Physical examination knowledge graph construction method and device, computer device, and storage medium

Abstract

Embodiments of the present application relate to a physical examination knowledge graph construction method and device, computer equipment and a storage medium, comprising: extracting medical concept entities in existing medical documents to obtain a plurality of target medical concept entities, wherein the target medical concept entities include disease entities and feature entities; extracting a plurality of preset parameters between the disease entities and the feature entities; linking the plurality of target medical concept entities with medical concept entities in a first physical examination knowledge graph based on the plurality of preset parameters to obtain a second physical examination knowledge graph; performing data processing on physical examination knowledge in the second physical examination knowledge graph to obtain a third physical examination knowledge graph; and quality evaluating the third physical examination knowledge graph, and obtaining a target physical examination knowledge graph when the quality evaluation result meets a preset condition, thereby avoiding too much manual intervention in the physical examination knowledge graph construction process and reducing the construction cost.

Description

Technical Field

[0001] Embodiments of the present invention relate to the field of data processing, and in particular to a method, device, computer equipment, and storage medium for constructing a physical examination knowledge graph. Background Art

[0002] A health checkup is a comprehensive physical examination of the patient using medical methods and techniques to assess their early health status and identify potential illnesses and health risks. This includes basic tests across clinical departments, such as ultrasound, electrocardiogram, and radiology equipment, as well as laboratory tests involving blood and urine. Because health checkups focus on health rather than simply disease, they are also called preventive health checkups. Therefore, a precise and comprehensive health checkup knowledge graph can improve health checkup efficiency.

[0003] Existing methods for constructing physical examination knowledge graphs rely on rules to infer a reasoning knowledge graph from a basic knowledge graph, then merge the basic and reasoning knowledge graphs to form a new basic knowledge graph. However, this rule-based approach requires a lot of manual rule creation, is costly, and limits the scope of knowledge acquisition. Summary of the Invention

[0004] In view of this, in order to solve the technical problems of high cost and high limitations of the above-mentioned physical examination knowledge graph construction method, the embodiments of the present invention provide a physical examination knowledge graph construction method, device, computer equipment and storage medium.

[0005] In a first aspect, an embodiment of the present invention provides a method for constructing a physical examination knowledge graph, comprising:

[0006] Extracting medical concept entities from existing medical documents to obtain multiple target medical concept entities, wherein the target medical concept entities include disease entities and feature entities;

[0007] Extracting a plurality of preset parameters between the disease entity and the characteristic entity;

[0008] Linking the multiple target medical concept entities with the medical concept entities in the first physical examination knowledge graph based on the multiple preset parameters to obtain a second physical examination knowledge graph;

[0009] performing data processing on the physical examination knowledge in the second physical examination knowledge graph to obtain a third physical examination knowledge graph;

[0010] A quality assessment is performed on the third physical examination knowledge graph, and when the quality assessment result meets a preset condition, a target physical examination knowledge graph is obtained.

[0011] In one possible implementation, the method further includes:

[0012] Extracting the association relationship between the disease entity and the feature entity based on a pre-trained relational classification model;

[0013] Based on the slot filling method, attribute information of the feature entity is extracted.

[0014] In one possible implementation, the method further includes:

[0015] Adding preset characters to the medical information sentence in the existing medical document to obtain a target medical information sentence;

[0016] Inputting the target medical sentence into the relational classification model to obtain target vectors corresponding to disease entities and feature entities;

[0017] The association relationship between the disease entity and the feature entity is calculated based on the target vector.

[0018] In one possible implementation, the method further includes:

[0019] Based on the slot filling method, semantic recognition is performed on the medical information sentences in the existing medical documents to identify the attribute information of the feature entities.

[0020] In one possible implementation, the method further includes:

[0021] Based on the entity linking method, the multiple target medical concept entities are linked with the medical concept entities in the first physical examination knowledge graph, and the hierarchical relationship between the multiple target medical concept entities and the medical concept entities in the first physical examination knowledge graph is determined;

[0022] Based on the hierarchical relationship, the multiple target medical concept entities are linked to the corresponding medical concept entities in the first physical examination knowledge graph to obtain a second physical examination knowledge graph.

[0023] In one possible implementation, the method further includes:

[0024] Based on semantic similarity, selecting multiple candidate medical concept entities corresponding to the multiple target medical concept entities from the medical concept entities in the first physical examination knowledge graph;

[0025] The hierarchical relationships between the multiple target medical concept entities and the corresponding multiple candidate entities are predicted based on a pre-trained relational classification model.

[0026] In one possible implementation, the method further includes:

[0027] Using rule reasoning method to obtain implicit medical concept entities in the second physical examination knowledge graph;

[0028] Using a link prediction method to predict the association relationship between the implicit medical concept entities;

[0029] A slot filling method is used to supplement the missing attribute information of the implicit medical concept entity;

[0030] The second physical examination knowledge graph is updated based on the association relationship between the implicit medical concept entities and the missing attribute information of the implicit medical concept entities to obtain a third physical examination knowledge graph.

