the Bert model can be used as the shared coding layer, that is, the Bert model As the Embedding layer, the entity recognition module is composed of Bi-LSTM and CRF layer, and the relationship extraction module is composed of fully connected layer and Sigmoid function.

Wherein, performing encoding processing on the original user query sentence can enhance the feature representation capability of the original user query sentence.

Further, the input of the original coded data into the entity recognition module in the entity-relationship joint extraction model is carried out for entity recognition, and one or more entities are obtained, including:

calculating the state value of the original coded data through an input gate in the entity recognition module;

calculating the activation value of the original coded data by using the forget gate in the entity recognition module;

calculating a state update value of the original encoded data according to the state value and the activation value;

using the output gate to calculate an initial text sequence corresponding to the state update value;

Input the initial text sequence into the fully connected layer to calculate the corresponding emission probability, and use the CRF layer to calculate the transition probability corresponding to the preset label;

The initial text sequence is marked according to the transition probability and the emission probability to obtain one or more entities.

In detail, the entity recognition module is composed of a Bi-LSTM and a CRF layer, wherein the Bi-LSTM (LongShort-TermMemory, bidirectional long-short-term memory network) is a time cyclic neural network, including: an input gate, a forgetting gate and the output gate.

In an optional embodiment, the calculation method of the state value includes:

Among them, _it represents the state value,

Represents the bias of the cell unit in the input gate, w _i represents the activation factor of the input gate, h _t-1 represents the peak value of the original encoded data at the time t-1 of the input gate, x _t represents the original encoded data at time t, b _i Indicates the weight of the cell unit in the input gate.

In an optional embodiment, the calculation method of the activation value includes:

where f _t represents the activation value,

Represents the bias of the cell unit in the forget gate, w _f represents the activation factor of the forget gate,

represents the peak value of the original coded data at time t-1 of the forget gate, x _t represents the original coded data input at time t, and b _f represents the weight of the cell unit in the forget gate.

In an optional embodiment, the calculation method of the status update value includes:

Among them, c _t represents the state update value, h _t-1 represents the peak value of the original encoded data at the time of input gate t-1,

Indicates the peak value of the original encoded data at the moment of forgetting gate t-1.

In an optional embodiment, the calculating the initial text sequence corresponding to the state update value using the output gate includes: calculating the initial text sequence using the following formula:

o _t ＝tan h(c _t )

Among them, o _t represents the initial text sequence, tan h represents the activation function of the output gate, and _ct represents the state update value.

Specifically, the multiple entities are input into the relationship extraction module in the entity-relationship joint extraction model to obtain the relationship between entities, and the original coded data, the preset label and the relative position of the tail entity The information is spliced and passed to the fully connected layer, and the probability of the starting position of the tail entity is calculated through the Sigmoid function, and finally the (head entity, relationship, tail entity) entity-relationship triplet is obtained through parsing.

S2. Input the original user query sentence into a preset intent recognition model to perform intent recognition, and obtain an intent recognition result.

In the embodiment of the present application, the input of the original user query sentence into a preset intent recognition model for intent recognition, and obtaining the intent recognition result include:

Using the input layer of the intent recognition model to vectorize the original user query statement to obtain a user query vector;

Using the convolution layer of the intention recognition model to perform convolution processing on the user query vector to obtain a convolution data set;

Inputting the convolution data set into the pooling layer and the fully connected layer in the intent recognition model to obtain classification results;

The classification result is marked with a preset intent recognition label to obtain an intent recognition result.

Wherein, in this solution, the intent recognition model may be a Text-CNN deep learning model.

In detail, the intent recognition model consists of four parts: an input layer, a convolutional layer, a pooling layer and a fully connected layer. The input layer needs to input a fixed-length text sequence, and the vectorization processing can use word vector tools such as word2vec, fastText or Glove, and can also use the Bert model for processing. The convolution layer generally includes a plurality of convolution kernels of different sizes, and the convolution kernel only performs one-dimensional sliding, that is, the width of the convolution kernel is equal to the dimension of the vector. The pooling layer uses Max-pool, which not only reduces the parameters of the intent recognition model, but also ensures that the input of a fixed-length fully connected layer is obtained from the output of the variable-length convolution layer. The role of the fully connected layer is a classifier. The original Text-CNN model uses a fully connected network with only one hidden layer, which is equivalent to inputting the features extracted by the convolution and pooling layers into an LR classifier for classification. .

For example, the intention recognition result is mainly divided into intention and non-intent. When the intention recognition result is intention, the recognized intention may be multiple intention types such as complications, related symptoms, recommended medicines, and whether it is possible or not.

S3. Perform scene classification on the original user query statement according to the entity, the relationship between the entities, and the intent recognition result, to obtain the scene category corresponding to the original user query statement.

In the embodiment of the present application, since one entity or multiple entities may be identified in the original user query statement, there may be a relationship or a specific intention between entities, according to the entity and the relationship between entities and the The intent recognition result performs scene classification on the original user query statement.

