Method and device for detecting object with medical insurance fraud behavior

Abstract

The invention discloses a method and device for detecting an object with a medical insurance fraud behavior. The method comprises the following steps: acquiring medical insurance settlement data generated in a preset period, the medical insurance settlement data at least comprising a window identifier; acquiring a first type of face image associated with the window identifier, and acquiring a second type of face image associated with the medical insurance card; determining a first type of face image different from the second type of face image as a to-be-detected object, and obtaining target medical insurance settlement data associated with the to-be-detected object; generating a knowledge graph according to the target medical insurance settlement data; scoring each to-be-detected object according to the knowledge graph to obtain a score value of each to-be-detected object, and determining a plurality of to-be-detected objects corresponding to a plurality of score values in a preset score interval as a to-be-detected group; and determining a target doctor and a target medical institution with the medical insurance fraud behavior according to the to-be-detected group and the knowledge graph.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to a method and device for detecting an object that has committed medical insurance fraud. Background Art

[0002] At present, elderly insured persons and insured persons with some special disease groups can entrust other people to collect medicines on their behalf at the pharmacy or drug warehouse of designated medical institutions with prescriptions; the above-mentioned other people who collect medicines on behalf of others are called drug dispensers; therefore, there are medical insurance frauds that purchase large quantities of medicines in the name of drug dispensers and then resell them for cash; in related technologies, only a single object is detected and identified, therefore, there are problems such as incomplete detection, low accuracy in identifying medical insurance fraud, and inability to associate objects with medical insurance fraud.

[0003] To address the above-mentioned problems, no effective solution has been proposed yet. Summary of the invention

[0004] The embodiments of the present application provide a method and device for detecting objects that have committed medical insurance fraud, so as to at least solve the technical problem of low recognition accuracy caused by the method for detecting objects that have committed medical insurance fraud in the related art only considering a single factor during detection.

[0005] According to one aspect of an embodiment of the present application, a method for detecting an object that has committed medical insurance fraud is provided, comprising: obtaining medical insurance settlement data generated within a preset period, wherein the medical insurance settlement data at least includes: a window identification, a medical insurance card that generated the consumption behavior, wherein the window identification is used to indicate the window where the consumption behavior was generated; obtaining a first type of facial image associated with the window identification, and obtaining a second type of facial image associated with the medical insurance card, wherein the first type of facial image is a facial image collected by a monitoring device of the consumption window within a preset period, and the second type of facial image is a facial image of an object bound to the medical insurance card; determining a first type of facial image different from the second type of facial image as an object to be detected, and obtaining target medical insurance settlement data associated with the object to be detected, wherein In the method, the target medical insurance settlement data includes at least: a target medical insurance card associated with the consumption behavior of the object to be detected, and multidimensional data associated with the target medical insurance card; a knowledge graph is generated according to the target medical insurance settlement data, wherein the knowledge graph is generated by determining multiple objects to be detected, multiple target medical insurance cards, and multidimensional data as multiple entities, and connecting multiple entities with associated relationships; each object to be detected is scored according to the knowledge graph to obtain a score value for each object to be detected, and multiple objects to be detected corresponding to multiple score values ​​in a preset score range are determined as a group to be detected, wherein the group to be detected is a group suspected of medical insurance fraud; target physicians and target medical institutions that have committed medical insurance fraud are determined according to the group to be detected and the knowledge graph.

[0006] Optionally, the multidimensional data includes: the insured person bound to each target medical insurance card, and the electronic prescription text data associated with each target medical insurance card, wherein the electronic prescription text data includes: the name of the drug, the medical institution, and the physician; the association relationship includes at least: a drug dispensing relationship, a doctor-patient relationship, a drug purchasing relationship, an employment relationship, and a prescription relationship; connecting multiple entities with an association relationship, including: when the target medical insurance card associated with the insured person and the target medical insurance card corresponding to the object to be tested are the same medical insurance card, determining that there is a drug dispensing relationship between the entity representing the insured person and the entity representing the object to be tested; when the target medical insurance card corresponding to the insured person and the target medical insurance card associated with the physician are the same medical insurance card, determining that there is a drug dispensing relationship between the entity representing the insured person and the entity representing the object to be tested. There is a doctor-patient relationship between the human entity and the entity representing the physician; when the target medical insurance card associated with the consumption behavior of the subject to be tested and the target medical insurance card associated with the electronic prescription text data are the same medical insurance card, it is determined that there is a drug purchasing relationship between the entity representing the subject to be tested and the entity representing the electronic prescription text data; when the electronic prescription text data associated with the physician is the same as the electronic prescription text data associated with the medical institution, it is determined that there is an employment relationship between the entity representing the physician and the entity representing the medical institution; when the name of the drug is the same as the name of any category of drugs contained in the electronic prescription text data, it is determined that there is a prescription relationship between the entity representing the drug name and the entity representing the electronic prescription text data.

[0007] Optionally, scoring each object to be detected according to the knowledge graph includes: determining scoring elements according to the knowledge graph, wherein the scoring elements include: first-category attribute data of each entity, second-category attribute data of each association relationship, the first-category attribute data is used to describe data related to drug purchasing behavior of each entity, and the second-category attribute data is used to describe data related to drug purchasing behavior jointly generated by multiple entities with an association relationship; determining a score for each object to be detected according to a functional relationship between multiple scoring elements.

[0008] Optionally, target physicians and target medical institutions that have committed medical insurance fraud are determined based on the group to be detected and the knowledge graph, including: generating a fully connected graph based on the knowledge graph, wherein the nodes in the fully connected graph are target entities used to represent the insured person bound to the target medical insurance card, and there is a connection relationship between any two nodes in the fully connected graph; determining a target object in the group to be detected based on the fully connected graph, wherein the target object is the object to be detected that has committed medical insurance fraud; and determining the target physician and target medical institution based on the target object and the knowledge graph.

[0009] Optionally, a fully connected graph is generated based on the knowledge graph, including: determining the target entity in the knowledge graph, and determining multiple consumption times corresponding to each target medical insurance card; connecting multiple target entities corresponding to multiple consumption times whose differences are less than a preset value to generate an association graph; determining a target adjacent node among multiple adjacent nodes of each target entity, and connecting multiple target adjacent nodes to generate a fully connected graph, wherein the target adjacent node has an adjacent node with a connection relationship with each node in the association graph.

[0010] Optionally, the target object is determined in the group to be detected according to the fully connected graph, including: obtaining the identity information of each insured person, determining the object to be filtered among the target objects according to the identity information, wherein the object to be filtered is a target object that has a target social relationship with the insured person, and the target social relationship includes at least: a blood relationship; for other objects to be detected in the group to be detected except the object to be filtered, determining the embedding vector of each other object to be detected according to the knowledge graph and a preset path, wherein the embedding vector is used to indicate the association relationship between other objects to be detected and each entity in the preset path in the knowledge graph, and the preset path contains multiple entities and multiple association relationships; clustering the target object among multiple other objects to be detected according to the distance of multiple embedding vectors, wherein the number of target objects is multiple.

[0011] Optionally, the target physician and the target medical institution are determined according to the target object and the knowledge graph, including: determining the medical insurance card to be tested that has an associated relationship with the target object in the knowledge graph, and determining the physician to be tested that has an associated relationship with the medical insurance card to be tested in the knowledge graph; determining the first page ranking PageRank score of each physician to be tested according to the number of physicians to be tested and the association relationship between multiple physicians to be tested recorded in the knowledge graph; determining the physician to be tested corresponding to the first PageRank score greater than the first preset score value as the target physician; and determining the target medical institution according to the target physician and the knowledge graph.

