Underwriting method, device, equipment and storage medium

Abstract

This application discloses an underwriting method, apparatus, device, and storage medium, belonging to the field of computer technology. Through the technical solution provided by the embodiments of this application, relevant information about the target object can be obtained during underwriting. Based on this information, a target graph network is generated. Since the target graph network includes multiple nodes related to the target object, it can completely reflect the situation of the target object. By processing the target graph network, the underwriting result of the target object can be obtained. The underwriting process does not require manual intervention, reducing the consumption of manpower and resources, and lowering underwriting costs.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular to an underwriting method, apparatus, device, and storage medium. Background Technology

[0002] With the advancement of the times, more and more users are purchasing insurance to mitigate risks. When users purchase insurance, insurers need to conduct underwriting based on the information provided by the user (policyholder). Underwriting refers to the process by which insurers, based on a comprehensive understanding and verification of the information about the insured object, assess and classify insurable risks, and then decide whether to underwrite and under what conditions.

[0003] In related technologies, when conducting underwriting, insurers often send the information provided by the user to underwriters, who then manually verify the information. However, manual underwriting requires significant manpower and resources, resulting in high costs. Summary of the Invention

[0004] This application provides an underwriting method, apparatus, device, and storage medium, which can reduce underwriting costs. The technical solution is as follows: On the one hand, an underwriting method is provided, the method comprising: Obtain the physical examination report of the target object, the physical examination report including the examination items and the corresponding examination results of the examination items; Based on the physical examination report of the target object, a target graph network is generated. The target graph network includes examination item nodes corresponding to the examination items, examination result nodes corresponding to the examination results, physical status nodes of the target object corresponding to the examination results, and edges, which are used to connect the interconnected nodes. Based on the target graph network, the underwriting result of the target object is obtained.

[0005] On the one hand, an underwriting device is provided, the device comprising: The physical examination report acquisition module is used to acquire the physical examination report of the target object. The physical examination report includes the examination items and the corresponding examination results of the examination items. The graph network generation module is used to generate a target graph network based on the physical examination report of the target object. The target graph network includes examination item nodes corresponding to the examination items, examination result nodes corresponding to the examination results, physical status nodes of the target object corresponding to the examination results, and edges, which are used to connect the interconnected nodes. The underwriting result acquisition module is used to acquire the underwriting result of the target object based on the target graph network.

[0006] In one possible implementation, the graph network generation module is configured to perform any of the following: The examination items and corresponding examination results are obtained from the physical examination report of the target object; a first named entity is obtained from the examination results, which is used to describe the physical condition of the target object; the target graph network is generated based on the examination items, the examination results, and the first named entity. The examination items and corresponding examination results are obtained from the physical examination report of the target object; the first named entity and the corresponding second named entity are obtained from the examination results, wherein the second named entity is used to describe the attributes of the corresponding first named entity; the target graph network is generated based on the examination items, the examination results, the first named entity and the corresponding second named entity.

[0007] In one possible implementation, the graph network generation module is configured to perform named entity recognition on the inspection result to obtain a first named entity and a plurality of second named entities; obtain first position information of the first named entity in the inspection result and second position information of the plurality of second named entities in the inspection result; and, based on the first position information and the second position information, obtain a second named entity corresponding to the first named entity from the plurality of second named entities.

[0008] In one possible implementation, the device further includes a weight determination module, configured to: obtain the word frequency of the text corresponding to the body state node in the text corresponding to the inspection result node for any inspection result node and the corresponding body state node; obtain the reverse file frequency of the text corresponding to the body state node in a corpus, the corpus including multiple text files; determine the weight of the edge connecting the inspection result node and the body state node based on the word frequency and the reverse file frequency; and set the weight of the edge connecting the inspection item node and the body state node to a target value if there is an edge between the inspection item node and the body state node for any inspection item node and any body state node.

[0009] In one possible implementation, the underwriting result acquisition module is used to encode multiple nodes in the target graph network to obtain node features corresponding to each node; obtain the adjacency matrix and degree matrix of the target graph network based on the edge weights; input the node features of the multiple nodes, the adjacency matrix and degree matrix of the target graph network into the underwriting model; perform graph convolution on the node features of the multiple nodes, the adjacency matrix and degree matrix of the target graph network through the underwriting model to obtain the graph network features of the target graph network; perform full connection and normalization on the graph network features, and output the underwriting result of the target object.

[0010] In one possible implementation, the underwriting result acquisition module is configured to perform any of the following: Embedding encoding is performed on multiple nodes in the target graph network to obtain the node features corresponding to each node; Semantic encoding is performed on multiple nodes in the target graph network to obtain the node features corresponding to each node.

[0011] In one possible implementation, the underwriting result acquisition module is used to regularize the adjacency matrix using the degree matrix through the underwriting model to obtain a regularized matrix of the adjacency matrix; and to process the regularized matrix of the adjacency matrix and the node features of the multiple nodes using the convolution weights of the underwriting model to obtain the graph network features of the target graph network.

[0012] In one possible implementation, the device further includes: The model parameter adjustment module is used to obtain the medical examination reports of multiple sample objects and the historical underwriting results of the multiple sample objects; generate a sample graph network for each of the sample objects based on the medical examination reports of the multiple sample objects; input the sample graph network of each of the sample objects into the underwriting model, and output the predicted underwriting result of each of the sample objects through the underwriting model; and adjust the model parameters of the underwriting model based on the predicted underwriting result of each of the sample objects and the corresponding historical underwriting result.

[0013] On one hand, a computer device is provided, the computer device including one or more processors and one or more memories, the one or more memories storing at least one computer program, the computer program being loaded and executed by the one or more processors to implement the underwriting method.

[0014] On one hand, a computer-readable storage medium is provided, wherein at least one computer program is stored in the computer-readable storage medium, the computer program being loaded and executed by a processor to implement the underwriting method.

[0015] On one hand, a computer program product or computer program is provided, which includes program code stored in a computer-readable storage medium. A processor of a computer device reads the program code from the computer-readable storage medium and executes the program code, causing the computer device to perform the aforementioned underwriting method.

[0016] The technical solution provided in this application embodiment enables the acquisition of the target object's medical examination report during underwriting. Based on the medical examination report, a target graph network is generated. Since the target graph network includes examination item nodes, examination result nodes, and physical status nodes, it can fully reflect the target object's physical condition. By processing the target graph network, the underwriting result of the target object can be obtained. The underwriting process does not require manual intervention, reducing the manpower and material resources consumed and lowering the underwriting cost. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a schematic diagram of the underwriting process in related technologies; Figure 2 This is a schematic diagram of the implementation environment of an underwriting method provided in an embodiment of this application; Figure 3 This is a flowchart of an underwriting method provided in an embodiment of this application; Figure 4 This is a flowchart of an underwriting method provided in an embodiment of this application; Figure 5 This is a schematic diagram of an interface provided in an embodiment of this application; Figure 6 This is a schematic diagram of an interface provided in an embodiment of this application; Figure 7 This is a schematic diagram of an interface provided in an embodiment of this application; Figure 8 This is a schematic diagram of a node feature extraction method provided in an embodiment of this application; Figure 9 This is a schematic diagram illustrating the determination of underwriting results provided in an embodiment of this application; Figure 10 This is a flowchart of an underwriting model training method provided in an embodiment of this application; Figure 11This is a schematic diagram of the structure of an underwriting model provided in an embodiment of this application; Figure 12 This is a schematic diagram of the structure of an underwriting device provided in an embodiment of this application; Figure 13 This is a schematic diagram of the structure of a terminal provided in an embodiment of this application; Figure 14 This is a schematic diagram of the structure of a server provided in an embodiment of this application. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this application clearer, the embodiments of this application will be described in further detail below with reference to the accompanying drawings.

[0020] In this application, the terms "first," "second," etc., are used to distinguish identical or similar items with essentially the same function. It should be understood that there is no logical or temporal dependency between "first," "second," and "nth," nor are there any restrictions on quantity or execution order.

[0021] In this application, the term "at least one" means one or more, and "multiple" means two or more, for example, multiple face images means two or more face images.

[0022] Artificial intelligence (AI) is the theory, methods, technology, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to achieve optimal results. In other words, AI is a comprehensive technology within computer science that attempts to understand the essence of intelligence and produce a new kind of intelligent machine that can react in a way similar to human intelligence. AI studies the design principles and implementation methods of various intelligent machines, enabling them to possess the functions of perception, reasoning, and decision-making.

[0023] Artificial intelligence (AI) is a comprehensive discipline encompassing a wide range of fields, including both hardware and software technologies. Fundamental AI technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing, operating / interactive systems, and mechatronics. AI software technologies primarily include computer vision, speech processing, natural language processing, and machine learning / deep learning.

[0024] Machine learning (ML) is a multidisciplinary field involving probability theory, statistics, approximation theory, convex analysis, and algorithm complexity theory, among others. It specifically studies how computers can simulate or implement human learning behavior to acquire new knowledge or skills, and reorganize existing knowledge sub-models to continuously improve their performance. Machine learning is the core of artificial intelligence and the fundamental way to endow computers with intelligence; its applications span all areas of artificial intelligence. Machine learning and deep learning typically include techniques such as artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and instruction-based learning.