Specifically, performing scene classification on the original user query statement according to the entity, the relationship between the entities, and the intent recognition result, to obtain the scene category corresponding to the original user query statement, including:

Classifying the scene category corresponding to the original user query statement in which the relationship between the entities is no relationship and the intention recognition result is no intention as the first scene;

Classifying the scene category corresponding to the original user query statement in which the relationship between the entities is non-relationship and the intention recognition result is intentional as the second scene;

The scene category corresponding to the original user query statement in which the relationship between the entities is related and the intention recognition result is intention is classified as the third scene.

For example, the original user query sentence "upper respiratory tract infection" belongs to a single or multiple entities and has no relationship and no intention, so it is classified as the first scene, and the original user query sentence "difference between upper respiratory tract infection and cold" has no relationship and intention, so it is classified as the second scene, and the original user query sentence "can I take roxithromycin for upper respiratory tract infection" is related and intentional, so it is classified as the third scene.

In detail, perform scene classification on the original user query statement according to the entity, the relationship between entities, and the intent recognition result, and pay more attention to the semantic information of the user, including the entity and entity in the original user query statement relationship and retrieval intent.

S4. Obtain a pre-built medical knowledge graph, and index the answer corresponding to the original user query statement in the medical knowledge graph according to the scene category.

In the embodiment of the present application, the method of acquiring a pre-built medical knowledge graph further includes:

Obtaining medical related data, performing structured processing on the medical related data, to obtain structured data;

performing entity extraction on the structured data to obtain entity information, and performing relationship extraction on the structured data to obtain related relationships;

A plurality of triplets are constructed according to the entity information and the correlation relationship, and a medical knowledge map is obtained by using the multiple triplets.

In detail, the medical-related data includes a large amount of medical-related data, such as common disease names, corresponding disease symptoms, medicines for treatment, disease cases, related examinations and medication instructions, etc. Structuring the medical-related data means defining the medical-related data to obtain structured data.

For example, the medical-related data includes upper respiratory tract infection, cold, diabetes, roxithromycin, etc., upper respiratory tract infection, cold and diabetes are defined as diseases, and roxithromycin is defined as medicine.

Specifically, the entity information includes but is not limited to medical entities, medical attribute entities, etc., common medical entities, such as diseases, symptoms, medicines, treatments, inspections, etc., common medical attributes, such as overview, etiology, disease, Medical treatment, treatment, medication instructions, drug efficacy, etc. The relevant information includes common complications, typical symptoms, departments visited, recommended medicines, and related examinations.

Further, in the embodiment of the present application, multiple triples are constructed according to the entity information and the correlation relationship, and the medical knowledge map is obtained by using the multiple triples. The triplet is an information representation form of "entity+relationship=entity". For example, the symptom of a cold is a runny nose, and the triplet is expressed as "cold+symptom=runny nose". You can take roxithromycin for upper respiratory tract infection, expressed as "upper respiratory tract infection + drug = roxithromycin" in triplets.

In the embodiment of the present application, the medical knowledge map is constructed based on the medical related data, which can intuitively reflect the correlation between multiple entities in the medical knowledge map, and improve the efficiency of further analysis using the medical knowledge map . Using the medical knowledge map as the underlying data support for medical information retrieval can not only rely on the huge relational network of the medical knowledge map to retrieve more extensive and accurate medical information, but also effectively associate various related information to make the search results more comprehensive. .

Specifically, the answer corresponding to the original user query statement is indexed in the medical knowledge graph according to the scene category. For the three situations of the original user query statement, the returned situations are as follows: in the first scene, the user enters "upper respiratory tract infection", retrieves all entities and entity attributes within the relationship corresponding to all current entities, and sorts them by entity category Distinguish, such as including complications, symptoms, medicines, questions and answers, cases, video articles, etc. In the second scenario, the user enters "difference between upper respiratory tract infection and cold", and a comparison of the same attributes of the upper "respiratory tract infection" and "cold" entities is retrieved. In the first scene, the user enters "Can I take roxithromycin for upper respiratory tract infection", and queries the relationship between the disease "upper respiratory tract infection" and the drug "roxithromycin" in the medical knowledge map, and gives Recommended medicines for "Upper Respiratory Tract Infections".

In the embodiment of the present application, the entity relationship is extracted from the original user query statement by using the preset entity relationship joint extraction model to obtain the entity and the relationship between the entities. The entity relationship reflects semantic information, and the original user query statement is input into the preset Intent recognition is carried out in the intent recognition model, and the intent recognition result is obtained, the user’s intention is determined, and the accuracy of subsequent question answering is improved. The original user query is classified according to the entity, the relationship between entities, and the intent recognition result. The scenario category indexes the answers corresponding to the original user query statements in the medical knowledge graph. The medical knowledge in the medical knowledge graph is highly relevant, and indexing according to the scenario category can more accurately extract the corresponding answers to the medical questions. Answer. Therefore, the method for answering medical questions based on the knowledge map proposed in this application can solve the problem of low accuracy in answering medical questions.

As shown in FIG. 2 , it is a functional block diagram of a medical question answering device based on a knowledge map provided by an embodiment of the present application.

The medical question answering device 100 based on the knowledge graph described in this application can be installed in an electronic device. According to the realized functions, the knowledge map-based medical question answering device 100 may include an entity relationship extraction module 101 , an intent recognition module 102 , a scene classification module 103 and an answer indexing module 104 . The module described in this application can also be called a unit, which refers to a series of computer program segments that can be executed by the processor of the electronic device and can complete fixed functions, and are stored in the memory of the electronic device.