[0012] Optionally, the target medical institution is determined based on the target physician and the knowledge graph, including: determining the medical institution to be detected that has an association relationship with the target physician in the knowledge graph; determining the second PageRank score of each medical institution to be detected based on the number of medical institutions to be detected and the association relationship between multiple medical institutions to be detected recorded in the knowledge graph; and determining the medical institution to be detected corresponding to the second PageRank score greater than the second preset score value as the target medical institution.

[0013] Optionally, after determining the target physician and the target medical institution, it also includes: determining a target chain in the knowledge graph, wherein the target chain is an edge in the knowledge graph that connects the entity corresponding to the object to be detected with medical insurance fraud, the entity representing the target physician, the entity representing the target medical institution, and the entity representing the target insured person, and the target insured person is the insured person associated with the object to be detected with medical insurance fraud in the knowledge graph.

[0014] According to another aspect of an embodiment of the present application, a device for detecting an object that has committed medical insurance fraud is also provided, including: a first acquisition module, used to acquire medical insurance settlement data generated within a preset period, wherein the medical insurance settlement data at least includes: a window identifier, a medical insurance card that generates consumption behavior, wherein the window identifier is used to indicate the window where the consumption behavior is generated; a second acquisition module, used to acquire a first type of facial image associated with the window identifier, and to acquire a second type of facial image associated with the medical insurance card, wherein the first type of facial image is a facial image collected by a monitoring device of the consumption window within a preset period, and the second type of facial image is a facial image of an object bound to the medical insurance card; a first determination module, used to determine a first type of facial image different from a second type of facial image as an object to be detected, and to acquire target medical insurance settlement data associated with the object to be detected , wherein the target medical insurance settlement data at least includes: a target medical insurance card associated with the consumption behavior of the object to be detected, and multidimensional data associated with the target medical insurance card; a generation module is used to generate a knowledge graph based on the target medical insurance settlement data, wherein the knowledge graph is generated by determining multiple objects to be detected, multiple target medical insurance cards, and multidimensional data as multiple entities, and connecting multiple entities with associated relationships; a second determination module is used to score each object to be detected according to the knowledge graph, obtain a score value for each object to be detected, and determine multiple objects to be detected corresponding to multiple score values ​​in a preset score range as a group to be detected, wherein the group to be detected is a group suspected of medical insurance fraud; a third determination module is used to determine the target physician and target medical institution that have committed medical insurance fraud based on the group to be detected and the knowledge graph.

[0015] In an embodiment of the present application, the medical insurance settlement data generated within a preset period is obtained, wherein the medical insurance settlement data at least includes: a window identification, a medical insurance card that generates consumption behavior, wherein the window identification is used to indicate the window that generates consumption behavior; a first type of facial image associated with the window identification is obtained, and a second type of facial image associated with the medical insurance card is obtained, wherein the first type of facial image is a facial image collected by a monitoring device of the consumption window within a preset period, and the second type of facial image is a facial image of an object bound to the medical insurance card; a first type of facial image different from the second type of facial image is determined as an object to be detected, and target medical insurance settlement data associated with the object to be detected is obtained, wherein the target medical insurance settlement data at least includes: a target medical insurance card associated with the consumption behavior of the object to be detected, and multidimensional data associated with the target medical insurance card; a knowledge graph is generated according to the target medical insurance settlement data, wherein the knowledge graph is generated by determining multiple objects to be detected, multiple target medical insurance cards, and multidimensional data as multiple entities, and connecting multiple entities with an associated relationship; each object to be detected is scored according to the knowledge graph to obtain a score value for each object to be detected, and multiple score values ​​in a preset score range are combined. The corresponding multiple objects to be detected are determined as the group to be detected, wherein the group to be detected is a group suspected of medical insurance fraud; the target physicians and target medical institutions that are suspected of medical insurance fraud are determined according to the group to be detected and the knowledge graph, and the medical insurance fraud is detected by taking the abnormal dispensing person (i.e., the object to be detected) as the entry point, and a knowledge graph is established based on the data related to the medical insurance card (i.e., the target medical insurance card) used by the abnormal dispensing person, and an innovative graph algorithm is used to analyze and mine the knowledge graph to identify the objects that are suspected of medical insurance fraud, thereby achieving the purpose of improving the accuracy of detecting medical insurance fraud; in addition, by mining and analyzing the association between the abnormal dispensing person (i.e., the object to be detected) and the knowledge graph, multiple objects that are suspected of medical insurance fraud (such as insured persons, physicians, medical institutions, etc.) with associated relationships are identified, thereby achieving the purpose of comprehensively analyzing the complex relationships hidden in the medical insurance settlement data, and realizing the technical effect of determining the illegal chain including the abnormal dispensing person, physicians, and medical institutions according to the abnormal dispensing person (i.e., the object to be detected), thereby solving the technical problem of low recognition accuracy caused by the method for detecting objects with medical insurance fraud in the related technology that only considers a single factor during detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0017] Figure 1 It is a hardware structure block diagram of a computer terminal for implementing a method for detecting an object having medical insurance fraud according to an embodiment of the present application;

[0018] Figure 2 is a flowchart of the steps of a method for detecting an object that has committed medical insurance fraud according to an embodiment of the present application;

[0019] Figure 3 is a scoring rule table according to an embodiment of the present application;

[0020] Figure 4 It is a structural schematic diagram of a device for detecting an object that has committed medical insurance fraud according to an embodiment of the present application. DETAILED DESCRIPTION

[0021] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present application.

[0022] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0023] In order to better understand the embodiments of the present application, the technical terms involved in the embodiments of the present application are explained as follows:

[0024] Drug dispenser: When insured persons who are elderly or have some special diseases entrust other persons to pick up drugs on their behalf at the pharmacy or drug warehouse of designated medical institutions based on prescriptions, the other persons entrusted are called drug dispensers.

[0025] In the related art, only a single object is detected to determine whether there is medical insurance fraud, and multiple objects with medical insurance fraud cannot be associated. There are problems such as low detection accuracy and inability to determine whether there are multiple objects that have jointly committed medical insurance fraud through a single detection. In order to solve this problem, the embodiments of the present application provide relevant solutions, which are described in detail below.

[0026] According to an embodiment of the present application, a method embodiment for detecting an object that has committed medical insurance fraud is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0027] The method embodiments provided in the embodiments of the present application can be executed in a mobile terminal, a computer terminal or a similar computing device. Figure 1 The hardware structure block diagram of a computer terminal for implementing a method for detecting an object with medical insurance fraud is shown. Figure 1 As shown, the computer terminal 10 may include one or more (102a, 102b, ..., 102n are used to illustrate) processors 102 (the processor 102 may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA), a memory 104 for storing data, and a transmission device 106 for communication functions. In addition, it may also include: a display, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of the BUS bus), a network interface, a power supply and / or a camera. It can be understood by those skilled in the art that Figure 1 The structure shown is only for illustration and does not limit the structure of the above electronic device. Figure 1 More or fewer components as shown, or with Figure 1 Different configurations are shown.

[0028] It should be noted that the one or more processors 102 and / or other data processing circuits described above may generally be referred to herein as "data processing circuits". The data processing circuits may be embodied in whole or in part as software, hardware, firmware, or any other combination thereof. In addition, the data processing circuit may be a single independent processing module, or may be incorporated in whole or in part into any of the other components in the computer terminal 10. As described in the embodiments of the present application, the data processing circuit acts as a processor control (e.g., selection of a variable resistor terminal path connected to an interface).