[0031] In a second aspect, an embodiment of the present invention provides a device for constructing a physical examination knowledge graph, comprising:

[0032] An extraction module is used to extract medical concept entities in existing medical documents to obtain multiple target medical concept entities, wherein the target medical concept entities include disease entities and feature entities;

[0033] The extraction module is further used to extract multiple preset parameters between the disease entity and the feature entity;

[0034] a linking module, configured to link the plurality of target medical concept entities with the medical concept entities in the first physical examination knowledge graph based on the plurality of preset parameters to obtain a second physical examination knowledge graph;

[0035] a processing module, configured to perform data processing on the physical examination knowledge in the second physical examination knowledge graph to obtain a third physical examination knowledge graph;

[0036] An evaluation module is used to perform quality evaluation on the third physical examination knowledge graph, and obtain a target physical examination knowledge graph when the quality evaluation result meets a preset condition.

[0037] In a third aspect, an embodiment of the present invention provides a computer device comprising: a processor and a memory, wherein the processor is configured to execute a physical examination knowledge graph construction program stored in the memory to implement the physical examination knowledge graph construction method described in the first aspect above.

[0038] In a fourth aspect, an embodiment of the present invention provides a storage medium, comprising: the storage medium stores one or more programs, and the one or more programs can be executed by one or more processors to implement the physical examination knowledge graph construction method described in the first aspect above.

[0039] The embodiment of the present invention provides a physical examination knowledge graph construction scheme, which extracts medical concept entities from existing medical documents to obtain multiple target medical concept entities, wherein the target medical concept entities include disease entities and feature entities; extracts multiple preset parameters between the disease entities and feature entities; links the multiple target medical concept entities with the medical concept entities in the first physical examination knowledge graph based on the multiple preset parameters to obtain a second physical examination knowledge graph; performs data processing on the physical examination knowledge in the second physical examination knowledge graph to obtain a third physical examination knowledge graph; performs quality assessment on the third physical examination knowledge graph, and obtains a target physical examination knowledge graph when the quality assessment result meets the preset conditions. Compared with the existing technology, which infers an inference knowledge graph from the basic knowledge graph based on rules, and then merges the basic knowledge graph and the inference knowledge graph to form a new basic knowledge graph; the rule-based approach requires more manual rule formulation, is more costly, and has a limited scope of knowledge acquisition. This scheme can avoid more manual intervention in the construction process of the physical examination knowledge graph and reduce the construction cost. The physical examination knowledge graph constructed by this scheme covers a wide range of knowledge and improves application efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 A schematic diagram of a process for constructing a physical examination knowledge graph provided by an embodiment of the present invention;

[0041] Figure 2 A schematic diagram of a process for constructing another physical examination knowledge graph provided by an embodiment of the present invention;

[0042] Figure 3 A schematic diagram of the structure of a physical examination knowledge graph construction device provided by an embodiment of the present invention;

[0043] Figure 4 A schematic structural diagram of a computer device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0044] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0045] To facilitate understanding of the embodiments of the present invention, specific embodiments will be further explained below with reference to the accompanying drawings. The embodiments do not limit the embodiments of the present invention.

[0046] Figure 1A flow chart of a method for constructing a physical examination knowledge graph provided by an embodiment of the present invention is shown as follows: Figure 1 As shown, the method specifically includes:

[0047] S11. Extract medical concept entities from existing medical documents to obtain multiple target medical concept entities, wherein the target medical concept entities include disease entities and feature entities.

[0048] In an embodiment of the present invention, medical concept entities in existing medical documents are first extracted, wherein a hybrid strategy of a BERT (Bidirectional Encoder Representations from Transformers) model, a CRF algorithm, and a dictionary back-marking can be used to extract medical concept entities in existing medical documents. The BERT model is a language representation model; the CRF algorithm is an undirected graph model, which is mainly used for sequence labeling such as word segmentation, part-of-speech tagging, and named entity recognition; existing medical documents may include but are not limited to unstructured medical documents such as physical examination reports, medical textbooks, and clinical records.

[0049] Furthermore, after extracting the medical concept entities, multiple target medical concept entities are obtained, wherein the target medical concept entities include disease entities and feature entities.

[0050] For example, if a physical examination data shows "lung cancer and high blood pressure (150 / 98 mmHG)", "lung cancer" is extracted as the disease entity; and "high blood pressure" is extracted as the feature entity.

[0051] S12. Extracting a plurality of preset parameters between the disease entity and the feature entity.

[0052] In an embodiment of the present invention, based on the disease entities and feature entities obtained above, multiple preset parameters between the disease entities and feature entities are further extracted, wherein the multiple preset parameters can be the association relationship between the disease entity and the feature entity, attribute information of the feature entity, etc.