[0025] Medical cloud in cloud technology refers to the creation of a healthcare service cloud platform using cloud computing, mobile technology, multimedia, 4G communication, big data, and the Internet of Things, combined with medical technology. This enables the sharing of medical resources and expands the scope of medical services. Because of the application of cloud computing technology, medical cloud improves the efficiency of medical institutions and facilitates access to medical care for residents. Features such as hospital appointment booking, electronic medical records, and medical insurance are all products of the integration of cloud computing and the medical field. Medical cloud also has advantages such as data security, information sharing, dynamic expansion, and global deployment. For example, the underwriting model provided in this application embodiment is deployed on a healthcare service cloud platform.

[0026] Semantic features: Features used to represent the semantics expressed by text. Different texts can correspond to the same semantic features; for example, the text "How's the weather today?" and the text "What's the weather like today?" can correspond to the same semantic feature. Computer devices can map characters in text into character vectors, and combine and operate on the character vectors according to the relationships between characters to obtain the semantic features of the text. For example, computer devices can use a bidirectional encoder representation (BERT).

[0027] Normalization: Mapping sequences with different value ranges to the interval (0, 1) to facilitate data processing. In some cases, normalized values ​​can be directly expressed as probabilities.

[0028] Embedded coding, mathematically speaking, represents a correspondence, that is, mapping data in space X to space Y using a function F. This function F is injective, and the mapping result preserves the structure. An injective function means that the mapped data uniquely corresponds to the original data, and preserving the structure means that the order of the original data remains the same. For example, if there are data X1 and X2 before mapping, after mapping we get Y1 corresponding to X1 and Y2 corresponding to X2. If the original data X1 > X2, then correspondingly, the mapped data Y1 > Y2. For words, this means mapping words to another space to facilitate subsequent machine learning and processing.

[0029] Graph Convolutional Network (GCN): A graph convolutional neural network is a type of deep learning network that, unlike traditional convolutional neural networks (CNNs), can be applied to graph networks.

[0030] In this embodiment of the application, the acquisition of the physical examination report must be authorized by the user before it can be used.

[0031] One related technology involves generating a structured electronic medical examination report by receiving a medical examination report sent from a front end, then obtaining the medical examination information in the electronic medical examination report and the types of insurance in the insurance policy, and then comparing the electronic medical examination report with the pre-set evaluation rules for each type of insurance and the judgment rules for each disease, and generating the corresponding underwriting result based on the comparison results. Figure 1 This is an underwriting example provided by related technology, which makes a judgment based on a family history of breast cancer. Figure 1 As can be seen, the relevant technologies use rules based on gender, age, and medical history to compare electronic medical examination reports in order to obtain the final underwriting result.

[0032] Analysis reveals the following problems with the relevant technology: 1. The scalability of related technologies is extremely poor: The judgment rules and standards for all types of insurance and diseases must be manually set in advance. Different insurance companies are likely to have different underwriting rules in the past. Therefore, the related technologies need to repeatedly add, delete and modify judgment rules, resulting in a lot of repetitive work and extremely poor scalability of the underwriting system.

[0033] 2. The predictive performance of related technologies fails to meet real-world requirements: Due to differences in insurance products within insurance companies and variations in data volume across different product types, underwriting conclusion prediction models often struggle to obtain sufficient effective data during the training phase. This makes it difficult for the models to fit underwriting rules well, resulting in poor model performance. Furthermore, data issues such as class imbalance can cause models to favor predicting a particular underwriting conclusion, further reducing model performance. In addition, in actual business operations, improper writing of medical examination reports often leads to redundant information appearing in multiple examination items, which also interferes with the training and prediction of underwriting models.

[0034] 3. Related technologies heavily rely on labeled data for training, but the amount of labeled data in real-world scenarios is very limited: Due to the high privacy and scarcity of insurance data, there is often very little labeled training data available, and existing technologies can only utilize the labeled data and cannot fully utilize this small amount of labeled data. Therefore, the performance of related technologies is greatly constrained.

[0035] Figure 2 This is a schematic diagram illustrating the implementation environment of an underwriting method provided in this application embodiment. See also: Figure 2 The implementation environment may include terminal 210 and server 240.

[0036] Terminal 210 is connected to server 240 via a wireless or wired network. Optionally, terminal 210 may be a smartphone, tablet, laptop, desktop computer, smartwatch, etc., but is not limited to these. Terminal 210 has an application installed and running that supports the display of underwriting results.

[0037] Server 240 is a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDN), and big data and artificial intelligence platforms. Server 240 provides background services for applications running on terminal 210.

[0038] Those skilled in the art will understand that the number of terminals described above can be more or less. For example, there may be only one terminal, or there may be dozens or hundreds of terminals, or even more, in which case other terminals may also be included in the above implementation environment. This application does not limit the number of terminals or the type of device in its embodiments.

[0039] After introducing the implementation environment of the embodiments of this application, the application scenarios of the embodiments of this application will be described below. In the following description, the terminal is the terminal 210 in the above implementation environment, and the server is the server 240 in the above implementation environment. The underwriting method provided by the embodiments of this application can be applied to scenarios where policyholders purchase various types of commercial medical insurance, such as when policyholders purchase critical illness insurance, hospitalization insurance, or outpatient insurance. The embodiments of this application do not limit this application; the policyholder here is the target object.

[0040] In scenarios where an insured purchases critical illness insurance, the insured provides a recent medical examination report to the insurer, who verifies the report. The insurer is the company providing the insurance service. In some embodiments, the insured uploads the medical examination report to a server via a terminal, which then verifies it. For example, the terminal may run an application provided by the insurer, through which the insured uploads the report. The server receives and verifies the report. The terminal is the one used by the insured, and the server is either a server provided by the insurer or a server associated with a cloud service. After successful verification, the insurer processes the medical examination report based on the technical solution provided in this application, generating a target graph network corresponding to the report. The insurer processes the target graph network through the server to obtain the underwriting result for the insured. The server then sends this underwriting result to the terminal used by the insured, who displays it to the insured. This underwriting result indicates whether the insured is eligible to purchase the critical illness insurance and the associated costs. The underwriting process does not require the involvement of underwriters; the server can quickly determine the underwriting result based on the insured's medical examination report, making the underwriting process highly efficient.

[0041] In some embodiments, the terminal is an application terminal provided by the insurance company to the policyholder. After arriving at the insurance company, the policyholder can upload a medical examination report through this terminal. The terminal can verify the medical examination report to determine whether it is the policyholder's medical examination report and its authenticity. After successful verification, the terminal can upload the medical examination report to the server, which generates a target graph network based on the report, processes the target graph network, and obtains the policyholder's underwriting result. The server then sends the policyholder's underwriting result to the underwriting terminal, which displays the result to the underwriter.

[0042] It should be noted that the above explanation is based on the scenario of the policyholder purchasing critical illness insurance. For other scenarios such as the policyholder purchasing hospitalization insurance or outpatient insurance, the implementation process is based on the same inventive concept as the purchase of critical illness insurance. The implementation process is described above and will not be repeated here.

[0043] The underwriting method provided in this application embodiment can be executed by a terminal, a server, or by a combination of both. That is, the terminal takes the medical examination report from its mobile phone, uploads the medical examination report to the server, the server receives the medical examination report, and obtains the underwriting result based on the medical examination report. This application embodiment does not limit this.

[0044] After introducing the implementation environment and application scenarios of the embodiments of this application, the underwriting method provided by the embodiments of this application will be described below.

[0045] Figure 3 This is a flowchart of an underwriting method provided in an embodiment of this application. See also: Figure 3 Taking the server as the executing entity as an example, the methods include: 301. The server obtains the physical examination report of the target object, which includes the examination items and the corresponding examination results.

[0046] The target audience refers to policyholders who wish to purchase insurance, and the results of the examinations reflect their physical condition. In some embodiments, the examination items in the medical examination report are presented as examination item titles, such as "Imaging Examination - Leg MRI," and the examination results are presented in text form, such as "Negative."

[0047] 302. Based on the physical examination report of the target object, the server generates a target graph network. The target graph network includes the examination item node corresponding to the examination item, the examination result node corresponding to the examination result, the physical status node of the target object corresponding to the examination result, and edges, which are used to connect the interconnected nodes.

[0048] The target graph network reflects the relationships between examination item nodes, examination result nodes, and body status nodes, which are represented by edges between nodes. If an edge exists between two nodes, it indicates a correlation. For example, given an examination item node "Imaging Examination - Leg MRI" and a test result node "Negative," an edge between them means the result node "Negative" was obtained through the examination item corresponding to "Imaging Examination - Leg MRI." Edges in the target graph network have weights, reflecting the degree of correlation between connected nodes. Higher weights indicate a stronger correlation, while lower weights indicate a weaker correlation.

[0049] 303. The server obtains the underwriting result of the target object based on the target graph network.

[0050] The underwriting result is used to indicate whether to insure the target, the scope of coverage, and the cost of coverage. Since the target graph network is generated based on the target's medical examination report, it can more accurately reflect the target's physical condition, and the underwriting result determined based on the target graph network is more accurate.