[0029] The memory 104 can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the method for detecting an object with medical insurance fraud in the embodiment of the present application. The processor 102 executes various functional applications and data processing by running the software programs and modules stored in the memory 104, that is, the above-mentioned method for detecting an object with medical insurance fraud is realized. The memory 104 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some examples, the memory 104 may further include a memory remotely arranged relative to the processor 102, and these remote memories may be connected to the computer terminal 10 via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0030] The transmission device 106 is used to receive or send data via a network. The specific example of the above network may include a wireless network provided by a communication provider of the computer terminal 10. In one example, the transmission device 106 includes a network adapter (Network Interface Controller, NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission device 106 can be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0031] The display may be, for example, a touch screen liquid crystal display (LCD) that enables a user to interact with a user interface of the computer terminal 10 .

[0032] The embodiment of the present application provides a corresponding method for detecting medical insurance fraud that can be run in the above-mentioned operating environment. Figure 2 is a flowchart of the steps of a method for detecting an object having medical insurance fraud according to an embodiment of the present application, such as Figure 2 As shown, the method comprises the following steps:

[0033] Step S202, obtaining the medical insurance settlement data generated within a preset period, wherein the medical insurance settlement data at least includes: a window identification, and a medical insurance card that generates the consumption behavior, wherein the window identification is used to indicate the window where the consumption behavior occurs.

[0034] The method provided in the embodiment of the present application uses abnormal drug dispensing persons as the entry point to determine the objects of medical insurance fraud. The above-mentioned objects include multiple types of participants related to medical insurance card consumption, such as drug dispensing persons, doctors, medical institutions, and insured persons. In step S202, first, data related to consumption behaviors occurring with medical insurance cards during a detection period (i.e., a preset period) are obtained (i.e., medical insurance settlement data). In this embodiment, data related to consumption behaviors occurring with medical insurance cards (i.e., medical insurance settlement data) not only include each medical insurance card that generates consumption behaviors, but also include window identifiers for indicating consumption windows, wherein each consumption window has a unique window identifier, and a consumption window is a window that uses medical insurance cards for settlement, and the consumption window exists in medical institutions (such as hospitals and pharmacies).

[0035] The method provided in the embodiment of the present application can be executed by any machine learning model, a device / electronic module storing a machine learning model, for example, it can be executed by a neural network supporting the machine learning model. In step 202, the neural network supporting the machine learning model can be loaded into the memory, for example, the raw data of the neural network supporting the machine learning model can be loaded from the non-volatile memory into the volatile memory, so that the processor runs the neural network supporting the machine learning model. The raw data of the neural network supporting the machine learning model refers to unprocessed data, and generally includes parameters and structural data of the neural network supporting the machine learning model. The structural data can be a calculation relationship based on the parameters, such as the forward propagation calculation relationship between the intermediate layers and between the neurons. Specifically, the structural data can include code related to the structure of the neural network supporting the machine learning model, such as code for performing related calculations between the intermediate layers and between the neurons.

[0036] In one embodiment, an area for loading a neural network supporting a machine learning model can be divided in the memory, which may include a structure data storage area and a parameter storage area. The structure data storage area is used to store structure-related codes, and the parameters referenced by it can point to the address of a specific parameter in the parameter storage area through a pointer. During the training process of the neural network supporting the machine learning model, it may be necessary to frequently update the parameters, and the parameter values ​​in the parameter storage area can be updated.

[0037] Step S204, obtain the first type of facial image associated with the window identifier, and obtain the second type of facial image associated with the medical insurance card, wherein the first type of facial image is the facial image collected by the monitoring equipment of the consumption window within a preset period, and the second type of facial image is the facial image of the object bound to the medical insurance card.

[0038] The embodiment of the present application determines the person who dispenses the medicine on behalf of others (i.e., the object to be detected) from among multiple objects that generate consumption behaviors within the detection period by comparing facial images, wherein the multiple objects that generate consumption behaviors within the detection period can be determined by the monitoring records of the consumption window. Therefore, after the window identification is obtained in step S202, in step S204, the face images of the people who purchase medicines recorded in the consumption window corresponding to each window identification within the detection period (i.e., the preset period) (i.e., the first type of face images) are obtained. The above-mentioned face images of the people who purchase medicines are recorded by the face detection device of the consumption window, and are the facial images of the objects that perform consumption behaviors in the consumption window within the detection period. After obtaining the face images of the people who purchase medicines within the detection period (i.e., the first type of face images), for each medical insurance card that generates consumption records within the detection period, the face images of the objects that use the medical insurance card for consumption (i.e., the first type of face images) and the face images of the objects bound to the medical insurance card (i.e., the insured person) (i.e., the second type of face images) are compared, and the person who dispenses the medicine on behalf of others (i.e., the object to be detected) is determined from among the multiple objects that generate consumption behaviors within the detection period according to the comparison results. In this embodiment, the face detection device can be a continuously running video recording device, or it can be a capture device (such as a camera) triggered by the card swiping signal of the medical insurance card. When the medical insurance card is swiped, the drug purchase signal is sent to the capture device through the drug purchase system. The camera triggers the capture immediately after obtaining the drug purchase signal, thereby ensuring that the system capture time is consistent with the drug purchase time.

[0039] It should be noted that the information collected in the embodiments of the present application (such as the above-mentioned first-category facial images, second-category facial images, medical insurance settlement data, and any subsequent multi-dimensional data associated with the medical insurance card, and any data needed to implement this solution) are all information and data authorized by the user or "fully authorized" by all parties; a corresponding "operation entrance" is provided during collection for the user to choose to authorize or refuse; and the collection, storage, use, processing, transmission, provision, disclosure and application of relevant data comply with the relevant laws, regulations and standards of the relevant countries and regions, and necessary confidentiality measures are taken without violating public order and good customs.

[0040] Step S206, determine the first type of facial image different from the second type of facial image as the object to be detected, and obtain target medical insurance settlement data associated with the object to be detected, wherein the target medical insurance settlement data at least includes: a target medical insurance card associated with the consumption behavior of the object to be detected, and multidimensional data associated with the target medical insurance card.

[0041] When comparing the facial images of the object bound to the medical insurance card (i.e., the second type of facial images) and the facial images of the object using the medical insurance card for card payment (i.e., the first type of facial images), the following two comparison results will be obtained: ① The two types of facial images associated with a medical insurance card (the first type of facial images and the second type of facial images) are the same, and ② The two types of facial images associated with a medical insurance card are different. In step S206, when the above-mentioned second type of comparison result appears during the facial image comparison process, the first type of facial image of the two facial images that are compared to obtain the second type of comparison result is determined as the person who dispenses the medicine on behalf of others (i.e., the object to be detected); by obtaining the medical insurance settlement data of the person who dispenses the medicine on behalf of others (i.e., the target medical insurance settlement data), a knowledge graph is established, and finally the object of medical insurance fraud is determined. Among them. Whether the first type of facial image is the same as the second type of facial image can be determined by the similarity of the two facial images. For example, for the same medical insurance card, if the similarity between the first type of facial image and the second type of facial image associated therewith is greater than or equal to a preset similarity (such as 80%), the two facial images are determined to be the same; otherwise, if the similarity between the first type of facial image and the second type of facial image associated therewith is less than the above-mentioned preset similarity (such as 80%), the two facial images are determined to be different. In this embodiment, the medical insurance settlement data of the person dispensing medicine (i.e., the target medical insurance settlement data) includes: the medical insurance card used by the person dispensing medicine (i.e., the target medical insurance card), and multidimensional data associated with the (target) medical insurance card: such as the binding object of the target medical insurance card, prescription data settled using the target medical insurance card, etc.

[0042] Step S208, generating a knowledge graph based on the target medical insurance settlement data, wherein the knowledge graph is generated by determining multiple objects to be detected, multiple target medical insurance cards, and multi-dimensional data as multiple entities, and connecting multiple entities with associated relationships.