[0053] S13. Based on the multiple preset parameters, the multiple target medical concept entities are linked with the medical concept entities in the first physical examination knowledge graph to obtain a second physical examination knowledge graph.

[0054] After extracting multiple preset parameters between disease entities and feature entities, based on the multiple preset parameters, the entity linking method can be used to link the target medical concept entity extracted from the document with the medical concept entity in the first physical examination knowledge graph, determine the hierarchical relationship (synonymous relationship, hierarchical relationship) between the medical concept entities, and fuse the medical concept entities according to the hierarchical relationship to obtain the second physical examination knowledge graph, where the first physical examination knowledge graph is the existing physical examination knowledge graph, which is obtained by integrating, disambiguating and updating the physical examination knowledge from other structured and third-party knowledge bases.

[0055] S14. Perform data processing on the physical examination knowledge in the second physical examination knowledge graph to obtain a third physical examination knowledge graph.

[0056] In an embodiment of the present invention, after obtaining the second physical examination knowledge graph, the second physical examination knowledge graph needs to be optimized, including but not limited to mining implicit information, inferring missing facts, discovering noise and defects in the data through inconsistency detection technology, and continuously improving and expanding the knowledge graph.

[0057] S15. Perform a quality assessment on the third physical examination knowledge graph. When the quality assessment result meets a preset condition, a target physical examination knowledge graph is obtained.

[0058] To conduct a quality assessment of the third physical examination knowledge graph, first, you can manually sample the entities, relationships, and attributes in the knowledge graph to evaluate the accuracy of the knowledge. Then, build a knowledge coverage test set based on the application to evaluate the knowledge coverage. Finally, apply the knowledge graph to the business and evaluate the quality of the knowledge graph based on the business results.

[0059] The embodiment of the present invention provides a method for constructing a physical examination knowledge graph, which extracts medical concept entities from existing medical documents to obtain multiple target medical concept entities, wherein the target medical concept entities include disease entities and feature entities; extracts multiple preset parameters between the disease entities and the feature entities; based on the multiple preset parameters, the multiple target medical concept entities are linked to the medical concept entities in the first physical examination knowledge graph to obtain a second physical examination knowledge graph; data processing is performed on the physical examination knowledge in the second physical examination knowledge graph to obtain a third physical examination knowledge graph; quality assessment is performed on the third physical examination knowledge graph, and when the quality assessment result meets the preset conditions, a target physical examination knowledge graph is obtained. Compared with the existing technology, the rule-based approach requires more manual rule formulation, is more costly, and has a limited scope of knowledge acquisition. The present method can avoid more manual intervention in the construction process of the physical examination knowledge graph and reduce the construction cost. The physical examination knowledge graph constructed by this solution covers a wide range of knowledge and improves application efficiency.

[0060] Figure 2 A flow chart of another method for constructing a physical examination knowledge graph provided by an embodiment of the present invention is shown as follows: Figure 2 As shown, the method specifically includes:

[0061] S21. Add preset characters to the medical information sentence in the existing medical document to obtain a target medical information sentence.

[0062] In an embodiment of the present invention, medical concept entities in existing medical documents are extracted to obtain disease entities and feature entities, and multiple preset parameters between the disease entities and the feature entities are further extracted, where the multiple preset parameters can be the association relationship between the disease entity and the feature entity, attribute information of the feature entity, etc.

[0063] Furthermore, the association relationship between disease entities and feature entities can be extracted using a pre-trained relational classification model (such as R-BERT). For example, "Severe hyperglycemia presents with typical "three more and one less" symptoms, namely, polydipsia, polyphagia, polyuria, and weight loss, which are more common in type 1 diabetes." The relationship between the disease entity "type 1 diabetes" and the symptoms (feature entities) "polydipsia", "polyphagia", "polyuria", and "weight loss" of the clinical manifestations of the disease is extracted.

[0064] Specifically, taking the R-BERT relational classification model as an example, preset characters are added to the input sentence to obtain the target medical information sentence, that is, it is processed into a form such as "[CLS] Severe hyperglycemia presents typical "three more and one less" symptoms, namely, $drinking more#, eating more, urinating more and losing weight, which is more common in $type 1 diabetes#." "[CLS]" is added before the sentence, and the "$" and "#" characters are added before and after the first and last entities respectively.

[0065] S22. Input the target medical sentence into the relational classification model to obtain target vectors corresponding to disease entities and feature entities.