[0051] The technical solution provided in this application embodiment enables the acquisition of the target object's medical examination report during underwriting. Based on the medical examination report, a target graph network is generated. Since the target graph network includes examination item nodes, examination result nodes, and physical status nodes, it can fully reflect the target object's physical condition. By processing the target graph network, the underwriting result of the target object can be obtained. The underwriting process does not require manual intervention, reducing the manpower and material resources consumed and lowering the underwriting cost.

[0052] Steps 301-303 above are a brief description of the technical solution provided in the embodiments of this application. The underwriting method provided in the embodiments of this application will be described in more detail below with some examples. See [link to relevant documentation]. Figure 4 The methods include: 401. The terminal obtains the physical examination report of the target object, which includes the examination items and the corresponding examination results.

[0053] The physical examination report is an electronic document that records the examination items and their corresponding results. Examination items indicate the methods used during the examination; for example, "blood test" and "MRI" are two separate examination items, representing two different methods. Different examination items reflect the individual's physical condition from different perspectives. The report contains multiple examination items, allowing for a more accurate assessment of the individual's health. Each examination item corresponds to a single result, reflecting the individual's overall health status. See also... Figure 5 The document provides a schematic diagram of a medical examination report 500, which includes multiple examination items 501, with the corresponding examination results 502 displayed below each examination item 501.

[0054] In one possible implementation, the terminal displays a first interface for obtaining a medical examination report. In response to an operation on this first interface, the terminal obtains the medical examination report of the target object through the target application.

[0055] In this implementation, the terminal can obtain the physical examination report through the target application, resulting in high efficiency of human-computer interaction.

[0056] The above implementation methods will be explained more clearly through several examples below.

[0057] Example 1: A terminal runs a target application, on which the target user's account is logged in. This target application is provided by an insurer (insurance company). The terminal displays a second interface of the target application, also referred to as the main interface. In response to an operation on the second interface, the terminal displays a first interface, which shows a medical examination report upload control. In response to clicking the medical examination report upload control, the terminal displays a medical examination report selection interface, which displays at least one candidate medical examination report. In response to selecting any one of the at least one medical examination report, the selected medical examination report is identified as the target user's medical examination report.

[0058] It should be noted that the medical examination reports displayed in the medical examination report selection interface can be either reports stored on the terminal or reports retrieved by the terminal from the cloud. For example, the terminal can connect to a medical examination report server that maintains a medical examination report database containing reports from multiple individuals. In response to clicking the medical examination report upload control, the terminal sends a medical examination report retrieval request to the medical examination report server, carrying the user account of the target individual. Upon receiving the request, the medical examination report server retrieves the user account from the request, queries the medical examination report database based on the user account, and obtains at least one medical examination report for the target individual. These different reports may be from different medical examination institutions or from different times. The medical examination report server then sends these at least one medical examination report to the terminal, which displays them in the medical examination report selection interface. In some embodiments, the medical examination report server is the server of a medical examination institution recognized by the insurance company.

[0059] For example, the terminal is a user device such as a mobile phone or tablet used by a user. This user can be the target individual or a relative of the target individual. The terminal runs the target application, on which the target individual's user account is logged. The terminal displays a second interface of the target application, which introduces different types of insurance and provides purchase entry points for each type of insurance; these purchase entry points are essentially purchase controls. In response to clicking any purchase entry point on the second interface, the terminal switches the second interface back to the first interface, which displays detailed information about the corresponding insurance, application instructions, and a medical examination report upload control. In response to clicking the medical examination report upload control, the terminal displays a medical examination report selection interface, which displays at least one medical examination report of the target individual. In response to selecting any of these medical examination reports, the selected report is identified as the target individual's medical examination report.

[0060] See Figure 6The terminal displays a second interface 601 of the target application, which shows multiple purchase controls. In response to a click on purchase control 602, the terminal switches the second interface 601 to a first interface 603, which displays a medical examination report upload control 604. In response to a click on medical examination report upload control 604, the terminal displays a medical examination report selection interface 606, which displays at least one medical examination report for the target object. In response to a click on medical examination report 606, the selected medical examination report 606 is identified as the medical examination report for the target object.

[0061] Example 2: The terminal runs a target application provided by an insurer (insurance company). The terminal displays a second interface of the target application, also known as the main interface. In response to an operation on the second interface, the terminal displays a first interface containing authentication controls. In response to clicking the authentication controls, the terminal displays an authentication interface used to verify the identity of the target object. Upon successful authentication of the target object through the authentication interface, the terminal retrieves the target object's medical examination report.

[0062] For example, the terminal could be a user device such as a mobile phone or tablet used by the target, or an underwriting terminal provided by an insurance company. The terminal runs the target application and displays a second interface for that application. This second interface introduces different types of insurance and provides purchase entry points for each type of insurance; these purchase entry points are essentially purchase controls. In response to a click on any purchase entry point on the second interface, the terminal switches the second interface back to the first interface, displaying detailed information about the corresponding insurance, underwriting instructions, and an identity verification control. In response to a click on the identity verification control, the terminal displays an identity verification interface, which includes an identity information input area or a facial verification control. If the identity verification interface includes an identity information input area, the target can enter their identity information there. This identity information is used to identify the target's identity, and the identity information input area includes an identity verification control. In response to a click on the identity verification control, the terminal verifies the identity information entered by the target in the identity information input area. Upon successful verification, the terminal retrieves the target's medical examination report from the medical examination report server. When the identity verification interface includes a face verification control, in response to clicking the face verification control, the terminal captures a facial image of the target object and verifies the facial image. Upon successful face image verification, the terminal retrieves the target object's medical examination report from the medical examination report server.

[0063] Example 3: A terminal is electrically connected to a scanning device. The terminal runs a target application provided by an insurer (insurance company). The terminal displays a second interface of the target application. In response to an operation on the second interface, the terminal displays a first interface containing scanning controls. In response to a click on these scanning controls, the terminal controls the scanning device to start scanning. The terminal identifies the file acquired by the scanning device as the medical examination report for the target individual.

[0064] For example, the terminal is an underwriting terminal provided by an insurance company, which is electrically connected to a scanning device. The terminal uses the target application and displays a second interface of that application. This second interface introduces different types of insurance and provides purchase entry points for each type of insurance; these purchase entry points are essentially purchase controls. In response to clicking any purchase entry point on the second interface, the terminal switches the second interface to a first interface, displaying detailed information about the corresponding insurance, underwriting instructions, and the scanning control. The user places a paper medical examination report in the scanning device. In response to clicking the scanning control, the terminal controls the scanning device to start working and acquire a scanned image of the paper medical examination report. While the terminal is controlling the scanning device, it can also prompt the user to perform operations such as page turning. After acquiring the scanned image of the paper medical examination report, the terminal can convert the scanned image into text, for example, through OCR (Optical Character Recognition). This text is the medical examination report of the target object.

[0065] 402. The terminal sends the physical examination report of the target object to the server.

[0066] In one possible implementation, the terminal displays a third interface, which includes a medical examination report sending control. This third interface can be the same as or different from the first interface; this embodiment does not limit this. In response to a click on the medical examination report sending control, the terminal sends the medical examination report of the target object to the server.

[0067] In this implementation, the terminal provides the user with a physical examination report sending control on the third interface. The user can control the terminal to send the physical examination report of the target object to the server through the physical examination report sending control, which has high efficiency of human-computer interaction.

[0068] For example, see Figure 7 The terminal displays a third interface 701, which includes a medical examination report sending control 702. In response to a click on the medical examination report sending control 702, the terminal sends the medical examination report of the target object to the server.

[0069] In one possible implementation, in response to obtaining the target object's medical examination report, the terminal sends the report to the server. That is, after obtaining the target object's medical examination report, the terminal can directly send it to the server without manual operation by the user.

[0070] 403. The server receives the physical examination report of the target object.

[0071] 404. Based on the physical examination report of the target object, the server generates a target graph network. The target graph network includes the examination item node corresponding to the examination item, the examination result node corresponding to the examination result, the physical status node of the target object corresponding to the examination result, and edges, which are used to connect the interconnected nodes.

[0072] In some embodiments, the target object's physical state node is also referred to as a disease node.

[0073] In one possible implementation, the server retrieves the examination items and corresponding examination results from the target object's medical examination report. The server then retrieves a first named entity from the examination results, which describes the target object's physical condition. Based on the examination items, the examination results, and the first named entity, the server generates the target graph network.

[0074] To provide a clearer explanation of the above embodiments, the embodiments will be described in three parts below.

[0075] Part 1: The server retrieves the examination item and the corresponding examination result from the physical examination report of the target object.

[0076] In one possible implementation, the server performs template matching on the target object's medical examination report to identify multiple examination items. The server then extracts the text below each examination item in the medical examination report as the corresponding examination result.

[0077] For example, the server stores templates for multiple examination items. The server can periodically update these templates to ensure they cover the latest examination items. The server uses these templates to match the target object's medical examination report, obtaining the examination items from the report. The server then extracts the text below each examination item in the report as the corresponding examination result. In some embodiments, after retrieving the examination items and their corresponding results from the medical examination report, the server can generate key-value pairs based on these items. The key of each pair is the examination item, and the value is the corresponding examination result. By processing the examination items and their corresponding results into key-value pairs, the server facilitates subsequent queries.