[0043] In step S208, when generating a knowledge graph based on the medical insurance settlement data (i.e., target medical insurance settlement data) of the person dispensing medicine on behalf of the test subject (i.e., the subject to be tested), the subject to be tested, the medical insurance card used by the person dispensing medicine on behalf of the test subject (i.e., the target medical insurance card), and the multidimensional data associated with the target medical insurance card are taken as entities, and edges are generated through the association relationships between these entities. The entities with association relationships are connected to generate a knowledge graph; the knowledge graph contains multiple triples, and each triple is in the following form: entity-relationship-attribute, wherein the "attribute" data includes: data used to describe the association relationship between entities, and data used to describe entities.

[0044] According to an optional embodiment of the present application, the multidimensional data includes: the insured person bound to each target medical insurance card, and the electronic prescription text data associated with each target medical insurance card, wherein the electronic prescription text data includes: the name of the drug, the medical institution, and the physician; the association relationship includes at least: the relationship of dispensing on behalf of another, the doctor-patient relationship, the relationship of purchasing another, the relationship of employment, and the relationship of prescription; connecting multiple entities with association relationships, including: when the target medical insurance card associated with the insured person and the target medical insurance card corresponding to the object to be detected are the same medical insurance card, determining that there is a relationship of dispensing on behalf of another between the entity representing the insured person and the entity representing the object to be detected; when the target medical insurance card corresponding to the insured person and the target medical insurance card associated with the physician are the same medical insurance card, determining It is determined that there is a doctor-patient relationship between the entity representing the insured person and the entity representing the physician; when the target medical insurance card associated with the consumption behavior of the subject to be tested and the target medical insurance card associated with the electronic prescription text data are the same medical insurance card, it is determined that there is a drug purchasing relationship between the entity representing the subject to be tested and the entity representing the electronic prescription text data; when the electronic prescription text data associated with the physician is the same as the electronic prescription text data associated with the medical institution, it is determined that there is an employment relationship between the entity representing the physician and the entity representing the medical institution; when the name of the drug is the same as the name of any type of drug contained in the electronic prescription text data, it is determined that there is a prescription relationship between the entity representing the drug name and the entity representing the electronic prescription text data.

[0045] In this embodiment, the multidimensional data associated with the medical insurance card (i.e., the target medical insurance card) used by the person dispensing medicine includes: the object bound to the target medical insurance card (i.e., the insured person), the prescription data settled using the target medical insurance card (i.e., the electronic prescription text data associated with the target medical insurance card), the physician associated when settling using the target medical insurance card, the medical institution corresponding to the consumption record using the target medical insurance card, the name of the medicine purchased when settling using the target medical insurance card, etc.; wherein, the above-mentioned physicians, medical insurance cards, and medical institutions can all be queried in the electronic prescription text data. That is, in this embodiment, when constructing the knowledge graph, at least the entities are: the person dispensing medicine (V1), the insured person (V2), the physician (V3), the medical institution (V4), the medicine (name of the medicine) (V5), and the electronic prescription (V6); in addition, the face of the person purchasing medicine (V7) (i.e., the face image of the person dispensing medicine, the first type of face image) and the face of the insured person (i.e., the second type of face image) can also be used as entities. As mentioned in step S208, constructing a knowledge graph requires connecting entities with associated relationships. In this embodiment, the associated relationships between entities include: proxy dispensing relationship (R1), doctor-patient relationship (R2), drug purchasing relationship (R3), prescription relationship (R4), employment relationship (R5) and other relationships. Among them, the proxy dispensing relationship is the "proxy dispensing person-insured person" relationship, which is the associated relationship between the insured person and the proxy dispensing person (i.e., the subject to be tested). When the current dispensing person and the insured person are associated with the same medical insurance card, it is determined that there is a proxy dispensing relationship between the two. The entity representing the insured person and the entity representing the person to be dispensed can be connected; among them, the target medical insurance card associated / corresponding to the proxy dispensing person is the medical insurance card used by the proxy dispensing person for settlement; the insured person is bound to the medical insurance card, that is, there is an associated relationship between the insured person and the medical insurance card. The drug purchasing relationship is the "drug dispensing agent-prescription" relationship, which is the association relationship between the drug dispensing agent (i.e., the subject to be tested) and the prescription. When the medical insurance card used to settle the prescription data and the medical insurance card used by the drug dispensing agent to settle the prescription (the target medical insurance card associated with the consumption behavior of the subject to be tested) are the same medical insurance card, it is determined that there is a drug purchasing relationship between the two, and the entity representing the drug dispensing agent and the entity representing the prescription can be connected; wherein, in this embodiment, the prescription exists in the form of electronic data, which is called electronic prescription text data, and each prescription corresponds to a set of electronic prescription text data. The medical insurance card used when settling the prescription data is the target medical insurance card associated with the electronic prescription text data. The doctor-patient relationship is the "insured person-physician" relationship, which is the relationship between the insured person and the physician. When the medical insurance card bound to the insured person and the medical insurance card associated with the physician are the same medical insurance, it is determined that there is a doctor-patient relationship between the two, and the entity representing the insured person and the entity representing the physician can be connected; wherein, when determining the medical insurance card associated with the physician, the prescription to which the physician belongs can be determined first (electronic text data, and the medical insurance card associated with the prescription is the medical insurance card associated with the physician.The employment relationship is the "physician-medical institution" relationship, which is the relationship between the physician and the medical institution. When the physician and the medical institution belong to the same prescription, it is determined that there is an employment relationship between the two. The entity representing the physician and the entity representing the medical institution can be connected; that is, there is an employment relationship between the physician and the medical institution recorded in the same prescription. The prescription relationship is the "electronic prescription-drug" relationship, which is the relationship between the drug and the prescription. When the drug is recorded in the prescription, it is determined that there is a prescription relationship between the two. The entity representing the drug and the entity representing the prescription (i.e., the electronic prescription text data) can be connected; among them, whether the drug is recorded in the prescription can be determined by matching the same drug name in the prescription. By defining these specific relationships, the knowledge graph can more accurately reflect the complex relationships in the medical insurance settlement data, which helps to identify potential medical insurance fraud, such as frequent drug dispensing, and improves the accuracy of detecting medical insurance fraud.

[0046] Step S210, scoring each object to be detected according to the knowledge graph to obtain a score value for each object to be detected, and determining multiple objects to be detected corresponding to multiple score values ​​in a preset score range as a group to be detected, wherein the group to be detected is a group suspected of medical insurance fraud.

[0047] In step S210, after the knowledge graph is generated, the knowledge graph is used to score each dispensing agent (i.e., the subject to be tested), and the scoring result (i.e., the score value) is used to determine whether the dispensing behavior of the dispensing agent complies with the regulations or involves medical insurance fraud; specifically, when the score value belongs to the preset score range, the dispensing agent corresponding to the score value is determined as a dispensing agent who may commit medical insurance fraud; if there are multiple dispensing agents of this type who may commit medical insurance fraud, the group they form is determined as the group to be tested, and the group to be tested is further analyzed in combination with the knowledge graph (i.e., the group to be tested); the group to be tested may also contain only one dispensing agent who may commit medical insurance fraud. The preset score range can be a manually set score range, or it can be determined according to the multiple score values ​​mentioned above after all the subjects to be tested are scored and the score values ​​of all the subjects to be tested are obtained; for example, the average value a and standard deviation (σ) of the multiple score values ​​can be calculated, and the interval [a-3σ, a+3σ] can be determined as the preset score range.

[0048] Optionally, scoring each object to be detected according to the knowledge graph includes: determining scoring elements according to the knowledge graph, wherein the scoring elements include: first-category attribute data of each entity, second-category attribute data of each association relationship, the first-category attribute data is used to describe data related to drug purchasing behavior of each entity, and the second-category attribute data is used to describe data related to drug purchasing behavior jointly generated by multiple entities with an association relationship; determining a score for each object to be detected according to a functional relationship between multiple scoring elements.