[0066] Input the target medical sentence into the BERT model to obtain the following vector H ′ 0, H ′ 1. H ′ 2:

[0067] H ′ 0=W0tanh(H0)+b0

[0068]

[0069] Among them, H0 is the vector representation of the target medical sentence; H ′ 0 represents the vector obtained after the sentence vector passes through the feedforward neural network; H h and H t Character vector representation of the first and last entities respectively; H ′ 1 and H ′ 2 is the vector representation of the first entity and the last entity obtained by summing and averaging the character vector representations of the entity and passing them through the feedforward neural network; W0, W1, and W2 represent the weight matrices of the model calculation at different stages; i represents the position of the first character of the first entity in the sentence; j represents the position of the last character of the first entity in the sentence; k represents the position of the first character of the last entity in the sentence; m represents the position of the last character of the last entity in the sentence; b0, b1, and b2 represent the biases of the model calculation at different stages.

[0070] S23. Calculate the association relationship between the disease entity and the feature entity based on the target vector.

[0071] The three target vectors obtained above are concatenated and passed through the fully connected layer. They are classified by the softmatx function to obtain the association relationship p between the disease entity and the feature entity, as shown in formulas 1 and 2:

[0072] h ″ =W3tanh(H ′ 0+H ′ 1+H ′ 2)+b3 Formula 1

[0073] p=softmax(h ″ ) Formula 2

[0074] Among them, h ″ It is the vector representation of the hidden layer of the softmatx function; W3 represents the weight matrix of the current stage; b3 represents the bias of the current stage.

[0075] S24. Based on the slot filling method, semantic recognition is performed on the medical information sentences in the existing medical document to identify the attribute information of the feature entity.

[0076] In an embodiment of the present invention, a slot filling method can be used to perform semantic recognition on medical information sentences in existing medical documents and extract attribute information of feature entities.

[0077] For example, in "lung cancer and high blood pressure (150 / 98 mmHG)", the indicator attribute "150 / 98 mmHG" of the feature entity "high blood pressure" is extracted.

[0078] S25. Based on semantic similarity, select multiple candidate medical concept entities corresponding to the multiple target medical concept entities from the medical concept entities in the first physical examination knowledge graph.

[0079] S26. Predicting the hierarchical relationship between the multiple target medical concept entities and the corresponding multiple candidate entities based on the pre-trained relationship classification model.

[0080] The entity linking method links multiple target medical concept entities extracted from the document with the medical concept entities in the first physical examination knowledge graph to determine the hierarchical relationship (synonymous relationship, hierarchical relationship). First, several candidate medical entities that are most similar to each target medical concept entity to be linked are selected from the first physical examination knowledge graph through semantic similarity, and then the hierarchical relationship between the target medical concept entity to be linked and the candidate medical concept entity is predicted through the pre-trained relationship classification model. Among them, the first physical examination knowledge graph is an existing physical examination knowledge graph, which is obtained by integrating, disambiguating and updating the physical examination knowledge from other structured and third-party knowledge bases.

[0081] S27. Based on the hierarchical relationship, link the multiple target medical concept entities to the corresponding medical concept entities in the first physical examination knowledge graph to obtain a second physical examination knowledge graph.

[0082] The target medical concept entity is fused with the medical concept entity in the first physical examination knowledge graph according to the hierarchical relationship to obtain the second physical examination knowledge graph.

[0083] S28. Use rule reasoning method to obtain implicit medical concept entities in the second physical examination knowledge graph.

[0084] S29. Use a link prediction method to predict the association relationship between the implicit medical concept entities.

[0085] S210: Use a slot filling method to supplement the missing attribute information of the implicit medical concept entity.

[0086] S211. Update the second physical examination knowledge graph based on the association relationship between the implicit medical concept entities and the missing attribute information of the implicit medical concept entities to obtain a third physical examination knowledge graph.

[0087] The following is a unified description of S28 to S211:

[0088] In an embodiment of the present invention, after obtaining the second physical examination knowledge graph, the second physical examination knowledge graph needs to be optimized.

[0089] Specifically, knowledge reasoning technology is used to mine implicit information from the second physical examination knowledge graph, infer missing facts, obtain implicit medical concept entities in the data through ontology or rule reasoning technology, predict implicit associations between implicit medical concept entities using link prediction methods, and complete missing attribute information of implicit medical concept entities using slot filling methods; the second physical examination knowledge graph is updated based on the associations between implicit medical concept entities and the missing attribute information of implicit medical concept entities, and the noise and defects in the data can be discovered through inconsistency detection technology, and the knowledge graph is continuously improved and expanded to obtain a complete and updated third physical examination knowledge graph.

[0090] Furthermore, the quality of the third physical examination knowledge graph is evaluated. First, the entities, relationships, and attributes in the knowledge graph can be manually sampled to evaluate the accuracy of the knowledge. Then, a knowledge coverage test set can be built based on the application to evaluate the coverage of the knowledge. Finally, the knowledge graph can be applied to the business and the quality of the knowledge graph can be evaluated based on the business results.