[0078] For example, the server can process the target object's medical examination report into the following format: Medical Examination Report - Summary of Abnormalities - Re-examination Recommendations - Positive / Negative Values: / ; Medical Examination Report - Summary of Abnormalities - Final Examination Recommendations - Positive / Negative Results - Value: / ; Medical Examination Report - Summary of Abnormalities - Overall Examination Opinion - Positive / Negative Value: / ; Physical Examination Report - Summary of Abnormalities - Final Examination Conclusion - Positive / Negative - Value: / ; Physical examination report - Imaging examination - CT scan - Positive and negative values: / ; Physical Examination Report - Imaging Examination - MRI - Original Name - Value: MRI; Physical examination report - Imaging examination - MRI - Structured MRI - Organs - Spinal canal - Value: Local stenosis; Physical Examination Report - Imaging Examination - MRI - Structured Diagnosis - Organs - Intervertebral Discs - Values: Posterior protrusion of intervertebral discs at C3-4, C4-5, C5-6, and C6-7; Physical Examination Report - Imaging Examination - Radiological Examination - X-ray of Other Parts - Values: / ; Physical Examination Report - Imaging Examination - Radiology Examination - Chest - Values: / ; Physical Examination Report - Imaging Examination - Radiological Examination - Lumbar Spine - Value: / .

[0079] In this context, CT stands for Computed Tomography, and MRI stands for Magnetic Resonance Imaging. The " / " indicates that the corresponding value does not exist. "Physical Examination Report_Imaging Examination_MRI_Structural_Organ_Spinal Canal_Value" represents an examination item or title, while "Local Stenosis" represents the examination result or content.

[0080] The second part is that the server obtains the first named entity from the inspection result, which is used to describe the physical state of the target object.

[0081] In some embodiments, the first named entity is also referred to as the disease entity, which corresponds to a disease and is the name of the corresponding disease.

[0082] In one possible implementation, the server performs named entity recognition on the examination result to obtain the first named entity in the examination result. That is, the server performs named entity recognition on the examination result to obtain the disease entity in the examination result. For example, for the examination result "Admission diagnosis: 1. Chronic kidney disease, stage 5 2. Metabolic acidosis 3. Electrolyte imbalance 4. Anemia", after the server performs named entity recognition on the examination result, it can obtain four first named entities: "chronic kidney disease", "metabolic acidosis", "electrolyte imbalance" and "anemia". These four first named entities can describe the physical condition of the target object.

[0083] The above implementation method is illustrated below with two examples.

[0084] Example 1: The server performs word segmentation on the inspection result, obtaining multiple reference words corresponding to the inspection result. The server queries these multiple reference words in a knowledge text database, and determines the queried reference words as the first named entities. The knowledge text database includes a disease entity dictionary and a disease entity knowledge graph, both of which store multiple first named entities.

[0085] For example, the server segments the check result at fixed intervals to obtain multiple reference words containing the same number of characters. The server then queries each of these reference words in a knowledge text database. When any one of the reference words is found to be identical to any word in the knowledge text database, the server designates that reference word as the first named entity.

[0086] Example 2: The server segments the inspection result using different methods, resulting in multiple reference word sequences corresponding to each method. Each reference word sequence includes multiple reference words from the inspection result. Different reference words within the same reference word sequence have the same number of characters, while reference words in different reference word sequences have different character counts. In response to multiple reference words belonging to different reference word sequences being identical to multiple words in the knowledge text database, the server identifies the reference word with the largest number of characters among the multiple reference words belonging to different reference word sequences as the first named entity.

[0087] Part Three: The server generates the target graph network based on the inspection item, the inspection result, and the first named entity.

[0088] In one possible implementation, the server generates an examination item node based on the examination item, an examination result node based on the examination result, and a body status node based on the first named entity. The server generates edges between the examination item node and its corresponding examination result node, and between the examination item node and its corresponding body status node, thus obtaining the target graph network. In this target graph network, body status nodes are connected to their corresponding examination result nodes by connecting examination item nodes, thereby distinguishing situations where different examination items simultaneously detect the same disease. One examination result node may correspond to multiple body status nodes, indicating that one examination item detects multiple diseases, and one body status node corresponds to one disease. Correspondingly, one body status node may also be connected to multiple examination result nodes, indicating that multiple examination items have detected the disease corresponding to that body status node.

[0089] For example, a physical condition node (such as "anemia") may be reflected in multiple examination items in a physical examination report. Therefore, in the target graph network, the "anemia" node may be connected to multiple examination result nodes (i.e., examination items that contain "anemia") through different examination item nodes. However, the relationship between different examination result nodes and "anemia" is different. For example, some examination items are diagnostic reports, and some are ultrasound reports. Therefore, examination item nodes (such as "physical examination report_imaging examination") are used to bridge the physical condition node and the corresponding examination result node.

[0090] The method for determining the weights of the edges in the target graph network is explained below.

[0091] In one possible implementation, for any inspection result node and its corresponding body state node, the server obtains the word frequency of the text corresponding to the body state node within the text corresponding to the inspection result node. The server obtains the inverse file frequency of the text corresponding to the body state node in a corpus comprising multiple text files. Based on the word frequency and the inverse file frequency, the server determines the weight of the edge connecting the inspection result node and the body state node.

[0092] Term Frequency (TF) refers to the number of times the text corresponding to the body state node appears in the text corresponding to the inspection result node, divided by the total number of words in the text corresponding to the inspection result node. Inverse Document Frequency (IDF) refers to the number of files in the corpus that contain the text corresponding to the body state node, divided by the total number of files in the corpus.

[0093] In some embodiments, the weight of the edge between the inspection result node and the corresponding body state node is represented by the following formula (1): A ij=TF-IDF ij (1) Where i is the number of the body status node, j is the number of the inspection result node, and A ij Let i be the weight of the edge between the body state node with number i and the inspection result node with number j.

[0094] In one possible implementation, for any check item node and any body status node, if there is an edge between the check item node and the body status node, the server sets the weight of the edge connecting the check item node and the body status node to a target value.

[0095] The target value is set by technicians according to the actual situation, for example, it can be set to 1.

[0096] In some embodiments, the weight of the edge between the inspection item node and the body state node is represented by the following formula (2): A ik =0, e(i,k) E(2) A ik =1, e(i,k)∈E Among them, A ik Let e(i,k) be the weight of the edge between the body state node i and the inspection item node k, and let e(i,k) be the weight of the edge between the body state node i and the inspection item node k. Let E be the set of edges in the target graph network.

[0097] Another implementation of step 404 will be described below.

[0098] In one possible implementation, the server obtains the examination item and the corresponding examination result from the target object's medical examination report. The server then obtains the first named entity and the corresponding second named entity from the examination result, whereby the second named entity describes the attributes of the corresponding first named entity. Based on the examination item, the examination result, the first named entity, and the corresponding second named entity, the server generates the target graph network.

[0099] To provide a clearer explanation of this implementation method, the above implementation method will be described in three parts below.

[0100] Part 1: The server retrieves the examination item and the corresponding examination result from the physical examination report of the target object.

[0101] In one possible implementation, the server performs template matching on the target object's medical examination report to identify multiple examination items. The server then extracts the text below each examination item in the medical examination report as the corresponding examination result.

[0102] For example, the server stores templates for multiple examination items. The server can periodically update these templates to ensure they cover the latest examination items. The server uses these templates to match the target object's medical examination report, obtaining the examination items from the report. The server then extracts the text below each examination item in the report as the corresponding examination result. In some embodiments, after retrieving the examination items and their corresponding results from the medical examination report, the server can generate key-value pairs based on these items. The key of each pair is the examination item, and the value is the corresponding examination result. By processing the examination items and their corresponding results into key-value pairs, the server facilitates subsequent queries.

[0103] The second part involves the server retrieving the first named entity and the corresponding second named entity from the inspection result. The second named entity is used to describe the attributes of the corresponding first named entity.

[0104] In some embodiments, the first named entity is also referred to as the disease entity, which corresponds to a disease and is the name of the corresponding disease. The second named entity is used to describe the attributes corresponding to the first named entity, indicating the severity of the disease corresponding to the first named entity. For example, for the first named entity "kidney disease", the corresponding second named entity may be "chronic" and "stage 5", which can describe the severity of "kidney disease".

[0105] In one possible implementation, the server performs named entity recognition on the inspection result to obtain the first named entity and a plurality of second named entities. The server obtains the first position information of the first named entity in the inspection result, and the second position information of the plurality of second named entities in the inspection result. Based on the first position information and the second position information, the server retrieves the second named entity corresponding to the first named entity from the plurality of second named entities.

[0106] The first position information of the first named entity in the inspection result refers to its absolute position in the inspection result, that is, the starting and ending character positions of the first named entity in the inspection result. Correspondingly, the second position information of the second named entity in the inspection result refers to its absolute position in the inspection result, that is, the starting and ending character positions of the second named entity in the inspection result.

[0107] The above implementation method is illustrated below with two examples.

[0108] Example 1: The server segments the inspection result into words, obtaining multiple reference words corresponding to the inspection result. The server queries these multiple reference words in a knowledge text database and an attribute text database. The reference words found in the knowledge text database are identified as first named entities, and the reference words found in the attribute text database are identified as second named entities. The knowledge text database includes a disease entity dictionary and a disease entity knowledge graph, both of which store multiple first named entities. The attribute text database stores multiple second named entities.