[0049] In step S210, the method of this embodiment may be used to score the person dispensing the medicine (i.e., the subject to be tested) according to the knowledge graph. Specifically, the following formula indicating the functional relationship between multiple scoring elements may be used: Score each dispensing agent, where score represents the score of each dispensing agent, N is the number of all insured persons associated with all the subjects to be tested (through the dispensing agent relationship), θ j represents the score of the insured person j. The score of each insured person (θ) can be determined according to the following formula: θ = log6r1 + log2r2 + r3 + log6r4 + log6r5 + log2r6 + r7, where log represents the logarithmic operation, r1, r2, r3, r4, r5, r6, r7, d i All represent scoring elements. As mentioned in the above embodiment, the knowledge graph constructed in the embodiment of the present application is composed of multiple triples (entity-relationship-attribute). In this embodiment, the "attribute" data in the knowledge graph is used as the above-mentioned scoring element to score the substitute drug dispenser; the "attribute" data includes data for describing "relationship" and data for describing "entity"; in this embodiment, the "attribute" data related to only one entity is classified as the first type of attribute data, that is, the data used to describe the "entity" is the first type of attribute data; the "attribute" data related to two entities associated with a "relationship" is classified as the second type of attribute data, that is, the data used to describe the "relationship" is the second type of attribute data. Figure 3 is a scoring rule table, based on Figure 3It can be seen that r3, r6 and r7 are data used to describe the insured person, belonging to the first type of attribute data, r3 is the violation mark associated with the insured person, used to indicate whether the insured person has been punished for using the medical insurance card illegally (if yes, r3 = 1; if not, r3 = 0); r6 is the age of the insured person in the year of drug purchase, the year of drug purchase is the year of the detection period, used to indicate the age of the object bound to the medical insurance card that generates consumption records during the detection period; r7 is the gender probability of the insured person (if male, r7 = 0.6; if female, r7 = 0.4). d1, d2, d3, d5 are data used to describe the "substitute relationship", belonging to the second type of attribute data; d1 is the number of insured persons associated with each object to be detected during the detection period, where each insured person is bound to a medical insurance card, so d1 can also represent the number of medical insurance cards used by each object to be detected during the detection period. d2 is the number of objects to be detected associated with each target medical insurance card, and the target medical insurance card is the medical insurance card used by the person dispensing the medicine on behalf of the person during the detection period. d3 is the number of times the drug is dispensed on behalf of the test object, which is used to indicate the number of times the test object uses the medical insurance card to settle the bill within the test period (i.e., the preset period). d5 is the amount of drug dispensing on behalf of the test object, which is used to indicate the specific settlement amount each time the dispensing person uses the medical insurance card to settle the bill within the test period. d4, r4, and r5 are data used to describe the "drug purchase relationship", which belongs to the second type of attribute data. d4 is the number of types of window identifiers associated with each test object, which is used to indicate how many medical institutions the test object uses the medical insurance card to consume within the test period. r4 is the amount of drug purchase, which is used to indicate the total amount of consumption generated by the insured person using the bound medical insurance card to consume within the test period. r5 is the number of drug purchases, which is used to indicate the number of times each insured person uses the bound medical insurance card to settle the bill within the test period. r1 and r2 are data used to describe the "prescription relationship", which also belongs to the second type of attribute data. r1 represents the number of types of target drugs purchased within the test period, where the target drug is recorded in the drug test list, and the drug test list can be generated by summarizing the drugs purchased by the insured person / dispensing person who has committed medical insurance fraud. r2 represents the number of target medical institutions for medical insurance settlement during the detection period, where the target medical institutions are the medical institutions (including hospitals and pharmacies) recorded in the medical institution detection list. The medical institution detection list can be generated by summarizing the medical institutions visited by insured persons / drug dispensers who have committed medical insurance fraud.In addition, in the embodiment of the present application, if the knowledge graph also includes entities representing face images (first-category face images and second-category face images), the association relationship between the entity representing the face of the person purchasing medicine (i.e., the first-category face image) and the entity representing the face of the person bound to the medical insurance card (i.e., the second-category face image) is: "Face relationship (R7)"; when the similarity between the first-category face image and the second-category face image is greater than the preset similarity, it is determined that there is a "face relationship" between the two, and the "attribute" corresponding to the face association relationship is the image similarity. In addition, in the present embodiment, the "attributes" corresponding to the doctor-patient relationship include: the number of visits and the amount of the visit; the "attributes" corresponding to the drug purchasing relationship include: the settlement serial number and the settlement amount; the "attributes" corresponding to the prescription relationship include the number of drug types and the settlement amount; the "attributes" of the employment relationship include: the doctor's suspicion score.

[0050] Step S212, determining the target physicians and target medical institutions that have committed medical insurance fraud based on the group to be detected and the knowledge graph.

[0051] After one or more dispensing agents who may have committed medical insurance fraud (i.e., the group to be detected) are determined through step S210, in step S212, the group to be detected is further analyzed in combination with the knowledge graph, and the dispensing agents who have committed medical insurance fraud are determined in the group to be detected, and the medical institutions (i.e., target medical institutions) that have committed medical insurance fraud, and the physicians (i.e., target physicians) who have committed medical insurance fraud are determined in the data recorded in the knowledge graph. In addition, it is also possible to determine whether the dispensing agents, medical institutions, and physicians who have committed medical insurance fraud are a group based on the associations recorded in the knowledge graph. If the dispensing agents, medical institutions, physicians, insured persons, etc. who have committed medical insurance fraud belong to a group, a violation chain can be determined. This violation chain is a data chain in the knowledge graph that connects entities representing dispensing agents, entities representing medical institutions, entities representing physicians, and entities representing insured persons.

[0052] Optionally, target physicians and target medical institutions that have committed medical insurance fraud are determined based on the group to be detected and the knowledge graph, including: generating a fully connected graph based on the knowledge graph, wherein the nodes in the fully connected graph are target entities used to represent the insured person bound to the target medical insurance card, and there is a connection relationship between any two nodes in the fully connected graph; determining a target object in the group to be detected based on the fully connected graph, wherein the target object is the object to be detected that has committed medical insurance fraud; and determining the target physician and target medical institution based on the target object and the knowledge graph.

[0053] In step S212, when determining the target physician and target medical institution that commits medical insurance fraud by jointly analyzing one or more drug dispensing agents who may commit medical insurance fraud (i.e., the group to be detected) and the knowledge graph, the method in this embodiment can be adopted: first, screen out the entities representing the insured persons in the knowledge graph (i.e., the target entities), and generate a fully connected graph in which each node (entity) is connected to other nodes (entities) by associating the target entities; next, further screen out the target objects in the group to be detected based on the fully connected graph, wherein the target objects are drug dispensing agents who have a greater possibility (for example, greater than 90) of committing medical insurance fraud.

[0054] According to an optional embodiment of the present application, a fully connected graph is generated based on a knowledge graph, including: determining a target entity in the knowledge graph, and determining multiple consumption moments corresponding to each target medical insurance card; connecting multiple target entities corresponding to multiple consumption moments whose differences are less than a preset value to generate an association graph; determining a target adjacent node among multiple adjacent nodes of each target entity, and connecting the multiple target adjacent nodes to generate a fully connected graph, wherein the target adjacent node is an adjacent node in the association graph that has a connection relationship with each node in the association graph.