[0091] The embodiment of the present invention provides a method for constructing a physical examination knowledge graph, which extracts medical concept entities from existing medical documents to obtain multiple target medical concept entities, wherein the target medical concept entities include disease entities and feature entities; extracts multiple preset parameters between the disease entities and the feature entities; links the multiple target medical concept entities with the medical concept entities in the first physical examination knowledge graph based on the multiple preset parameters to obtain a second physical examination knowledge graph; performs data processing on the physical examination knowledge in the second physical examination knowledge graph to obtain a third physical examination knowledge graph; performs quality assessment on the third physical examination knowledge graph, and obtains a target physical examination knowledge graph when the quality assessment result meets the preset conditions. This method uses a technology based on pre-trained deep learning to extract entities, relationships, and attributes, with low labor costs, a wide range of extracted knowledge, and high efficiency, and can improve the application efficiency of the physical examination knowledge graph.

[0092] Figure 3 A schematic diagram of a physical examination knowledge graph construction device provided by an embodiment of the present invention is shown in FIG. Figure 3 As shown, specifically including:

[0093] An extraction module 301 is used to extract medical concept entities from existing medical documents to obtain multiple target medical concept entities, wherein the target medical concept entities include disease entities and feature entities;

[0094] The extraction module 301 is further used to extract multiple preset parameters between the disease entity and the feature entity;

[0095] A linking module 302 is configured to link the plurality of target medical concept entities with the medical concept entities in the first physical examination knowledge graph based on the plurality of preset parameters to obtain a second physical examination knowledge graph;

[0096] A processing module 303 is configured to perform data processing on the physical examination knowledge in the second physical examination knowledge graph to obtain a third physical examination knowledge graph;

[0097] The evaluation module 304 is used to perform quality evaluation on the third physical examination knowledge graph, and obtain the target physical examination knowledge graph when the quality evaluation result meets the preset conditions.

[0098] In a possible implementation, the extraction module 301 is specifically configured to extract the association relationship between the disease entity and the feature entity based on a pre-trained relational classification model; and extract the attribute information of the feature entity based on a slot filling method.

[0099] In one possible embodiment, the extraction module 301 is also used to add preset characters to the medical information sentences in the existing medical document to obtain a target medical information sentence; input the target medical sentence into the relational classification model to obtain a target vector corresponding to the disease entity and the feature entity; and calculate the association relationship between the disease entity and the feature entity based on the target vector.

[0100] In a possible implementation, the extraction module 301 is further configured to perform semantic recognition on the medical information sentences in the existing medical document based on the slot filling method, and identify the attribute information of the feature entity.

[0101] In one possible implementation, the linking module 302 is specifically used to link the multiple target medical concept entities with the medical concept entities in the first physical examination knowledge graph based on the entity linking method, and determine the hierarchical relationship between the multiple target medical concept entities and the medical concept entities in the first physical examination knowledge graph; based on the hierarchical relationship, the multiple target medical concept entities are linked to the corresponding medical concept entities in the first physical examination knowledge graph to obtain a second physical examination knowledge graph.

[0102] In one possible embodiment, the link module 302 is also used to select multiple candidate medical concept entities corresponding to the multiple target medical concept entities from the medical concept entities in the first physical examination knowledge graph based on semantic similarity; and predict the hierarchical relationship between the multiple target medical concept entities and the corresponding multiple candidate entities based on a pre-trained relational classification model.

[0103] In one possible embodiment, the processing module 303 is specifically used to obtain the implicit medical concept entities existing in the second physical examination knowledge graph using a rule reasoning method; predict the association relationship between the implicit medical concept entities using a link prediction method; supplement the missing attribute information of the implicit medical concept entities using a slot filling method; update the second physical examination knowledge graph based on the association relationship between the implicit medical concept entities and the missing attribute information of the implicit medical concept entities to obtain a third physical examination knowledge graph.

[0104] The physical examination knowledge graph construction device provided in this embodiment can be as follows Figure 3 The physical examination knowledge graph construction device shown in can be executed as follows Figure 1-2 All steps of the physical examination knowledge graph construction method, and then realize Figure 1-2 For details, please refer to the technical effect of the physical examination knowledge graph construction method shown in the figure. Figure 1-2 For the sake of brevity, the relevant description will not be repeated here.

[0105] Figure 4A schematic diagram of the structure of a computer device provided in an embodiment of the present invention is provided. Figure 4 The computer device 400 shown includes: at least one processor 401, memory 402, at least one network interface 404 and other user interfaces 403. The various components in the computer device 400 are coupled together via a bus system 405. It is understood that the bus system 405 is used to achieve connection and communication between these components. In addition to including a data bus, the bus system 405 also includes a power bus, a control bus, and a status signal bus. However, for the sake of clarity, the bus system 405 is not described in detail. Figure 4 Various buses are labeled as bus system 405 .

[0106] The user interface 403 may include a display, a keyboard, or a pointing device (eg, a mouse, a trackball, a touchpad, or a touch screen).