[0109] For example, the server segments the inspection result at fixed intervals, obtaining multiple reference words containing the same number of characters. The server queries each of these reference words in a knowledge text database. When any one of the reference words is found to be identical to any word in the knowledge text database, the server designates that reference word as the first named entity. The server then queries each of these reference words in an attribute text database. When any one of the reference words is found to be identical to any word in the attribute text database, the server designates that reference word as the second named entity. The server then obtains the first position information of the first named entity in the inspection result, as well as the second position information of the multiple second named entities in the inspection result. In response to the fact that the second location information of any of the plurality of second named entities meets the target condition with the first location information, the second named entity is determined as the second named entity corresponding to the first named entity. The fact that the second location information meets the target condition with the first location information means that the difference information between the second location information and the first location information meets the difference information condition, that is, the difference between the second location information and the first location information is less than or equal to the difference threshold. The difference threshold is set by the technician according to the actual situation, and this application embodiment does not limit it.

[0110] Example 2: The server segments the inspection result using different methods, obtaining multiple reference word sequences corresponding to each method. Each reference word sequence includes multiple reference words from the inspection result. Different reference words within the same reference word sequence have the same number of characters, while reference words in different reference word sequences have different character counts. In response to multiple reference words belonging to different reference word sequences being identical to multiple words in a knowledge text database, the server identifies the reference word with the most characters among the multiple reference words belonging to different reference word sequences as the first named entity. Similarly, in response to multiple reference words belonging to different reference word sequences being identical to multiple words in an attribute text database, the server identifies the reference word with the most characters among the multiple reference words belonging to different reference word sequences as the second named entity. The server obtains the first position information of the first named entity in the inspection result, and the second position information of the multiple second named entities in the inspection result. In response to the fact that the second location information of any of the plurality of second named entities meets the target condition with the first location information, the second named entity is determined as the second named entity corresponding to the first named entity. The fact that the second location information meets the target condition with the first location information means that the difference information between the second location information and the first location information meets the difference information condition, that is, the difference between the second location information and the first location information is less than or equal to the difference threshold. The difference threshold is set by the technician according to the actual situation, and this application embodiment does not limit it.

[0111] Part Three: The server generates the target graph network based on the inspection item, the inspection result, the first named entity, and the corresponding second named entity.

[0112] In one possible implementation, the server generates inspection item nodes based on the inspection item, inspection result nodes based on the inspection result, body state nodes based on the first named entity, and attribute nodes based on the second named entity. The server generates edges between inspection item nodes and their corresponding inspection result nodes, between inspection item nodes and their corresponding body state nodes, and between body state nodes and their corresponding attribute nodes, thus obtaining the target graph network. Attribute nodes are only connected to body state nodes; however, some body state nodes may not have any connected attribute nodes.

[0113] The method for determining the weights of the edges in the target graph network is explained below.

[0114] In one possible implementation, for any inspection result node and its corresponding body state node, the word frequency of the text corresponding to the body state node within the text corresponding to the inspection result node is obtained. The server obtains the inverse file frequency of the text corresponding to the body state node in a corpus comprising multiple text files. Based on the word frequency and the inverse file frequency, the server determines the weight of the edge connecting the inspection result node and the body state node.

[0115] This implementation method is described above in the formula (1), and will not be repeated here.

[0116] In one possible implementation, for any check item node and any body status node, if there is an edge between the check item node and the body status node, the server sets the weight of the edge connecting the check item node and the body status node to a target value.

[0117] This implementation method is described in the above description of formula (2), and will not be repeated here.

[0118] In one possible implementation, for any body state node and any attribute node, if there is an edge between the body state node and the attribute node, the server sets the weight of the edge connecting the body state node and the attribute node to a target value.

[0119] In some embodiments, the weight of the edge between the body state node and the attribute node is represented by the following formula (3): A im =0, e(i,m) E(3) A im =1, e(i,m)∈E Among them, A im Let e(i, m) be the weight of the edge between the body state node i and the attribute node m, and let E be the set of edges in the target graph network.

[0120] 405. The server obtains the underwriting result of the target object based on the target graph network.

[0121] In one possible implementation, the server encodes multiple nodes in the target graph network to obtain the node features corresponding to each node. Based on the edge weights, the server obtains the adjacency matrix and degree matrix of the target graph network. The server inputs the node features of the multiple nodes, the adjacency matrix, and the degree matrix of the target graph network into an underwriting model. The underwriting model performs graph convolution on the node features of the multiple nodes, the adjacency matrix, and the degree matrix of the target graph network to obtain the graph network features of the target graph network. The graph network features are then fully connected and normalized, and the underwriting result for the target object is output.

[0122] To provide a clearer explanation of the above embodiments, the following description will be divided into four parts.

[0123] The first part involves the server encoding multiple nodes in the target graph network to obtain the node features corresponding to each node.

[0124] In one possible implementation, the server performs embedding encoding on multiple nodes in the target graph network to obtain the node features corresponding to each node.

[0125] The server performs embedded coding on multiple nodes in the target graph network, mapping the text corresponding to each node to a vector. This vector represents the node features of each node, and the language model used for embedded coding is word2vec (word-to-vector). In some embodiments, this vector is called the node feature vector of each node. Embedded coding can map the same text to the same vector and different text to different vectors. In some embodiments, after the server performs embedded coding on multiple nodes in the target graph network to obtain the node features corresponding to each node, it can store the node features of each node in the form of a matrix. For example, the server can number the multiple node features and store them in the matrix in the order of their numbers. Each row of the matrix stores one node feature, and the row number of each node feature is its index, which facilitates subsequent querying.

[0126] In one possible implementation, the server performs semantic encoding on multiple nodes in the target graph network to obtain the node features corresponding to each node.

[0127] In some embodiments, the server can semantically encode multiple nodes using a semantic feature extraction model to obtain node features corresponding to each node. This semantic feature extraction model can be a Bidirectional Encoder Representations from Transformers (BERT) model, a Medical BERT model (MedBERT), a Lite BERT (ALBERT) model, etc., and this application does not limit the specific model. In some embodiments, after the server semantically encodes multiple nodes in the target graph network to obtain node features corresponding to each node, it can store the node features of each node in the form of a matrix. For example, the server can number the multiple node features and store them in the matrix according to the order of the numbers. Each row of the matrix stores one node feature, and the row number of each node feature is its index, which facilitates subsequent querying.

[0128] See Figure 8 After obtaining the physical examination report 801 of the target object, the server generates a target graph network 802 based on the physical examination report 801. The server encodes multiple nodes in the target graph network 802 to obtain the node features corresponding to each node.

[0129] Part Two: Based on the weight of the edge, the server obtains the adjacency matrix and degree matrix of the target graph network.

[0130] The adjacency matrix is ​​a square matrix with the same number of rows and columns as the number of nodes in the target graph network. Each row and column of the adjacency matrix corresponds to one node. The adjacency matrix consists of 0s and N, where N is the weight of the corresponding edge. 0 indicates that there is no edge between the two corresponding nodes. Here, the two corresponding nodes refer to the node in the row and the node in the column containing 0 in the adjacency matrix.

[0131] In one possible implementation, the server generates an initial adjacency matrix, the number of rows and columns of which are the same as the number of nodes in the target graph network. The server fills this initial adjacency matrix with 0s or Ns based on the edge relationships between multiple nodes in the target graph network, thus obtaining the adjacency matrix of the target graph network. Specifically, if there is no edge between any two nodes in the target graph network, the server fills the corresponding position of those two nodes in the initial adjacency matrix with 0s. If there is an edge between any two nodes in the target graph network, the server fills the corresponding position of those two nodes in the initial adjacency matrix with Ns, which represents the weight of the edge between the two nodes. The server then generates a degree matrix for the target graph network based on the number of edges between each node and other nodes. This degree matrix represents how many nodes each node has an edge with.

[0132] The third part involves the server inputting the node features of the multiple nodes, the adjacency matrix and degree matrix of the target graph network into the underwriting model. The underwriting model then performs graph convolution on the node features of the multiple nodes, the adjacency matrix and degree matrix of the target graph network to obtain the graph network features of the target graph network.

[0133] The structure and training method of the underwriting model are described in steps 901-904 following step 407.

[0134] In one possible implementation, the server uses the degree matrix to regularize the adjacency matrix using the underwriting model, obtaining a regularized matrix of the adjacency matrix. The server then uses the convolution weights of the underwriting model to process the regularized matrix of the adjacency matrix and the node features of the multiple nodes to obtain the graph network features of the target graph network. The underwriting model can have one or more convolution weights. If the underwriting model has one convolution weight, it means that the underwriting model has one convolutional layer; if the underwriting model has multiple convolution weights, it means that the underwriting model has multiple convolutional layers. This application embodiment does not limit the number of convolutional layers in the underwriting model.

[0135] In some embodiments, the server can use the degree matrix to regularize the adjacency matrix using the following formula (4) to obtain the regularized matrix of the adjacency matrix: (4) in, This is the regularization matrix of the adjacency matrix. For this degree matrix, Let be the adjacency matrix.