[0055] In this embodiment, an innovative association graph algorithm is provided to construct a fully connected subgraph (i.e., a fully connected graph) of the knowledge graph. The core idea of ​​the innovative association graph algorithm is: "A medical insurance card with a record of dispensing medicine at the same pharmacy within a preset time range (such as 10 minutes, 30 minutes, 60 minutes, etc.) of a drug dispensing record is connected to this card by an edge to newly construct an association, wherein each time an association is constructed between two identical medical insurance cards, it is equivalent to adding 1 to the weight of the edge connecting the nodes representing the two medical insurance cards; through the above method, the problem of detecting medical insurance fraud is converted into finding a fully connected subgraph in the knowledge graph." The specific steps of constructing a fully connected graph through the innovative association graph algorithm are as follows: Step 1: Traverse each insured person entity (i.e., target entity) in the knowledge graph, and obtain all edges and adjacent points connected to it (entities directly connected to the insured person entity without other entities in between). Among them, the adjacent points of the insured person entity include entities representing the medical insurance card bound to the insured person, or, through the edges of the insured person entity, the entity representing the medical insurance card bound to the insured person can be found in the knowledge graph; therefore, after obtaining the edges and adjacent points connected to the insured person entity, multiple consumption times of each target medical insurance card can be queried to determine the multiple consumption times of the consumption records generated; the multiple consumption times corresponding to the target medical insurance card are determined as the multiple consumption times of the insured person bound to the target medical insurance card; in this embodiment, to construct a fully connected graph, it is first necessary to filter out the association graph composed of the insured person entity according to the consumption time in the knowledge graph through step 1, wherein the association graph is generated by connecting multiple insured person entities whose consumption time difference is less than a preset value (i.e., a preset time range). Next, step 2 is performed: traverse multiple adjacent nodes of each insured person entity to determine from these multiple adjacent nodes the nodes (i.e., the target adjacent nodes) that have a connection relationship with each node in the association graph, and connect the multiple selected target adjacent nodes to generate a fully connected graph. Through the above steps 1 and 2, multiple fully connected graphs can be screened out from the knowledge graph, each fully connected graph represents a group of insured persons who may commit medical insurance fraud, that is, multiple insured persons represented by multiple nodes contained in the same fully connected graph constitute a group of insured persons who may commit medical insurance fraud; in order to reduce the amount of data processing, in this embodiment, after obtaining multiple fully connected graphs, a deduplication operation can be performed. In addition, in order to avoid randomness affecting the results, this embodiment can limit the weight of the search edge to be greater than or equal to 2, that is, when searching for the target adjacent node used to generate the fully connected graph in the knowledge graph, only the edge with a weight value greater than 2 (connected to the insured entity) is searched.

[0056] According to another optional embodiment of the present application, the target object is determined in the group to be detected according to the fully connected graph, including: obtaining the identity information of each insured person, determining the object to be filtered among the target objects according to the identity information, wherein the object to be filtered is a target object that has a target social relationship with the insured person, and the target social relationship includes at least: a blood relationship; for other objects to be detected in the group to be detected except the object to be filtered, determining the embedding vector of each other object to be detected according to the knowledge graph and a preset path, wherein the embedding vector is used to indicate the association relationship between other objects to be detected and each entity in the preset path in the knowledge graph, and the preset path contains multiple entities and multiple association relationships; clustering the target object among multiple other objects to be detected according to the distance of multiple embedding vectors, wherein the number of target objects is multiple.

[0057] In this embodiment, a graph embedding algorithm (such as Metapath2Vec++) can be used to screen out the dispensing agents (i.e., target objects) who are more likely to commit medical insurance fraud (e.g., greater than 90%) in the knowledge graph; in this embodiment of the present application, the above-mentioned target objects are determined as dispensing agents who have committed medical insurance fraud. The core idea of ​​the graph embedding algorithm is: "Using a random walk approach, by simultaneously capturing the semantic and structural relationships between different types of nodes (entities), the deep features between different types of nodes are mined; and the association relationship between different types of nodes is further determined based on the deep features." Specifically in this embodiment, a graph embedding algorithm is used to mine the hidden relationship between the dispensing agent (i.e., the object to be detected) and the insured person, so as to exclude dispensing agents who are less likely to commit medical insurance fraud (e.g., less than 90%). In the actual drug purchasing scenario, some associated behaviors are highly suspicious, but there are also reasonable special cases. For example, illegal drug dispensers (i.e., target objects) often have the characteristic of using multiple medical insurance cards. At the same time, compliant drug dispensers (such as friends and relatives) also have such characteristics. Therefore, in order to effectively distinguish these two types of objects to be detected, a scenario rationalization judgment is added in this embodiment. Specifically, before using the graph embedding algorithm to mine the hidden relationship between the insured and the drug dispenser, the objects to be detected that have a target social relationship such as a blood relationship or a friend relationship with the drug dispenser are marked according to the identity information of the insured (gender, age, etc., which can be queried in the knowledge graph) and the identity information of the drug dispenser recorded in the knowledge graph (gender, age, etc., which can be queried in the knowledge graph). The mark is used to indicate that the graph embedding algorithm is not executed for these objects to be detected (i.e., objects to be filtered) that have a blood relationship or a friend relationship with the insured. Next, the graph embedding algorithm is executed on the unmarked objects to be detected in the detection group (i.e., other objects to be detected). In the method provided in this embodiment, when using the graph embedding algorithm for deep feature mining, the following metapath (i.e., preset path) is pre-set: Path 1: Drug purchase face (V7) R → 7 Dispenser (V1) R → 3 Electronic prescription (V6) R → 4 Medicine (V5); Path 2: Dispenser (V1) R → 1 Insured Person (V2) R → 2 Physician (V3) R → 5Drugs (V5), when executing the graph embedding algorithm, the nodes in the knowledge graph are jumped in the order defined by one or two meta-walking paths above, and the hidden relationship between the dispensing agent and the insured person is mined during the walking process. For example, when walking according to path 2, when jumping from the dispensing agent entity (node), the edge representing the dispensing relationship (R1) is selected from the multiple edges connected to the dispensing agent entity (V1), and the edge is jumped to the entity representing the insured person (V2). After walking along the preset path, a low-dimensional vector (i.e., embedding vector) is output for each other object to be detected; these low-dimensional vectors can capture the similarities and differences between entities (nodes), so that they can be used for subsequent analysis. When subsequent analysis is performed based on the low-dimensional vectors, each low-dimensional vector represents the corresponding other objects to be detected. In this embodiment, the distance between every two low-dimensional vectors is calculated, and the other objects to be detected are clustered (classified) according to the distance; a density-based spatial clustering algorithm (Density-Based Spatial Clustering of Applications with Noise, DBSCAN) can be used for clustering: the other objects to be detected represented by multiple low-dimensional vectors with a distance value less than a preset distance value are classified as substitute drug dispensers (i.e., target objects) with a high possibility (for example, greater than 90%) of committing medical insurance fraud. Through the method of the embodiment of the present application, the social association relationship between the insured person and the person dispensing the medicine on his behalf who has a correlation with the drug purchasing behavior is determined based on the identity information of the insured person, and the objects to be detected who have a blood relationship or other emotional relationship, such as family members, friends or colleagues, are filtered out. The number of insured persons in the fully connected graph involved in each person dispensing the medicine on his behalf is counted, and the group of people dispensing the medicine on his behalf with repeated insured persons in the association analysis results is screened. Based on the semantic information of the entity, the group of people dispensing the medicine on his behalf is divided by a clustering algorithm to form a community of people dispensing the medicine on his behalf composed of multiple target objects. This achieves the technical effect of being able to effectively identify abnormal drug purchasing and medical treatment patterns in scenarios where there are groups with complex social relationships.

[0058] According to some optional embodiments of the present application, a target physician and a target medical institution are determined based on a target object and a knowledge graph, including: determining in the knowledge graph a medical insurance card to be tested that has an associated relationship with the target object, and determining in the knowledge graph a physician to be tested that has an associated relationship with the medical insurance card to be tested; determining a first page ranking PageRank score for each physician to be tested based on the number of physicians to be tested and the associated relationship between multiple physicians to be tested recorded in the knowledge graph; determining a physician to be tested corresponding to a first PageRank score greater than a first preset score value as a target physician; and determining a target medical institution based on the target physician and the knowledge graph.