In this embodiment, the functions of each module/unit are as follows:

The entity relationship extraction module 101 is configured to obtain an original user query statement, and use a preset entity relationship joint extraction model to perform entity relationship extraction on the original user query statement to obtain entities and relationships between entities;

The intent recognition module 102 is configured to input the original user query sentence into a preset intent recognition model for intent recognition, and obtain an intent recognition result;

The scene classification module 103 is configured to classify the scene of the original user query statement according to the entity, the relationship between the entities and the intention recognition result, and obtain the scene category corresponding to the original user query statement;

The answer indexing module 104 is configured to obtain a pre-built medical knowledge map, and index the answer corresponding to the original user query statement in the medical knowledge map according to the scene category.

In detail, the specific implementation of each module of the knowledge graph-based medical question answering device 100 is as follows:

Step 1: Obtain the original user query statement, and use the preset entity-relationship joint extraction model to perform entity-relationship extraction on the original user query statement to obtain entities and relationships between entities.

In the embodiment of the present application, the original user query sentence is a query sentence that the patient wants to inquire about medical problems. For example, the original user query sentence is: "upper respiratory tract infection", "difference between upper respiratory tract infection and cold" or " Can I take roxithromycin for upper respiratory tract infection?"

Specifically, using the preset entity-relationship joint extraction model to perform entity-relationship extraction on the original user query statement to obtain entities and relationships between entities includes:

encoding the original user query statement by using the shared encoding layer in the joint entity-relationship extraction model to obtain original encoded data;

Inputting the original coded data into the entity recognition module in the entity-relationship joint extraction model for entity recognition to obtain one or more entities;

The multiple entities are input into the relationship extraction module in the entity-relationship joint extraction model to obtain the relationship between the entities.

In detail, the entity-relationship joint extraction model includes a shared coding layer, an entity recognition module and a relationship extraction module, wherein the shared coding layer is an Embedding layer, and the Bert model can be used as the shared coding layer, that is, the Bert model As the Embedding layer, the entity recognition module is composed of Bi-LSTM and CRF layer, and the relationship extraction module is composed of fully connected layer and Sigmoid function.

Wherein, performing encoding processing on the original user query sentence can enhance the feature representation capability of the original user query sentence.

Further, the input of the original coded data into the entity recognition module in the entity-relationship joint extraction model is carried out for entity recognition, and one or more entities are obtained, including:

calculating the state value of the original coded data through an input gate in the entity recognition module;

calculating the activation value of the original coded data by using the forget gate in the entity recognition module;

calculating a state update value of the original encoded data according to the state value and the activation value;

using the output gate to calculate an initial text sequence corresponding to the state update value;

Input the initial text sequence into the fully connected layer to calculate the corresponding emission probability, and use the CRF layer to calculate the transition probability corresponding to the preset label;

The initial text sequence is marked according to the transition probability and the emission probability to obtain one or more entities.

In detail, the entity recognition module is composed of a Bi-LSTM and a CRF layer, wherein the Bi-LSTM (LongShort-TermMemory, bidirectional long-short-term memory network) is a time cyclic neural network, including: an input gate, a forgetting gate and the output gate.

In an optional embodiment, the calculation method of the state value includes:

Among them, _it represents the state value,

Represents the bias of the cell unit in the input gate, w _i represents the activation factor of the input gate, h _t-1 represents the peak value of the original encoded data at the time t-1 of the input gate, x _t represents the original encoded data at time t, b _i Indicates the weight of the cell unit in the input gate.

In an optional embodiment, the calculation method of the activation value includes:

where f _t represents the activation value,

Represents the bias of the cell unit in the forget gate, w _f represents the activation factor of the forget gate,

represents the peak value of the original coded data at time t-1 of the forget gate, x _t represents the original coded data input at time t, and b _f represents the weight of the cell unit in the forget gate.

In an optional embodiment, the calculation method of the status update value includes:

Among them, c _t represents the state update value, h _t-1 represents the peak value of the original encoded data at the time of input gate t-1,

Indicates the peak value of the original encoded data at the moment of forgetting gate t-1.

In an optional embodiment, the calculating the initial text sequence corresponding to the state update value using the output gate includes: calculating the initial text sequence using the following formula:

o _t ＝tan h(c _t )

Among them, o _t represents the initial text sequence, tan h represents the activation function of the output gate, and _ct represents the state update value.

Specifically, the multiple entities are input into the relationship extraction module in the entity-relationship joint extraction model to obtain the relationship between entities, and the original coded data, the preset label and the relative position of the tail entity The information is spliced and passed to the fully connected layer, and the probability of the starting position of the tail entity is calculated through the Sigmoid function, and finally the (head entity, relationship, tail entity) entity-relationship triplet is obtained through parsing.

Step 2: Input the original user query sentence into a preset intent recognition model for intent recognition, and obtain an intent recognition result.

In the embodiment of the present application, the input of the original user query sentence into a preset intent recognition model for intent recognition, and obtaining the intent recognition result include:

Using the input layer of the intent recognition model to vectorize the original user query statement to obtain a user query vector;

Using the convolution layer of the intention recognition model to perform convolution processing on the user query vector to obtain a convolution data set;

Inputting the convolution data set into the pooling layer and the fully connected layer in the intent recognition model to obtain classification results;

The classification result is marked with a preset intent recognition label to obtain an intent recognition result.