[0136] The server uses the convolution weights of the underwriting model according to the following formula (5) to process the regular matrix of the adjacency matrix and the node features of the multiple nodes to obtain the graph network features of the target graph network.

[0137] (5) in, The graph network features of the target graph network are as follows: For the residual function, For node features, These are the convolution weights of the first convolutional layer of the target graph network. These are the convolution weights of the second convolutional layer of the target graph network.

[0138] Part Four: The server performs full connection and normalization on the graph network features using the underwriting model, and outputs the underwriting result for the target object.

[0139] Combining the third part above, the server can use the following formula (6) to fully connect and normalize the graph network features through the underwriting model, and output the underwriting result of the target object.

[0140] (6) in, The underwriting result for this target object. For normalization function, It is a fully connected function.

[0141] 406. The server sends the underwriting result of the target object to the terminal.

[0142] 407. The terminal displays the underwriting result.

[0143] The underwriting results include five categories: underwriting, additional premium, exclusion, rejection, and others. Underwriting means that the target individual can purchase the corresponding insurance service. Additional premium means that the target individual needs to pay an additional premium when purchasing the corresponding insurance service. Exclusion means that a certain type of disease needs to be excluded when purchasing the corresponding insurance service. For example, if the target individual is diagnosed with disease A in the future, the insurance will not be underwritten. Rejection means that the target individual cannot purchase the corresponding insurance service. Others are other situations.

[0144] See Figure 9 The server processes the target graph network to obtain graph network feature 901, performs full connection and normalization on graph network feature 901, and maps graph network feature 901 to 5 underwriting results. In some embodiments, the underwriting method shown in steps 401-407 above is also called intelligent underwriting.

[0145] All of the above-mentioned optional technical solutions can be combined in any way to form the optional embodiments of this application, and will not be described in detail here.

[0146] During the experiment, a simple baseline model was designed for comparison. This baseline model uses a language model (word2vec or MedBERT) to model the examination items and results obtained from the preprocessing of the physical examination report to obtain node feature vectors. The node feature vectors are then passed through a classifier to obtain the underwriting results.

[0147] The experiment used 8000 data points, and the classification task was a 5-class classification task, which corresponds to the 5 underwriting results mentioned above. The experimental results are shown in Table 1 below: Table 1

[0148] As can be seen from Table 1 above, the underwriting model proposed in this application has a significant performance improvement compared with the baseline model. Among them, the precision, recall and F1 score have all improved to varying degrees, especially the F1 score, which has improved by 44%, proving the feasibility of constructing a graph of the medical examination report and using a graph convolutional neural network for prediction.

[0149] The technical solution provided in this application embodiment enables the acquisition of the target object's medical examination report during underwriting. Based on the medical examination report, a target graph network is generated. Since the target graph network includes examination item nodes, examination result nodes, and physical status nodes, it can fully reflect the target object's physical condition. By processing the target graph network, the underwriting result of the target object can be obtained. The underwriting process does not require manual intervention, reducing the manpower and material resources consumed and lowering the underwriting cost.

[0150] For insurance companies, the technical solution provided in this application can improve underwriting efficiency and accuracy, and significantly reduce labor and time costs. For policyholders, the technical solution provides a simplified application process, reduces their workload, and allows for real-time response to their needs. In the future, it can also help insurance companies quickly build a one-stop service for underwriting and policy maintenance, further enhancing the intelligence level of insurance business.

[0151] The structure and training method of the underwriting model described in step 405 above are explained below. (See [link to relevant documentation]). Figure 10 The training method for this underwriting model includes: 1001. The server retrieves the physical examination reports of multiple sample objects and the historical underwriting results of these multiple sample objects.

[0152] Historical underwriting results are the underwriting results obtained by insurance company underwriters through manual underwriting. When training the underwriting model, historical underwriting results are used as labels for the medical examination reports of the corresponding sample objects. The purpose of training the underwriting model is to make the underwriting results output by the underwriting model as close as possible to the corresponding historical underwriting results after the medical examination report of the sample object is input into the underwriting model.

[0153] 1002. Based on the physical examination reports of these multiple sample objects, the server generates a sample graph network for each sample object.

[0154] The method by which the server generates a sample graph network for each sample object based on the medical examination reports of the multiple sample objects is the same inventive concept as the method by which the target graph network is generated based on the medical examination report of the target object in step 404 above. The implementation process is described in step 404 above and will not be repeated here.

[0155] 1003. The server inputs the sample graph network of each sample object into the underwriting model, and outputs the predicted underwriting results of each sample object through the underwriting model.

[0156] The method of inputting the sample graph network of each sample object into the underwriting model and outputting the predicted underwriting result of each sample object through the underwriting model belongs to the same inventive concept as the method of obtaining the underwriting result of the target object based on the target graph network in step 405 above. The implementation process is described in step 405 above and will not be repeated here.

[0157] The structure of this underwriting model is explained below.

[0158] See Figure 11 The underwriting model 1100 includes a graph convolutional layer 1101, a residual layer 1102, a fully connected layer 1103, and a normalization layer 1104. In some embodiments, the underwriting model 1100 has two graph convolutional layers 1101, which are used to perform graph convolution on the input data. The residual layer 1102 is located in the second graph convolutional layer and is used to perform residual connections on the data to prevent data loss during graph convolution. The fully connected layer 1103 is used to perform fully connected processing on the data, and the normalization layer 1104 is used to normalize the data.

[0159] It should be noted that, Figure 11 This is one possible structure of the underwriting model. In other embodiments, the underwriting model may also have other structures, and this application does not limit this.

[0160] 1004. The server adjusts the model parameters of the underwriting model based on the predicted underwriting results and the corresponding historical underwriting results for each sample object.

[0161] In one possible implementation, the server constructs a binary cross-entropy loss function and a cross-entropy error loss function based on each sample object, the predicted underwriting result, and the corresponding historical underwriting result. The server then adjusts the model parameters of the underwriting model based on these two loss functions, effectively training the model. The binary cross-entropy loss function is a supervised training loss function, while the cross-entropy error loss function is a semi-supervised training loss function. This combination of supervised and semi-supervised training allows for more efficient use of the sample data (medical examination reports of multiple sample objects and their historical underwriting results), improving the training effectiveness of the underwriting model.

[0162] For example, the server constructs a binary cross-entropy loss function as shown in Equation (7) and a cross-entropy error loss function as shown in Equation (8).

[0163] (7) in, The cross-entropy loss function is used for binary classification. This refers to the medical examination report number of the sample subjects. For the number Historical underwriting results For the number The predicted underwriting results.

[0164] (8) in, Let cross-entropy be the error loss function. and This refers to the medical examination report number of the sample subjects. The number of medical examination reports for the sample subjects. This refers to all medical examination reports corresponding to a specific historical underwriting result. For the number Historical underwriting results For the number The predicted underwriting results.

[0165] Figure 12 This is a schematic diagram of an underwriting device structure provided in an embodiment of this application. See also... Figure 12 The device includes: a physical examination report acquisition module 1201, a graph network generation module 1202, and an underwriting result acquisition module 1203.

[0166] The physical examination report acquisition module 1201 is used to acquire the physical examination report of the target object, the physical examination report including the examination items and the corresponding examination results of the examination items; Graph network generation module 1202 is used to generate a target graph network based on the physical examination report of the target object. The target graph network includes examination item nodes corresponding to the examination items, examination result nodes corresponding to the examination results, physical status nodes of the target object corresponding to the examination results, and edges, which are used to connect the interconnected nodes. The underwriting result acquisition module 1203 is used to acquire the underwriting result of the target object based on the target graph network.

[0167] In one possible implementation, the graph network generation module 1202 is configured to perform any of the following: The examination items and corresponding examination results are obtained from the physical examination report of the target object; a first named entity is obtained from the examination results, which is used to describe the physical condition of the target object; the target graph network is generated based on the examination items, the examination results, and the first named entity. The examination items and corresponding examination results are obtained from the physical examination report of the target object; the first named entity and the corresponding second named entity are obtained from the examination results, wherein the second named entity is used to describe the attributes of the corresponding first named entity; the target graph network is generated based on the examination items, the examination results, the first named entity and the corresponding second named entity.

[0168] In one possible implementation, the graph network generation module 1202 is configured to perform named entity recognition on the inspection result to obtain a first named entity and a plurality of second named entities; obtain first position information of the first named entity in the inspection result and second position information of the plurality of second named entities in the inspection result; and, based on the first position information and the second position information, obtain a second named entity corresponding to the first named entity from the plurality of second named entities.

[0169] In one possible implementation, the device further includes a weight determination module, configured to: obtain the word frequency of the text corresponding to the body state node in the text corresponding to the inspection result node for any inspection result node and the corresponding body state node; obtain the reverse file frequency of the text corresponding to the body state node in a corpus, the corpus including multiple text files; determine the weight of the edge connecting the inspection result node and the body state node based on the word frequency and the reverse file frequency; and set the weight of the edge connecting the inspection item node and the body state node to a target value if there is an edge between the inspection item node and the body state node for any inspection item node and any body state node.