[0107] It is understood that the memory 402 in the embodiment of the present invention can be a volatile memory or a non-volatile memory, or can include both volatile and non-volatile memories. Among them, the non-volatile memory can be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), or a flash memory. The volatile memory can be a random access memory (RAM), which is used as an external cache. By way of example and not limitation, many forms of RAM are available, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate synchronous DRAM (DDRSDRAM), enhanced synchronous DRAM (ESDRAM), synchronous link DRAM (SLDRAM), and direct RAM bus random access memory (DRRAM). The memory 402 described herein is intended to include, but is not limited to, these and any other suitable types of memory.

[0108] In some embodiments, the memory 402 stores the following elements, executable units or data structures, or a subset thereof, or an extended set thereof: an operating system 4021 and application programs 4022 .

[0109] The operating system 4021 includes various system programs, such as a framework layer, a core library layer, and a driver layer, for implementing various basic services and handling hardware-based tasks. Application programs 4022 include various application programs, such as a media player and a browser, for implementing various application services. Programs implementing the methods of the embodiments of the present invention may be included in application programs 4022.

[0110] In an embodiment of the present invention, by calling a program or instruction stored in the memory 402, specifically, a program or instruction stored in the application 4022, the processor 401 is configured to execute the method steps provided in each method embodiment, for example, including:

[0111] Medical concept entities in existing medical documents are extracted to obtain multiple target medical concept entities, wherein the target medical concept entities include disease entities and feature entities; multiple preset parameters between the disease entities and feature entities are extracted; based on the multiple preset parameters, the multiple target medical concept entities are linked to the medical concept entities in the first physical examination knowledge graph to obtain a second physical examination knowledge graph; data processing is performed on the physical examination knowledge in the second physical examination knowledge graph to obtain a third physical examination knowledge graph; quality assessment is performed on the third physical examination knowledge graph, and when the quality assessment result meets the preset conditions, the target physical examination knowledge graph is obtained.

[0112] In one possible implementation, the association relationship between the disease entity and the feature entity is extracted based on a pre-trained relational classification model; and the attribute information of the feature entity is extracted based on a slot filling method.

[0113] In one possible implementation, preset characters are added to the medical information sentence in the existing medical document to obtain a target medical information sentence; the target medical sentence is input into the relational classification model to obtain a target vector corresponding to the disease entity and the feature entity; and the association relationship between the disease entity and the feature entity is calculated based on the target vector.

[0114] In one possible implementation, based on the slot filling method, semantic recognition is performed on the medical information sentences in the existing medical document to identify the attribute information of the feature entity.

[0115] In one possible implementation, based on the entity linking method, the multiple target medical concept entities are linked to the medical concept entities in the first physical examination knowledge graph to determine the hierarchical relationship between the multiple target medical concept entities and the medical concept entities in the first physical examination knowledge graph; based on the hierarchical relationship, the multiple target medical concept entities are linked to the corresponding medical concept entities in the first physical examination knowledge graph to obtain a second physical examination knowledge graph.

[0116] In one possible embodiment, based on semantic similarity, multiple candidate medical concept entities corresponding to the multiple target medical concept entities are selected from the medical concept entities in the first physical examination knowledge graph; and the hierarchical relationship between the multiple target medical concept entities and the corresponding multiple candidate entities is predicted based on a pre-trained relational classification model.

[0117] In one possible implementation, a rule reasoning method is used to obtain the implicit medical concept entities present in the second physical examination knowledge graph; a link prediction method is used to predict the association relationship between the implicit medical concept entities; a slot filling method is used to supplement the missing attribute information of the implicit medical concept entities; and the second physical examination knowledge graph is updated based on the association relationship between the implicit medical concept entities and the missing attribute information of the implicit medical concept entities to obtain a third physical examination knowledge graph.

[0118] The methods disclosed in the above embodiments of the present invention can be applied to or implemented by processor 401. Processor 401 may be an integrated circuit chip with signal processing capabilities. During implementation, each step of the above method can be completed by hardware integrated logic circuits in processor 401 or by software instructions. The above processor 401 may be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. The methods, steps, and logic block diagrams disclosed in the embodiments of the present invention can be implemented or executed. The general-purpose processor may be a microprocessor or any conventional processor. The steps of the methods disclosed in conjunction with the embodiments of the present invention can be directly implemented and executed by a hardware decoding processor, or by a combination of hardware and software units in the decoding processor. The software units can be located in storage media well-known in the art, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, etc. The storage medium is located in the memory 402 , and the processor 401 reads the information in the memory 402 and completes the steps of the above method in combination with its hardware.

[0119] It is understood that the embodiments described herein may be implemented using hardware, software, firmware, middleware, microcode, or a combination thereof. For hardware implementation, the processing unit may be implemented in one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), general-purpose processors, controllers, microcontrollers, microprocessors, other electronic units for performing the functions described herein, or a combination thereof.