Wherein, in this solution, the intent recognition model may be a Text-CNN deep learning model.

In detail, the intent recognition model consists of four parts: an input layer, a convolutional layer, a pooling layer and a fully connected layer. The input layer needs to input a fixed-length text sequence, and the vectorization processing can use word vector tools such as word2vec, fastText or Glove, and can also use the Bert model for processing. The convolution layer generally includes a plurality of convolution kernels of different sizes, and the convolution kernel only performs one-dimensional sliding, that is, the width of the convolution kernel is equal to the dimension of the vector. The pooling layer uses Max-pool, which not only reduces the parameters of the intent recognition model, but also ensures that the input of a fixed-length fully connected layer is obtained from the output of the variable-length convolution layer. The function of the fully connected layer is a classifier. The original Text-CNN model uses a fully connected network with only one hidden layer, which is equivalent to inputting the features extracted by the convolution and pooling layers into an LR classifier for classification. .

For example, the intention recognition result is mainly divided into intention and non-intent. When the intention recognition result is intention, the recognized intention may be multiple intention types such as complications, related symptoms, recommended medicines, and whether it is possible or not.

Step 3: Perform scene classification on the original user query statement according to the entity, the relationship between the entities, and the intent recognition result, to obtain the scene category corresponding to the original user query statement.

In the embodiment of the present application, since one entity or multiple entities may be identified in the original user query statement, there may be a relationship or a specific intention between entities, according to the entity and the relationship between entities and the The intent recognition result performs scene classification on the original user query statement.

Specifically, performing scene classification on the original user query statement according to the entity, the relationship between the entities, and the intent recognition result, to obtain the scene category corresponding to the original user query statement, including:

Classifying the scene category corresponding to the original user query statement in which the relationship between the entities is no relationship and the intention recognition result is no intention as the first scene;

Classifying the scene category corresponding to the original user query statement in which the relationship between the entities is non-relationship and the intention recognition result is intentional as the second scene;

The scene category corresponding to the original user query statement in which the relationship between the entities is related and the intention recognition result is intention is classified as the third scene.

For example, the original user query sentence "upper respiratory tract infection" belongs to a single or multiple entities and has no relationship and no intention, so it is classified as the first scene, and the original user query sentence "difference between upper respiratory tract infection and cold" has no relationship and intention, so it is classified as the second scene, and the original user query sentence "can I take roxithromycin for upper respiratory tract infection" is related and intentional, so it is classified as the third scene.

In detail, perform scene classification on the original user query statement according to the entity, the relationship between entities, and the intent recognition result, and pay more attention to the semantic information of the user, including the entity and entity in the original user query statement relationship and retrieval intent.

Step 4: Obtain a pre-built medical knowledge map, and index the answer corresponding to the original user query statement in the medical knowledge map according to the scene category.

In the embodiment of the present application, the method of acquiring a pre-built medical knowledge graph further includes:

Obtaining medical related data, performing structured processing on the medical related data, to obtain structured data;

performing entity extraction on the structured data to obtain entity information, and performing relationship extraction on the structured data to obtain related relationships;

A plurality of triplets are constructed according to the entity information and the correlation relationship, and a medical knowledge map is obtained by using the multiple triplets.

In detail, the medical-related data includes a large amount of medical-related data, such as common disease names, corresponding disease symptoms, medicines for treatment, disease cases, related examinations and medication instructions, etc. Structuring the medical-related data means defining the medical-related data to obtain structured data.

For example, the medical-related data includes upper respiratory tract infection, cold, diabetes, roxithromycin, etc., upper respiratory tract infection, cold and diabetes are defined as diseases, and roxithromycin is defined as medicine.

Specifically, the entity information includes but is not limited to medical entities, medical attribute entities, etc., common medical entities, such as diseases, symptoms, medicines, treatments, inspections, etc., common medical attributes, such as overview, etiology, disease, Medical treatment, treatment, medication instructions, drug efficacy, etc. The relevant information includes common complications, typical symptoms, departments visited, recommended medicines, and related examinations.

Further, in the embodiment of the present application, multiple triples are constructed according to the entity information and the correlation relationship, and the medical knowledge map is obtained by using the multiple triples. The triplet is an information representation form of "entity+relationship=entity". For example, the symptom of a cold is a runny nose, and the triplet is expressed as "cold+symptom=runny nose". You can take roxithromycin for upper respiratory tract infection, expressed as "upper respiratory tract infection + drug = roxithromycin" in triplets.

In the embodiment of the present application, the medical knowledge map is constructed based on the medical related data, which can intuitively reflect the correlation between multiple entities in the medical knowledge map, and improve the efficiency of further analysis using the medical knowledge map . Using the medical knowledge map as the underlying data support for medical information retrieval can not only rely on the huge relational network of the medical knowledge map to retrieve more extensive and accurate medical information, but also effectively associate various related information to make the search results more comprehensive. .