[0059] After multiple substitute dispensing agents (i.e., target objects) who have committed medical insurance fraud have been determined by the method in the above embodiment, the knowledge graph can be further combined to analyze the (target) medical institutions and (target) physicians who have committed medical insurance fraud. In this embodiment, a network centrality algorithm is used to identify the target physician / target physician group based on the knowledge graph and the community of substitute dispensing agents composed of multiple target objects. The specific steps are as follows: locate the entity representing the target object in the knowledge graph; further, the medical insurance card represented by the medical insurance card entity connected to the target object on an edge (connected by an edge, that is, there is an association relationship) is determined as the medical insurance card to be tested; the physician represented by the physician entity connected to the medical insurance card to be tested on an edge in the knowledge graph is the physician to be tested. The physician who has committed medical insurance fraud (i.e., the target physician) is determined among the physicians to be tested by calculating the page ranking (PageRank) score, wherein when the value of the PageRank score (i.e., the first PageRank score) of the physician to be tested is greater than the (first) preset score value, the physician to be tested is the target physician. The PageRank score of each physician to be tested is based on the formula: Calculated, where d is the damping factor, which takes a value of 0.85, N is the number of physicians to be tested, A represents the probability transfer matrix of each physician to be tested, and each element in the probability transfer matrix A is the probability of the physician to be tested jumping to each node connected to it in the knowledge graph, which can be determined by the association relationship between multiple physicians to be tested recorded in the knowledge graph (when on the same edge, the probability of jumping to the node corresponding to each physician to be tested). The above-mentioned first preset score value can be the average of the PageRank scores of all physicians to be tested; or, the first preset score value = Q3 + 1.5 × (Q3-Q1), where Q3 is the value of the (first) PageRank score ranked in the 75% position from large to small among the PageRank scores of all physicians to be tested; Q1 is the value of the (first) PageRank score ranked in the 25% position from large to small among the PageRank scores of all physicians to be tested.

[0060] According to other optional embodiments of the present application, the target medical institution is determined based on the target physician and the knowledge graph, including: determining the medical institution to be detected that has an associated relationship with the target physician in the knowledge graph; determining the second PageRank score of each medical institution to be detected based on the number of medical institutions to be detected and the associated relationship between the multiple medical institutions to be detected recorded in the knowledge graph; and determining the medical institution to be detected corresponding to the second PageRank score greater than the second preset score value as the target medical institution.

[0061] After multiple dispensing agents (i.e., target objects) who have committed medical insurance fraud are determined by the method in the above embodiment, the (target) medical institutions and (target) physicians who have committed medical insurance fraud can be further analyzed in combination with the knowledge graph. In this embodiment, a network centrality algorithm is used to identify (target) medical institutions that have committed medical insurance fraud in the knowledge graph according to the knowledge graph and a target physician group composed of multiple target physicians. The specific steps are as follows: locate the entity representing the target physician in the knowledge graph; further, determine the medical institution represented by the medical institution entity connected to the target physician on an edge (connected by an edge, that is, there is an association relationship) as the medical institution to be detected; the medical institution represented by the medical institution entity connected to the entity representing the target physician on an edge in the knowledge graph is the medical institution to be detected. The medical institution (i.e., the target medical institution) that has committed medical insurance fraud is determined in the medical institution to be detected by calculating the page ranking (PageRank) score, wherein, when the value of the PageRank score (i.e., the second PageRank score) of the medical institution to be detected is greater than the (second) preset score value, the medical institution to be detected is the target medical institution. The PageRank score of each medical institution to be tested is based on the formula: Calculated, where d is the damping factor, which takes a value of 0.85, M is the number of medical institutions to be tested, B represents the probability transfer matrix of each medical institution to be tested, and each element in the probability transfer matrix B is the probability of the medical institution to be tested jumping to each node connected to it in the knowledge graph, which can be determined by the association relationship between multiple medical institutions to be tested recorded in the knowledge graph (when on the same edge, the probability of jumping to the node corresponding to each medical institution to be tested). The above-mentioned second preset score value can be the average of the PageRank scores of all medical institutions to be tested; or, the second preset score value = Q3 ′ +1.5×(Q3 ′ -Q1 ′ ), where Q3 ′ Q1 is the value of the (second) PageRank score that ranks in the 75% position from the largest to the smallest among all the PageRank scores of the medical institutions to be tested; ′ It is the value of the (second) PageRank score that ranks in the 25th percentile from the largest to the smallest among the PageRank scores of all the doctors to be tested.

[0062] According to an optional embodiment of the present application, after determining the target physician and the target medical institution, it also includes: determining a target chain in the knowledge graph, wherein the target chain is an edge in the knowledge graph that connects an entity corresponding to the object to be detected that has medical insurance fraud, an entity representing the target physician, an entity representing the target medical institution, and an entity representing the target insured person, and the target insured person is an insured person associated with the object to be detected that has medical insurance fraud in the knowledge graph.

[0063] The method provided in the embodiment of the present application can also perform group identification, and generate a "drug dispenser-insured person-physician-medical institution" violation chain (i.e., target chain) according to the target drug dispenser (i.e., target object) group, the target insured person (group) corresponding to the target drug dispenser (group), the target physician (group), and the target medical institution (group); specifically, after the target object, the target medical institution, and the target physician are determined by the method in the above embodiment, the insured person associated with the target object can be queried in the knowledge graph; and, it is queried whether the target object, the target medical institution, and the target physician have an associated relationship (i.e., whether they are located on the same edge), and the group that cooperates to commit medical insurance fraud is summarized. The determination of the target chain can clearly show the complete chain of medical insurance fraud, enrich the information used to analyze medical insurance fraud, and improve the accuracy of detecting medical insurance fraud.

[0064] Through the above steps, we can use the abnormal person dispensing medicine as the starting point, and based on the medical insurance settlement behavior data, electronic prescription data, and face image data of the relevant insured persons, gradually focus the detection objects from the person dispensing medicine to the insured persons, doctors, and medical institutions; and based on this, we can identify the chain of violations and improve the accuracy of detecting medical insurance fraud. In addition, this method can also be applied to any other scenario field where there is a need to analyze and identify complex relationships and abnormal behaviors, such as the field of network security.

[0065] Figure 4 is a schematic diagram of a structure of a device for detecting an object having medical insurance fraud according to an embodiment of the present application, such as Figure 4As shown, the device includes: a first acquisition module 40, used to acquire medical insurance settlement data generated within a preset period, wherein the medical insurance settlement data at least includes: a window identifier, a medical insurance card that generates a consumption behavior, wherein the window identifier is used to indicate the window that generates the consumption behavior; a second acquisition module 42, used to acquire a first type of facial image associated with the window identifier, and acquire a second type of facial image associated with the medical insurance card, wherein the first type of facial image is a facial image collected by a monitoring device of the consumption window within a preset period, and the second type of facial image is a facial image of an object bound to the medical insurance card; a first determination module 44, used to determine a first type of facial image different from a second type of facial image as an object to be detected, and acquire target medical insurance settlement data associated with the object to be detected, wherein the target medical insurance settlement data at least includes : a target medical insurance card associated with the consumption behavior of the object to be detected, and multidimensional data associated with the target medical insurance card; a generation module 46, used to generate a knowledge graph based on the target medical insurance settlement data, wherein the knowledge graph is generated by determining multiple objects to be detected, multiple target medical insurance cards, and multidimensional data as multiple entities, and connecting multiple entities with an associated relationship; a second determination module 48, used to score each object to be detected according to the knowledge graph, obtain a score value for each object to be detected, and determine multiple objects to be detected corresponding to multiple score values ​​in a preset score range as a group to be detected, wherein the group to be detected is a group suspected of medical insurance fraud; a third determination module 410, used to determine the target physician and target medical institution that have committed medical insurance fraud based on the group to be detected and the knowledge graph.