[0120] For software implementation, the technology described herein can be implemented by a unit that performs the functions described herein. The software code can be stored in a memory and executed by a processor. The memory can be implemented in the processor or outside the processor.

[0121] The computer device provided in this embodiment may be Figure 4 The computer device shown in , can execute Figure 1-2 All steps of the physical examination knowledge graph construction method, and then realize Figure 1-2 For details, please refer to the technical effect of the physical examination knowledge graph construction method shown in the figure. Figure 1-2 For the sake of brevity, the relevant description will not be repeated here.

[0122] An embodiment of the present invention further provides a storage medium (computer-readable storage medium). The storage medium stores one or more programs. The storage medium may include volatile memory, such as random access memory; the memory may also include non-volatile memory, such as read-only memory, flash memory, hard disk, or solid-state drive; and the memory may also include a combination of the aforementioned types of memory.

[0123] When one or more programs in the storage medium can be executed by one or more processors, the above-mentioned physical examination knowledge graph construction method executed on the computer device side can be implemented.

[0124] The processor is configured to execute a physical examination knowledge graph construction program stored in the memory to implement the following steps of a physical examination knowledge graph construction method executed on the computer device side:

[0125] Medical concept entities in existing medical documents are extracted to obtain multiple target medical concept entities, wherein the target medical concept entities include disease entities and feature entities; multiple preset parameters between the disease entities and feature entities are extracted; based on the multiple preset parameters, the multiple target medical concept entities are linked to the medical concept entities in the first physical examination knowledge graph to obtain a second physical examination knowledge graph; data processing is performed on the physical examination knowledge in the second physical examination knowledge graph to obtain a third physical examination knowledge graph; quality assessment is performed on the third physical examination knowledge graph, and when the quality assessment result meets the preset conditions, the target physical examination knowledge graph is obtained.

[0126] In one possible implementation, the association relationship between the disease entity and the feature entity is extracted based on a pre-trained relational classification model; and the attribute information of the feature entity is extracted based on a slot filling method.

[0127] In one possible implementation, preset characters are added to the medical information sentence in the existing medical document to obtain a target medical information sentence; the target medical sentence is input into the relational classification model to obtain a target vector corresponding to the disease entity and the feature entity; and the association relationship between the disease entity and the feature entity is calculated based on the target vector.

[0128] In one possible implementation, based on the slot filling method, semantic recognition is performed on the medical information sentences in the existing medical document to identify the attribute information of the feature entity.

[0129] In one possible implementation, based on the entity linking method, the multiple target medical concept entities are linked to the medical concept entities in the first physical examination knowledge graph to determine the hierarchical relationship between the multiple target medical concept entities and the medical concept entities in the first physical examination knowledge graph; based on the hierarchical relationship, the multiple target medical concept entities are linked to the corresponding medical concept entities in the first physical examination knowledge graph to obtain a second physical examination knowledge graph.

[0130] In one possible embodiment, based on semantic similarity, multiple candidate medical concept entities corresponding to the multiple target medical concept entities are selected from the medical concept entities in the first physical examination knowledge graph; and the hierarchical relationship between the multiple target medical concept entities and the corresponding multiple candidate entities is predicted based on a pre-trained relational classification model.

[0131] In one possible implementation, a rule reasoning method is used to obtain the implicit medical concept entities present in the second physical examination knowledge graph; a link prediction method is used to predict the association relationship between the implicit medical concept entities; a slot filling method is used to supplement the missing attribute information of the implicit medical concept entities; and the second physical examination knowledge graph is updated based on the association relationship between the implicit medical concept entities and the missing attribute information of the implicit medical concept entities to obtain a third physical examination knowledge graph.

[0132] Professionals should also be further aware that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the above description has generally described the components and steps of each example according to their functions. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0133] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein may be implemented using hardware, a software module executed by a processor, or a combination of the two. The software module may be placed in a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.