[0170] In one possible implementation, the underwriting result acquisition module 1203 is used to encode multiple nodes in the target graph network to obtain node features corresponding to each node; obtain the adjacency matrix and degree matrix of the target graph network based on the edge weights; input the node features of the multiple nodes, the adjacency matrix and degree matrix of the target graph network into the underwriting model; perform graph convolution on the node features of the multiple nodes, the adjacency matrix and degree matrix of the target graph network through the underwriting model to obtain the graph network features of the target graph network; perform full connection and normalization on the graph network features, and output the underwriting result of the target object.

[0171] In one possible implementation, the underwriting result acquisition module 1203 is configured to perform any of the following: Embedding encoding is performed on multiple nodes in the target graph network to obtain the node features corresponding to each node; Semantic encoding is performed on multiple nodes in the target graph network to obtain the node features corresponding to each node.

[0172] In one possible implementation, the underwriting result acquisition module 1203 is used to regularize the adjacency matrix using the degree matrix through the underwriting model to obtain the regularized matrix of the adjacency matrix; and to process the regularized matrix of the adjacency matrix and the node features of the multiple nodes using the convolution weights of the underwriting model to obtain the graph network features of the target graph network.

[0173] In one possible implementation, the device further includes: The model parameter adjustment module is used to obtain the medical examination reports of multiple sample objects and the historical underwriting results of the multiple sample objects; generate a sample graph network for each of the sample objects based on the medical examination reports of the multiple sample objects; input the sample graph network of each of the sample objects into the underwriting model, and output the predicted underwriting result of each of the sample objects through the underwriting model; and adjust the model parameters of the underwriting model based on the predicted underwriting result of each of the sample objects and the corresponding historical underwriting result.

[0174] It should be noted that the underwriting device provided in the above embodiments, when performing underwriting based on medical examination reports, is only illustrated by the division of the above functional modules. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the computer device can be divided into different functional modules to complete all or part of the functions described above. Furthermore, the underwriting device and underwriting method embodiments provided in the above embodiments belong to the same concept, and their specific implementation process is detailed in the method embodiments, which will not be repeated here.

[0175] The technical solution provided in this application embodiment enables the acquisition of the target object's medical examination report during underwriting. Based on the medical examination report, a target graph network is generated. Since the target graph network includes examination item nodes, examination result nodes, and physical status nodes, it can fully reflect the target object's physical condition. By processing the target graph network, the underwriting result of the target object can be obtained. The underwriting process does not require manual intervention, reducing the manpower and material resources consumed and lowering the underwriting cost.

[0176] This application provides a computer device for performing the above-described method. This computer device can be implemented as a terminal or a server. The structure of the terminal will be described below: Figure 13 This is a schematic diagram of the structure of a terminal provided in an embodiment of this application. The terminal 1300 can be a smartphone, tablet computer, laptop computer, desktop computer, smartwatch, etc., but is not limited to these.

[0177] Typically, terminal 1300 includes one or more processors 1301 and one or more memories 1302.

[0178] Processor 1301 may include one or more processing cores, such as a quad-core processor, an octa-core processor, etc. Processor 1301 may be implemented using at least one hardware form selected from DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), and PLA (Programmable Logic Array). Processor 1301 may also include a main processor and a coprocessor. The main processor, also known as a CPU (Central Processing Unit), is used to process data in the wake-up state; the coprocessor is a low-power processor used to process data in the standby state. In some embodiments, processor 1301 may integrate a GPU (Graphics Processing Unit), which is responsible for rendering and drawing the content to be displayed on the screen. In some embodiments, processor 1301 may also include an AI (Artificial Intelligence) processor, which is used to handle computational operations related to machine learning.

[0179] The memory 1302 may include one or more computer-readable storage media, which may be non-transitory. The memory 1302 may also include high-speed random access memory and non-volatile memory, such as one or more disk storage devices or flash memory devices. In some embodiments, the non-transitory computer-readable storage media in the memory 1302 are used to store at least one computer program, which is executed by the processor 1301 to implement the underwriting method provided in the method embodiments of this application.

[0180] In some embodiments, the terminal 1300 may also optionally include a peripheral device interface 1303 and at least one peripheral device. The processor 1301, memory 1302, and peripheral device interface 1303 can be connected via a bus or signal line. Each peripheral device can be connected to the peripheral device interface 1303 via a bus, signal line, or circuit board. Specifically, the peripheral device includes at least one of the following: a radio frequency circuit 1304, a display screen 1305, a camera assembly 1306, an audio circuit 1307, and a power supply 1309.

[0181] Peripheral interface 1303 can be used to connect at least one I / O (Input / Output) related peripheral device to processor 1301 and memory 1302. In some embodiments, processor 1301, memory 1302 and peripheral interface 1303 are integrated on the same chip or circuit board; in some other embodiments, any one or two of processor 1301, memory 1302 and peripheral interface 1303 can be implemented on separate chips or circuit boards, which is not limited in this embodiment.

[0182] The radio frequency (RF) circuit 1304 is used to receive and transmit RF (Radio Frequency) signals, also known as electromagnetic signals. The RF circuit 1304 communicates with communication networks and other communication devices via electromagnetic signals. The RF circuit 1304 converts electrical signals into electromagnetic signals for transmission, or converts received electromagnetic signals back into electrical signals. Optionally, the RF circuit 1304 includes: an antenna system, an RF transceiver, one or more amplifiers, a tuner, an oscillator, a digital signal processor, a codec chipset, a user identity module card, etc.

[0183] Display screen 1305 is used to display a user interface (UI). This UI may include graphics, text, icons, video, and any combination thereof. When display screen 1305 is a touch display screen, it also has the ability to collect touch signals on or above its surface. These touch signals can be input as control signals to processor 1301 for processing. In this case, display screen 1305 can also be used to provide virtual buttons and / or a virtual keyboard, also known as soft buttons and / or a soft keyboard.

[0184] The camera assembly 1306 is used to capture images or videos. Optionally, the camera assembly 1306 includes a front-facing camera and a rear-facing camera. Typically, the front-facing camera is located on the front panel of the terminal, and the rear-facing camera is located on the back of the terminal.

[0185] The audio circuit 1307 may include a microphone and a speaker. The microphone is used to collect sound waves from the user and the environment, and convert the sound waves into electrical signals that are input to the processor 1301 for processing, or input to the radio frequency circuit 1304 to realize voice communication.

[0186] Power supply 1309 is used to supply power to the various components in terminal 1300. Power supply 1309 can be AC ​​power, DC power, a disposable battery, or a rechargeable battery.

[0187] In some embodiments, the terminal 1300 further includes one or more sensors 1310. The one or more sensors 1310 include, but are not limited to: an acceleration sensor 1311, a gyroscope sensor 1312, a pressure sensor 1313, an optical sensor 1315, and a proximity sensor 1316.

[0188] Accelerometer 1311 can detect the magnitude of acceleration on the three coordinate axes of a coordinate system established with terminal 1300.

[0189] The gyroscope sensor 1312 can detect the orientation and rotation angle of the terminal 1300. The gyroscope sensor 1312 can work in conjunction with the accelerometer sensor 1311 to collect the user's 3D movements on the terminal 1300.

[0190] The pressure sensor 1313 can be installed on the side bezel of the terminal 1300 and / or on the lower layer of the display screen 1305. When the pressure sensor 1313 is installed on the side bezel of the terminal 1300, it can detect the user's grip signal on the terminal 1300, and the processor 1301 can perform left / right hand recognition or quick operation based on the grip signal collected by the pressure sensor 1313. When the pressure sensor 1313 is installed on the lower layer of the display screen 1305, the processor 1301 can control the operable controls on the UI interface based on the user's pressure operation on the display screen 1305.

[0191] The optical sensor 1315 is used to collect ambient light intensity. In one embodiment, the processor 1301 can control the display brightness of the display screen 1305 based on the ambient light intensity collected by the optical sensor 1315.

[0192] The proximity sensor 1316 is used to detect the distance between the user and the front of the terminal 1300.

[0193] Those skilled in the art will understand that Figure 13 The structure shown does not constitute a limitation on terminal 1300 and may include more or fewer components than shown, or combine certain components, or use different component arrangements.

[0194] The aforementioned computer equipment can also be implemented as a server. The structure of a server is described below: Figure 14 This is a schematic diagram of a server structure provided in an embodiment of this application. The server 1400 can vary significantly due to different configurations or performance. It may include one or more Central Processing Units (CPUs) 1401 and one or more memories 1402. The one or more memories 1402 store at least one computer program, which is loaded and executed by the one or more processors 1401 to implement the methods provided in the various method embodiments described above. Of course, the server 1400 may also have wired or wireless network interfaces, a keyboard, and input / output interfaces for input and output. The server 1400 may also include other components for implementing device functions, which will not be elaborated upon here.

[0195] In an exemplary embodiment, a computer-readable storage medium is also provided, such as a memory including a computer program that can be executed by a processor to perform the underwriting method in the above embodiments. For example, the computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a compact disc read-only memory (CD-ROM), magnetic tape, floppy disk, and optical data storage device, etc.

[0196] In an exemplary embodiment, a computer program product or computer program is also provided, which includes program code stored in a computer-readable storage medium. A processor of a computer device reads the program code from the computer-readable storage medium and executes the program code, causing the computer device to perform the aforementioned underwriting method.