Specifically, the answer corresponding to the original user query statement is indexed in the medical knowledge graph according to the scene category. For the three situations of the original user query statement, the returned situations are as follows: in the first scene, the user enters "upper respiratory tract infection", retrieves all entities and entity attributes within the relationship corresponding to all current entities, and sorts them by entity category Distinguish, such as including complications, symptoms, medicines, questions and answers, cases, video articles, etc. In the second scenario, the user enters "difference between upper respiratory tract infection and cold", and a comparison of the same attributes of the upper "respiratory tract infection" and "cold" entities is retrieved. In the first scene, the user enters "Can I take roxithromycin for upper respiratory tract infection", and queries the relationship between the disease "upper respiratory tract infection" and the drug "roxithromycin" in the medical knowledge map, and gives Recommended medicines for "Upper Respiratory Tract Infections".

In the embodiment of the present application, the entity relationship is extracted from the original user query statement by using the preset entity relationship joint extraction model to obtain the entity and the relationship between the entities. The entity relationship reflects semantic information, and the original user query statement is input into the preset Intent recognition is carried out in the intent recognition model, and the intent recognition result is obtained, the user’s intention is determined, and the accuracy of subsequent question answering is improved. The original user query is classified according to the entity, the relationship between entities, and the intent recognition result. The scenario category indexes the answers corresponding to the original user query statements in the medical knowledge graph. The medical knowledge in the medical knowledge graph is highly relevant, and indexing according to the scenario category can more accurately extract the corresponding answers to the medical questions. Answer. Therefore, the medical question answering device based on the knowledge map proposed in this application can solve the problem of low accuracy in answering medical questions.

As shown in FIG. 3 , it is a schematic structural diagram of an electronic device implementing a knowledge graph-based medical question answering method provided by an embodiment of the present application.

The electronic device 1 may include a processor 10, a memory 11, a communication bus 12, and a communication interface 13, and may also include a computer program stored in the memory 11 and operable on the processor 10, such as a knowledge map based medical question answering program.

Wherein, the processor 10 may be composed of integrated circuits in some embodiments, for example, may be composed of a single packaged integrated circuit, or may be composed of multiple integrated circuits with the same function or different functions packaged, including one or Combinations of multiple central processing units (Central Processing unit, CPU), microprocessors, digital processing chips, graphics processors and various control chips, etc. The processor 10 is the control core (ControlUnit) of the electronic device, and utilizes various interfaces and lines to connect the various parts of the entire electronic device, by running or executing programs or modules stored in the memory 11 (for example, executing based on medical question answering program of knowledge graph, etc.), and call the data stored in the memory 11 to execute various functions of the electronic device and process data.

The memory 11 includes at least one type of readable storage medium, and the readable storage medium includes flash memory, mobile hard disk, multimedia card, card type memory (for example: SD or DX memory, etc.), magnetic memory, magnetic disk, optical disk, etc. , the computer-readable storage medium may be non-volatile or volatile. The storage 11 may be an internal storage unit of the electronic device in some embodiments, such as a mobile hard disk of the electronic device. The memory 11 may also be an external storage device of the electronic device in other embodiments, such as a plug-in mobile hard disk equipped on the electronic device, a smart memory card (SmartMediaCard, SMC), a secure digital (SecureDigital, SD) card, Flash card (FlashCard), etc. Further, the memory 11 may also include both an internal storage unit of the electronic device and an external storage device. The memory 11 can not only be used to store application software and various data installed in electronic devices, such as codes of medical question answering programs based on knowledge graphs, but can also be used to temporarily store data that has been output or will be output.

The communication bus 12 may be a peripheral component interconnect standard (PCI for short) bus or an extended industry standard architecture (EISA for short) bus or the like. The bus can be divided into address bus, data bus, control bus and so on. The bus is configured to realize connection and communication between the memory 11 and at least one processor 10 and the like.

The communication interface 13 is used for communication between the electronic device and other devices, including a network interface and a user interface. Optionally, the network interface may include a wired interface and/or a wireless interface (such as a WI-FI interface, a Bluetooth interface, etc.), which are generally used to establish a communication connection between the electronic device and other electronic devices. The user interface may be a display (Display) or an input unit (such as a keyboard (Keyboard)). Optionally, the user interface may also be a standard wired interface or a wireless interface. Optionally, in some embodiments, the display may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, an OLED (Organic Light-Emitting Diode, Organic Light Emitting Diode) touch device, and the like. Wherein, the display may also be properly referred to as a display screen or a display unit, and is used for displaying information processed in the electronic device and for displaying a visualized user interface.

FIG. 3 only shows an electronic device with components. Those skilled in the art can understand that the structure shown in FIG. 3 does not constitute a limitation to the electronic device 1, and may include fewer or more components, or combinations of certain components, or different arrangements of components.

For example, although not shown, the electronic device may also include a power supply (such as a battery) for supplying power to various components. Preferably, the power supply may be logically connected to the at least one processor 10 through a power management device, so that Realize functions such as charge management, discharge management, and power consumption management. The power supply may also include one or more DC or AC power supplies, recharging devices, power failure detection circuits, power converters or inverters, power status indicators and other arbitrary components. The electronic device may also include various sensors, a Bluetooth module, a Wi-Fi module, etc., which will not be described in detail here.

It should be understood that the embodiments are only for illustration, and are not limited by the structure in terms of the scope of the patent application.