[0134] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for constructing a physical examination knowledge graph, characterized in that: include: Extracting medical concept entities from existing medical documents to obtain multiple target medical concept entities, wherein the target medical concept entities include disease entities and feature entities; Extracting a plurality of preset parameters between the disease entity and the characteristic entity; Linking the multiple target medical concept entities with the medical concept entities in the first physical examination knowledge graph based on the multiple preset parameters to obtain a second physical examination knowledge graph; performing data processing on the physical examination knowledge in the second physical examination knowledge graph to obtain a third physical examination knowledge graph; Performing a quality assessment on the third physical examination knowledge graph, and obtaining a target physical examination knowledge graph when the quality assessment result meets a preset condition; The extraction of multiple preset parameters between the disease entity and the feature entity includes: Extracting the association relationship between the disease entity and the feature entity based on a pre-trained relational classification model; Extracting attribute information of the feature entity based on a slot filling method; The pre-trained relationship classification model is used to extract the association relationship between the disease entity and the feature entity, including: Adding preset characters to the medical information sentence in the existing medical document to obtain a target medical information sentence; Inputting the target medical information sentence into the relational classification model to obtain target vectors corresponding to disease entities and feature entities; Calculating the association relationship between the disease entity and the feature entity based on the target vector; The method of extracting attribute information of the feature entity based on the slot filling method includes: Based on the slot filling method, semantic recognition is performed on the medical information sentences in the existing medical document to identify the attribute information of the feature entity; Input the target medical information sentence into the BERT model to obtain vectors H′0, H′1, and H′2: H′0=W0tanh(H0)+b0 Among them, H0 is the vector representation of the target medical information sentence; H′0 represents the vector obtained after the sentence vector passes through the feedforward neural network; H h and H t Represent the character vector representation of the first entity and the last entity respectively; H′1 and H′2 are the vector representations of the first entity and the last entity obtained by summing and averaging the character vector representations of each entity through a feedforward neural network; W0, W1, and W2 represent the weight matrices at different stages of model calculation; i represents the position of the first character of the first entity in the sentence; j represents the position of the last character of the first entity in the sentence; k represents the position of the first character of the last entity in the sentence; m represents the position of the last character of the last entity in the sentence; b0, b1, and b2 represent the biases at different stages of model calculation respectively; The three target vectors are concatenated and passed through the fully connected layer, and then classified by the softmatx function to obtain the association relationship p between the disease entity and the feature entity, as shown in formulas 1 and 2: h″=W3tanh(H′0+H′1+H′2)+b3 Formula 1 p=softmax(h″) Formula 2 Among them, h″ is the vector representation of the hidden layer of the softmatx function; W3 represents the weight matrix of the current stage; b3 represents the bias of the current stage.

2. The method according to claim 1, characterized in that: The linking of the plurality of target medical concept entities with the medical concept entities in the first physical examination knowledge graph based on the plurality of preset parameters to obtain a second physical examination knowledge graph includes: Based on the entity linking method, the multiple target medical concept entities are linked with the medical concept entities in the first physical examination knowledge graph, and the hierarchical relationship between the multiple target medical concept entities and the medical concept entities in the first physical examination knowledge graph is determined; Based on the hierarchical relationship, the multiple target medical concept entities are linked to the corresponding medical concept entities in the first physical examination knowledge graph to obtain a second physical examination knowledge graph.

3. The method according to claim 2, characterized in that The entity linking method is based on linking the multiple target medical concept entities with the medical concept entities in the first physical examination knowledge graph, and determining the hierarchical relationship between the multiple target medical concept entities and the medical concept entities in the first physical examination knowledge graph, including: Based on semantic similarity, selecting multiple candidate medical concept entities corresponding to the multiple target medical concept entities from the medical concept entities in the first physical examination knowledge graph; The hierarchical relationships between the multiple target medical concept entities and the corresponding multiple candidate entities are predicted based on a pre-trained relational classification model.

4. The method according to claim 3, characterized in that The performing data processing on the physical examination knowledge in the second physical examination knowledge graph to obtain a third physical examination knowledge graph includes: Using rule reasoning method to obtain implicit medical concept entities in the second physical examination knowledge graph; Using a link prediction method to predict the association relationship between the implicit medical concept entities; A slot filling method is used to supplement the missing attribute information of the implicit medical concept entity; The second physical examination knowledge graph is updated based on the association relationship between the implicit medical concept entities and the missing attribute information of the implicit medical concept entities to obtain a third physical examination knowledge graph.

5. A device for constructing a physical examination knowledge graph, using the method according to any one of claims 1 to 4, characterized in that: include: An extraction module is used to extract medical concept entities in existing medical documents to obtain multiple target medical concept entities, wherein the target medical concept entities include disease entities and feature entities; The extraction module is further used to extract multiple preset parameters between the disease entity and the feature entity; a linking module, configured to link the plurality of target medical concept entities with the medical concept entities in the first physical examination knowledge graph based on the plurality of preset parameters to obtain a second physical examination knowledge graph; a processing module, configured to perform data processing on the physical examination knowledge in the second physical examination knowledge graph to obtain a third physical examination knowledge graph; An evaluation module is used to perform quality evaluation on the third physical examination knowledge graph, and obtain a target physical examination knowledge graph when the quality evaluation result meets a preset condition.

6. A computer device, characterized in that: include: A processor and a memory, wherein the processor is used to execute a physical examination knowledge graph construction program stored in the memory to implement the physical examination knowledge graph construction method according to any one of claims 1 to 4.

7. A storage medium, characterized in that: The storage medium stores one or more programs, and the one or more programs can be executed by one or more processors to implement the physical examination knowledge graph construction method according to any one of claims 1 to 4.

Images