[0197] In some embodiments, the computer program involved in the present application embodiments may be deployed and executed on a computer device, or executed on multiple computer devices located in one location, or executed on multiple computer devices distributed in multiple locations and interconnected through a communication network. Multiple computer devices distributed in multiple locations and interconnected through a communication network may constitute a blockchain system.

[0198] Those skilled in the art will understand that all or part of the steps of the above embodiments can be implemented by hardware or by a program instructing related hardware. The program can be stored in a computer-readable storage medium, such as a read-only memory, a disk, or an optical disk.

[0199] The above are merely optional embodiments of this application and are not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. An underwriting method, characterized in that, The method includes: Obtain the physical examination report of the target object, the physical examination report including the examination items and the corresponding examination results of the examination items; An edge is generated between the inspection item node corresponding to the inspection item and the inspection result node corresponding to the inspection result, and an edge is generated between the inspection item node and the body state node of the target object corresponding to the inspection result, to obtain a target graph network. The edges in the target graph network are used to connect the interconnected nodes. In the target graph network, the body state node is connected to the corresponding inspection result node by connecting the inspection item node. Multiple nodes in the target graph network are encoded to obtain the node features corresponding to each node; based on the edge weights in the target graph network and the node features, the underwriting result of the target object is obtained, where the weights are used to reflect the degree of correlation between two connected nodes.

2. The method according to claim 1, characterized in that, The obtained target graph network includes any of the following: Obtain the examination items and corresponding examination results from the physical examination report of the target object; obtain a first named entity from the examination results, the first named entity being used to describe the physical condition of the target object; The target graph network is generated based on the inspection items, the inspection results, and the first named entity; The examination items and corresponding examination results are obtained from the physical examination report of the target object; the first named entity and the corresponding second named entity are obtained from the examination results, wherein the second named entity is used to describe the attributes of the corresponding first named entity; the target graph network is generated based on the examination items, the examination results, the first named entity and the corresponding second named entity.

3. The method according to claim 2, characterized in that, The step of obtaining the first named entity and the corresponding second named entity from the inspection result includes: Named entity recognition is performed on the inspection results to obtain the first named entity and multiple second named entities; Obtain the first position information of the first named entity in the inspection result, and the second position information of the plurality of second named entities in the inspection result; Based on the first location information and the second location information, a second named entity corresponding to the first named entity is obtained from the plurality of second named entities.

4. The method according to claim 1, characterized in that, The method for determining the weight of the edge includes: For any inspection result node and its corresponding body state node, obtain the word frequency of the text corresponding to the body state node in the text corresponding to the inspection result node; obtain the reverse file frequency of the text corresponding to the body state node in the corpus, the corpus including multiple text files; based on the word frequency and the reverse file frequency, determine the weight of the edge connecting the inspection result node and the body state node; For any inspection item node and any body status node, if there is an edge between the inspection item node and the body status node, the weight of the edge connecting the inspection item node and the body status node is set to the target value.

5. The method according to claim 1, characterized in that, The process of obtaining the underwriting result of the target object based on the edge weights and node features in the target graph network includes: Based on the weights of the edges, the adjacency matrix and degree matrix of the target graph network are obtained; The node features of the multiple nodes, the adjacency matrix and degree matrix of the target graph network are input into the underwriting model. The underwriting model performs graph convolution on the node features of the multiple nodes, the adjacency matrix and degree matrix of the target graph network to obtain the graph network features of the target graph network. The graph network features are then fully connected and normalized to output the underwriting result of the target object.

6. The method according to claim 5, characterized in that, Encoding multiple nodes in the target graph network to obtain the node features corresponding to each node includes any one of the following: Embedding encoding is performed on multiple nodes in the target graph network to obtain the node features corresponding to each node; Semantic encoding is performed on multiple nodes in the target graph network to obtain the node features corresponding to each node.

7. The method according to claim 5, characterized in that, The graph network features of the target graph network obtained by performing graph convolution on the node features of the multiple nodes, the adjacency matrix, and the degree matrix of the target graph network through the underwriting model include: Using the aforementioned underwriting model, the adjacency matrix is ​​regularized using the degree matrix to obtain the regularized matrix of the adjacency matrix; the convolution weights of the underwriting model are used to process the regularized matrix of the adjacency matrix and the node features of the multiple nodes to obtain the graph network features of the target graph network.

8. The method according to claim 5, characterized in that, The method further includes: Obtain the medical examination reports of multiple sample subjects and the historical underwriting results of the multiple sample subjects; Based on the physical examination reports of the multiple sample objects, a sample graph network is generated for each of the sample objects; The sample graph network of each of the sample objects is input into the underwriting model, and the underwriting model outputs the predicted underwriting results for each of the sample objects. Based on the predicted underwriting results and corresponding historical underwriting results for each of the sample objects, the model parameters of the underwriting model are adjusted.

9. An underwriting device, characterized in that, The device includes: The physical examination report acquisition module is used to acquire the physical examination report of the target object. The physical examination report includes the examination items and the corresponding examination results of the examination items. The graph network generation module is used to generate edges between the inspection item nodes corresponding to the inspection items and the inspection result nodes corresponding to the inspection results, and to generate edges between the inspection item nodes and the body state nodes of the target object corresponding to the inspection results, thereby obtaining a target graph network. The edges in the target graph network are used to connect mutually related nodes. In the target graph network, the body state nodes are connected to the corresponding inspection result nodes by connecting the inspection item nodes. The underwriting result acquisition module is used to encode multiple nodes in the target graph network to obtain the node features corresponding to each node; based on the weights of the edges in the target graph network and the node features, the underwriting result of the target object is obtained, wherein the weights are used to reflect the correlation between two connected nodes.

10. The apparatus according to claim 9, characterized in that, The graph network generation module is configured to perform any of the following: Obtain the examination items and corresponding examination results from the physical examination report of the target object; obtain a first named entity from the examination results, the first named entity being used to describe the physical condition of the target object; The target graph network is generated based on the inspection items, the inspection results, and the first named entity; The examination items and corresponding examination results are obtained from the physical examination report of the target object; the first named entity and the corresponding second named entity are obtained from the examination results, wherein the second named entity is used to describe the attributes of the corresponding first named entity; the target graph network is generated based on the examination items, the examination results, the first named entity and the corresponding second named entity.

11. The apparatus according to claim 10, characterized in that, The graph network generation module is used for: Named entity recognition is performed on the inspection results to obtain the first named entity and multiple second named entities; Obtain the first position information of the first named entity in the inspection result, and the second position information of the plurality of second named entities in the inspection result; Based on the first location information and the second location information, a second named entity corresponding to the first named entity is obtained from the plurality of second named entities.

12. The apparatus according to claim 9, characterized in that, The device further includes a weight determination module, used for: For any inspection result node and its corresponding body status node, obtain the word frequency of the text corresponding to the body status node in the text corresponding to the inspection result node; Obtain the reverse file frequency of the text corresponding to the body state node in the corpus, the corpus including multiple text files; based on the word frequency and the reverse file frequency, determine the weight of the edge connecting the inspection result node and the body state node; For any inspection item node and any body status node, if there is an edge between the inspection item node and the body status node, the weight of the edge connecting the inspection item node and the body status node is set to the target value.

13. The apparatus according to claim 9, characterized in that, The underwriting result acquisition module is also used for: Based on the weights of the edges, the adjacency matrix and degree matrix of the target graph network are obtained; The node features of the multiple nodes, the adjacency matrix and degree matrix of the target graph network are input into the underwriting model. The underwriting model performs graph convolution on the node features of the multiple nodes, the adjacency matrix and degree matrix of the target graph network to obtain the graph network features of the target graph network. The graph network features are fully connected and normalized to output the underwriting result of the target object.

14. The apparatus according to claim 13, characterized in that, The underwriting result acquisition module is used to perform any of the following: Embedding encoding is performed on multiple nodes in the target graph network to obtain the node features corresponding to each node; Semantic encoding is performed on multiple nodes in the target graph network to obtain the node features corresponding to each node.

15. The apparatus according to claim 13, characterized in that, The underwriting result acquisition module is used for: Using the aforementioned underwriting model, the adjacency matrix is ​​regularized using the degree matrix to obtain the regularized matrix of the adjacency matrix; the convolution weights of the underwriting model are used to process the regularized matrix of the adjacency matrix and the node features of the multiple nodes to obtain the graph network features of the target graph network.

16. The apparatus according to claim 13, characterized in that, The device further includes a model parameter adjustment module, used for: Obtain the medical examination reports of multiple sample subjects and the historical underwriting results of the multiple sample subjects; Based on the physical examination reports of the multiple sample objects, a sample graph network is generated for each of the sample objects; The sample graph network of each of the sample objects is input into the underwriting model, and the underwriting model outputs the predicted underwriting results for each of the sample objects. Based on the predicted underwriting results and corresponding historical underwriting results for each of the sample objects, the model parameters of the underwriting model are adjusted.

17. A computer device, characterized in that, The computer device includes one or more processors and one or more memories, wherein at least one computer program is stored in the one or more memories, and the computer program is loaded and executed by the one or more processors to implement the underwriting method as described in any one of claims 1 to 8.

18. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores at least one computer program, which is loaded and executed by a processor to implement the underwriting method as described in any one of claims 1 to 8.

19. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the underwriting method according to any one of claims 1 to 8.

Images