The medical question answering program based on the knowledge map stored in the memory 11 of the electronic device 1 is a combination of multiple instructions, and when running in the processor 10, it can realize:

Obtain the original user query statement, and use the preset entity-relationship joint extraction model to perform entity-relationship extraction on the original user query statement to obtain entities and relationships between entities;

inputting the original user query sentence into a preset intent recognition model to perform intent recognition, and obtain an intent recognition result;

performing scene classification on the original user query statement according to the entity, the relationship between the entities, and the intent recognition result, to obtain the scene category corresponding to the original user query statement;

A pre-built medical knowledge map is obtained, and an answer corresponding to the original user query statement is indexed in the medical knowledge map according to the scene category.

Specifically, for the specific implementation method of the above instructions by the processor 10, reference may be made to the description of relevant steps in the corresponding embodiments in the drawings, and details are not repeated here.

Further, if the integrated modules/units of the electronic device 1 are realized in the form of software function units and sold or used as independent products, they can be stored in a computer-readable storage medium. The computer-readable storage medium may be volatile or non-volatile. For example, the computer-readable medium may include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a removable hard disk, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM, Read-Only Memory) ).

The present application also provides a computer-readable storage medium, the readable storage medium stores a computer program, and when the computer program is executed by a processor of an electronic device, it can realize:

Obtain the original user query statement, and use the preset entity-relationship joint extraction model to perform entity-relationship extraction on the original user query statement to obtain entities and relationships between entities;

inputting the original user query sentence into a preset intent recognition model to perform intent recognition, and obtain an intent recognition result;

performing scene classification on the original user query statement according to the entity, the relationship between the entities, and the intent recognition result, to obtain the scene category corresponding to the original user query statement;

A pre-built medical knowledge map is obtained, and an answer corresponding to the original user query statement is indexed in the medical knowledge map according to the scene category.

In the several embodiments provided in this application, it should be understood that the disclosed devices, devices and methods can be implemented in other ways. For example, the device embodiments described above are only illustrative. For example, the division of the modules is only a logical function division, and there may be other division methods in actual implementation.

The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical units, that is, they may be located in one place, or may be distributed to multiple network units. Part or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

In addition, each functional module in each embodiment of the present application may be integrated into one processing unit, each unit may exist separately physically, or two or more units may be integrated into one unit. The above-mentioned integrated units can be implemented in the form of hardware, or in the form of hardware plus software function modules.

It will be apparent to those skilled in the art that the present application is not limited to the details of the exemplary embodiments described above, but that the present application can be implemented in other specific forms without departing from the spirit or essential characteristics of the present application.

Therefore, the embodiments should be regarded as exemplary and not restrictive in all points of view, and the scope of the application is defined by the appended claims rather than the foregoing description, and it is intended that the scope of the present application be defined by the appended claims rather than by the foregoing description. All changes within the meaning and range of equivalents of the elements are embraced in this application. Any reference sign in a claim should not be construed as limiting the claim concerned.

The blockchain referred to in this application is a new application mode of computer technologies such as distributed data storage, point-to-point transmission, consensus mechanism, and encryption algorithm. Blockchain (Blockchain), essentially a decentralized database, is a series of data blocks associated with each other using cryptographic methods. Each data block contains a batch of network transaction information, which is used to verify its Validity of information (anti-counterfeiting) and generation of the next block. The blockchain can include the underlying platform of the blockchain, the platform product service layer, and the application service layer.

The embodiments of the present application may acquire and process relevant data based on artificial intelligence technology. Among them, artificial intelligence (AI) is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.

In addition, it is obvious that the word "comprising" does not exclude other elements or steps, and the singular does not exclude the plural. A plurality of units or devices stated in the system claims may also be realized by one unit or device through software or hardware. The terms first, second, etc. are used to denote names and do not imply any particular order.

Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application without limitation. Although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be Make modifications or equivalent replacements without departing from the spirit and scope of the technical solutions of the present application.

Claims (20)

1. A method for answering medical questions based on a knowledge map, wherein the method includes:

   Obtain the original user query statement, and use the preset entity-relationship joint extraction model to perform entity-relationship extraction on the original user query statement to obtain entities and relationships between entities;

   inputting the original user query sentence into a preset intent recognition model to perform intent recognition, and obtain an intent recognition result;

   performing scene classification on the original user query statement according to the entity, the relationship between the entities, and the intent recognition result, to obtain the scene category corresponding to the original user query statement;

   A pre-built medical knowledge map is obtained, and an answer corresponding to the original user query statement is indexed in the medical knowledge map according to the scene category.
2. The method for answering medical questions based on knowledge graphs according to claim 1, wherein said entity relationship extraction is performed on said original user query statement using a preset entity relationship joint extraction model to obtain entities and relationships between entities, include:

   encoding the original user query statement by using the shared encoding layer in the joint entity-relationship extraction model to obtain original encoded data;

   Inputting the original coded data into the entity recognition module in the entity-relationship joint extraction model for entity recognition to obtain one or more entities;

   The multiple entities are input into the relationship extraction module in the entity-relationship joint extraction model to obtain the relationship between the entities.
3. The method for answering medical questions based on knowledge graphs according to claim 2, wherein the input of the original coded data into the entity recognition module in the entity-relationship joint extraction model is carried out for entity recognition, and one or more entities, including:

   calculating the state value of the original coded data through an input gate in the entity recognition module;

   calculating the activation value of the original coded data by using the forget gate in the entity recognition module;

   calculating a state update value of the original encoded data according to the state value and the activation value;

   using the output gate to calculate an initial text sequence corresponding to the state update value;

   Input the initial text sequence into the fully connected layer to calculate the corresponding emission probability, and use the CRF layer to calculate the transition probability corresponding to the preset label;

   The initial text sequence is marked according to the transition probability and the emission probability to obtain one or more entities.
4. The medical question answering method based on knowledge graph as claimed in claim 3, wherein said calculating the state value of said original encoded data through said input gate in said entity recognition module comprises:

   The state value of the original coded data is calculated by using the following calculation formula:

   

   Among them, _it represents the state value,

   

   Indicates the bias of the cell unit in the input gate, w _i indicates the activation factor of the input gate, h _t-1 indicates the peak value of the original encoded data at the time t-1 of the input gate, x _t indicates the original encoded data at the time t, b _i Indicates the weight of the cell unit in the input gate.
5. The method for answering medical questions based on knowledge graphs according to claim 1, wherein said inputting said original user query sentence into a preset intent recognition model for intent recognition to obtain an intent recognition result includes:

   Using the input layer of the intent recognition model to vectorize the original user query statement to obtain a user query vector;

   Using the convolution layer of the intention recognition model to perform convolution processing on the user query vector to obtain a convolution data set;

   The convolution data set is input into the pooling layer and the fully connected layer in the intention recognition model to obtain classification results;

   The classification result is marked with a preset intent recognition label to obtain an intent recognition result.
6. The method for answering medical questions based on a knowledge graph according to claim 1, wherein said obtaining a pre-built medical knowledge graph comprises:

   Obtaining medical related data, performing structured processing on the medical related data, to obtain structured data;

   performing entity extraction on the structured data to obtain entity information, and performing relationship extraction on the structured data to obtain related relationships;

   A plurality of triplets are constructed according to the entity information and the correlation relationship, and a medical knowledge map is obtained by using the multiple triplets.
7. The method for answering medical questions based on knowledge graphs according to claim 1, wherein said original user query statement is classified according to the scene according to the entity, the relationship between the entities, and the intent recognition result, to obtain The scene category corresponding to the original user query statement includes:

   Classifying the scene category corresponding to the original user query statement in which the relationship between the entities is no relationship and the intention recognition result is no intention as the first scene;

   Classifying the scene category corresponding to the original user query statement in which the relationship between the entities is non-relationship and the intention recognition result is intentional as the second scene;

   The scene category corresponding to the original user query statement in which the relationship between the entities is related and the intention recognition result is intention is classified as the third scene.
8. A device for answering medical questions based on a knowledge map, wherein the device includes:

   The entity relationship extraction module is used to obtain the original user query statement, and use the preset entity relationship joint extraction model to perform entity relationship extraction on the original user query statement to obtain the entity and the relationship between the entities;

   An intent recognition module, configured to input the original user query sentence into a preset intent recognition model for intent recognition, and obtain an intent recognition result;

   A scene classification module, configured to classify the scene of the original user query statement according to the entity, the relationship between the entities and the intention recognition result, and obtain the scene category corresponding to the original user query statement;

   An answer indexing module, configured to obtain a pre-built medical knowledge map, and index the answer corresponding to the original user query statement in the medical knowledge map according to the scene category.
9. An electronic device, wherein the electronic device includes:

   at least one processor; and,

   a memory communicatively coupled to the at least one processor; wherein,

   The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor, so that the at least one processor can perform the following knowledge map-based medical treatment Question answer method:

   Obtain the original user query statement, and use the preset entity-relationship joint extraction model to perform entity-relationship extraction on the original user query statement to obtain entities and relationships between entities;

   inputting the original user query sentence into a preset intent recognition model to perform intent recognition, and obtain an intent recognition result;

   performing scene classification on the original user query statement according to the entity, the relationship between the entities, and the intent recognition result, to obtain the scene category corresponding to the original user query statement;

   A pre-built medical knowledge map is obtained, and an answer corresponding to the original user query statement is indexed in the medical knowledge map according to the scene category.
10. The electronic device according to claim 9, wherein said entity relationship extraction is performed on said original user query statement by using a preset entity relationship joint extraction model to obtain entities and relationships between entities, including:

    encoding the original user query statement by using the shared encoding layer in the joint entity-relationship extraction model to obtain original encoded data;

    Inputting the original coded data into the entity recognition module in the entity-relationship joint extraction model for entity recognition to obtain one or more entities;

    The multiple entities are input into the relationship extraction module in the entity-relationship joint extraction model to obtain the relationship between the entities.
11. The electronic device according to claim 10, wherein said inputting the original coded data into the entity recognition module in the entity-relationship joint extraction model for entity recognition to obtain one or more entities comprises:

    calculating the state value of the original coded data through an input gate in the entity recognition module;

    calculating the activation value of the original coded data by using the forget gate in the entity recognition module;

    calculating a state update value of the original encoded data according to the state value and the activation value;

    using the output gate to calculate an initial text sequence corresponding to the state update value;

    Input the initial text sequence into the fully connected layer to calculate the corresponding emission probability, and use the CRF layer to calculate the transition probability corresponding to the preset label;

    The initial text sequence is marked according to the transition probability and the emission probability to obtain one or more entities.
12. The electronic device according to claim 11, wherein said calculating the state value of said original coded data through said input gate in said entity recognition module comprises:

    The state value of the original coded data is calculated by using the following calculation formula:

    

    Among them, _it represents the state value,

    

    Indicates the bias of the cell unit in the input gate, w _i indicates the activation factor of the input gate, h _t-1 indicates the peak value of the original encoded data at the time t-1 of the input gate, x _t indicates the original encoded data at the time t, b _i Indicates the weight of the cell unit in the input gate.
13. The electronic device according to claim 9, wherein said inputting said original user query sentence into a preset intention recognition model to perform intention recognition, and obtaining an intention recognition result comprises:

    Using the input layer of the intent recognition model to vectorize the original user query statement to obtain a user query vector;

    Using the convolution layer of the intent recognition model to perform convolution processing on the user query vector to obtain a convolution data set;

    Inputting the convolution data set into the pooling layer and the fully connected layer in the intent recognition model to obtain classification results;

    The classification result is marked with a preset intent recognition label to obtain an intent recognition result.
14. The electronic device according to claim 9, wherein said acquiring a pre-built medical knowledge graph comprises:

    Obtaining medical related data, performing structured processing on the medical related data, to obtain structured data;

    performing entity extraction on the structured data to obtain entity information, and performing relationship extraction on the structured data to obtain related relationships;

    A plurality of triplets are constructed according to the entity information and the correlation relationship, and a medical knowledge map is obtained by using the multiple triplets.
15. A computer-readable storage medium, storing a computer program, wherein, when the computer program is executed by a processor, the following method for answering medical questions based on a knowledge map is realized:

    Obtain the original user query statement, and use the preset entity-relationship joint extraction model to perform entity-relationship extraction on the original user query statement to obtain entities and relationships between entities;

    inputting the original user query sentence into a preset intent recognition model to perform intent recognition, and obtain an intent recognition result;

    performing scene classification on the original user query statement according to the entity, the relationship between the entities, and the intent recognition result, to obtain the scene category corresponding to the original user query statement;

    Obtain a pre-built medical knowledge map, and index the answer corresponding to the original user query statement in the medical knowledge map according to the scene category.
16. The computer-readable storage medium according to claim 15, wherein said entity relationship extraction is performed on the original user query statement using a preset entity relationship joint extraction model to obtain entities and relationships between entities, comprising:

    encoding the original user query statement by using the shared encoding layer in the joint entity-relationship extraction model to obtain original encoded data;

    Inputting the original coded data into the entity recognition module in the entity-relationship joint extraction model for entity recognition to obtain one or more entities;

    The multiple entities are input into the relationship extraction module in the entity-relationship joint extraction model to obtain the relationship between the entities.
17. The computer-readable storage medium according to claim 16, wherein said inputting said original encoded data into an entity recognition module in said entity-relationship joint extraction model for entity recognition, and obtaining one or more entities, including :

    calculating the state value of the original coded data through an input gate in the entity recognition module;

    calculating the activation value of the original coded data by using the forget gate in the entity recognition module;

    calculating a state update value of the original encoded data according to the state value and the activation value;

    using the output gate to calculate an initial text sequence corresponding to the state update value;

    Input the initial text sequence into the fully connected layer to calculate the corresponding emission probability, and use the CRF layer to calculate the transition probability corresponding to the preset label;

    The initial text sequence is marked according to the transition probability and the emission probability to obtain one or more entities.
18. The computer-readable storage medium according to claim 17, wherein the calculating the state value of the original encoded data through the input gate in the entity recognition module comprises:

    The state value of the original coded data is calculated by using the following calculation formula:

    

    Among them, _it represents the state value,

    

    Indicates the bias of the cell unit in the input gate, w _i indicates the activation factor of the input gate, h _t-1 indicates the peak value of the original encoded data at the time t-1 of the input gate, x _t indicates the original encoded data at the time t, b _i Indicates the weight of the cell unit in the input gate.
19. The computer-readable storage medium according to claim 15, wherein said inputting said original user query sentence into a preset intent recognition model for intent recognition, and obtaining an intent recognition result comprises:

    Using the input layer of the intent recognition model to vectorize the original user query statement to obtain a user query vector;

    Using the convolution layer of the intention recognition model to perform convolution processing on the user query vector to obtain a convolution data set;

    Inputting the convolution data set into the pooling layer and the fully connected layer in the intent recognition model to obtain classification results;

    The classification result is marked with a preset intent recognition label to obtain an intent recognition result.
20. The computer-readable storage medium according to claim 15, wherein said acquiring a pre-built medical knowledge graph comprises:

    Obtaining medical related data, performing structured processing on the medical related data, to obtain structured data;

    performing entity extraction on the structured data to obtain entity information, and performing relationship extraction on the structured data to obtain related relationships;

    A plurality of triplets are constructed according to the entity information and the correlation relationship, and a medical knowledge map is obtained by using the multiple triplets.

Images