[0066] It should be noted that Figure 4 The preferred implementation of the illustrated embodiment can be found in Figure 2 The relevant description of the illustrated embodiment will not be repeated here.

[0067] An embodiment of the present application also provides a non-volatile storage medium, in which a computer program is stored, wherein the above method of detecting an object that has committed medical insurance fraud is executed by running the computer program on a device where the non-volatile storage medium is located.

[0068] The above-mentioned non-volatile storage medium is used to store a program that performs the following functions: obtaining medical insurance settlement data generated within a preset period, wherein the medical insurance settlement data at least includes: a window identifier, a medical insurance card that generates consumption behavior, wherein the window identifier is used to indicate the window where the consumption behavior is generated; obtaining a first type of facial image associated with the window identifier, and obtaining a second type of facial image associated with the medical insurance card, wherein the first type of facial image is a facial image collected by a monitoring device at the consumption window within a preset period, and the second type of facial image is a facial image of an object bound to the medical insurance card; determining a first type of facial image different from the second type of facial image as an object to be detected, and obtaining target medical insurance settlement data associated with the object to be detected, wherein the target medical insurance settlement The calculation data at least includes: a target medical insurance card associated with the consumption behavior of the object to be detected, and multidimensional data associated with the target medical insurance card; a knowledge graph is generated according to the target medical insurance settlement data, wherein the knowledge graph is generated by determining multiple objects to be detected, multiple target medical insurance cards, and multidimensional data as multiple entities, and connecting multiple entities with associated relationships; scoring each object to be detected according to the knowledge graph to obtain a score value for each object to be detected, and determining multiple objects to be detected corresponding to multiple score values ​​in a preset score range as a group to be detected, wherein the group to be detected is a group suspected of medical insurance fraud; determining target physicians and target medical institutions that have committed medical insurance fraud according to the group to be detected and the knowledge graph.

[0069] An embodiment of the present application also provides an electronic device, including a memory and a processor, wherein a computer program is stored in the memory, and the processor is configured to execute the above method of detecting an object that has committed medical insurance fraud through the computer program.

[0070] The processor in the above-mentioned electronic device is used to run a program that performs the following functions: obtaining medical insurance settlement data generated within a preset period, wherein the medical insurance settlement data at least includes: a window identifier, a medical insurance card that generates consumption behavior, wherein the window identifier is used to indicate the window where the consumption behavior is generated; obtaining a first type of facial image associated with the window identifier, and obtaining a second type of facial image associated with the medical insurance card, wherein the first type of facial image is a facial image collected by a monitoring device of the consumption window within a preset period, and the second type of facial image is a facial image of an object bound to the medical insurance card; determining a first type of facial image different from the second type of facial image as an object to be detected, and obtaining target medical insurance settlement data associated with the object to be detected, wherein the target medical insurance settlement The calculation data at least includes: a target medical insurance card associated with the consumption behavior of the object to be detected, and multidimensional data associated with the target medical insurance card; a knowledge graph is generated according to the target medical insurance settlement data, wherein the knowledge graph is generated by determining multiple objects to be detected, multiple target medical insurance cards, and multidimensional data as multiple entities, and connecting multiple entities with associated relationships; scoring each object to be detected according to the knowledge graph to obtain a score value for each object to be detected, and determining multiple objects to be detected corresponding to multiple score values ​​in a preset score range as a group to be detected, wherein the group to be detected is a group suspected of medical insurance fraud; determining target physicians and target medical institutions that have committed medical insurance fraud according to the group to be detected and the knowledge graph.

[0071] An embodiment of the present application also provides a computer program product, including computer instructions, which, when executed by a processor, implement the above method for detecting objects that have committed medical insurance fraud.

[0072] It should be noted that the various modules in the above-mentioned device for detecting objects that have committed medical insurance fraud can be program modules (for example, a set of program instructions that implement a certain specific function) or hardware modules. For the latter, it can be expressed in the following forms, but is not limited to this: the expression form of each of the above-mentioned modules is a processor, or the functions of each of the above-mentioned modules are implemented by a processor.

[0073] The serial numbers of the above-mentioned embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0074] In the above embodiments of the present application, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.

[0075] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only schematic. For example, the division of the units can be a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0076] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0077] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0078] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the relevant technology or all or part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, disk or optical disk, etc. Various media that can store program codes.

[0079] The above is only a preferred implementation of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A method for detecting an object that has committed medical insurance fraud, characterized in that: include: Acquire the medical insurance settlement data generated within a preset period, wherein the medical insurance settlement data at least includes: a window identifier and a medical insurance card that generates the consumption behavior, wherein the window identifier is used to indicate the window where the consumption behavior is generated; Acquire a first type of facial image associated with the window identifier, and acquire a second type of facial image associated with the medical insurance card, wherein the first type of facial image is a facial image collected by the monitoring device of the consumption window within the preset period, and the second type of facial image is a facial image of the object bound to the medical insurance card; Determine the first type of face image different from the second type of face image as an object to be detected, and obtain target medical insurance settlement data associated with the object to be detected, wherein the target medical insurance settlement data at least includes: a target medical insurance card associated with the consumption behavior of the object to be detected, and multidimensional data associated with the target medical insurance card; Generate a knowledge graph according to the target medical insurance settlement data, wherein the knowledge graph is generated by determining the plurality of objects to be detected, the plurality of target medical insurance cards, and the multidimensional data as a plurality of entities, and connecting the plurality of entities having an associated relationship; Scoring each of the objects to be detected according to the knowledge graph to obtain a score value for each of the objects to be detected, and determining a plurality of the objects to be detected corresponding to a plurality of the score values ​​in a preset score range as a group to be detected, wherein the group to be detected is a group suspected of medical insurance fraud; Target physicians and target medical institutions that have committed medical insurance fraud are determined based on the group to be detected and the knowledge graph.

2. The method according to claim 1, characterized in that The multi-dimensional data includes: each insured person bound to the target medical insurance card, and each electronic prescription text data associated with the target medical insurance card, wherein the electronic prescription text data includes: drug name, medical institution, and physician; the association relationship includes at least: drug dispensing relationship, doctor-patient relationship, drug purchasing relationship, employment relationship, and prescription relationship; Connecting a plurality of entities having an associated relationship comprises: In the case where the target medical insurance card associated with the insured person and the target medical insurance card corresponding to the subject to be detected are the same medical insurance card, determining that there exists the proxy dispensing relationship between the entity representing the insured person and the entity representing the subject to be detected; In the case where the target medical insurance card corresponding to the insured person and the target medical insurance card associated with the physician are the same medical insurance card, determining that the doctor-patient relationship exists between the entity representing the insured person and the entity representing the physician; In the case where the target medical insurance card associated with the consumption behavior of the subject to be detected and the target medical insurance card associated with the electronic prescription text data are the same medical insurance card, determining that the drug purchase relationship exists between the entity representing the subject to be detected and the entity representing the electronic prescription text data; In the case where the electronic prescription text data associated with the physician is the same as the electronic prescription text data associated with the medical institution, determining that the employment relationship exists between the entity representing the physician and the entity representing the medical institution; In the case where the drug name is identical to any type of drug name included in the electronic prescription text data, it is determined that the entity representing the drug name and the entity representing the electronic prescription text data have the prescription relationship.

3. The method according to claim 2, characterized in that Scoring each of the objects to be detected according to the knowledge graph includes: Determine a scoring element according to the knowledge graph, wherein the scoring element includes: first-category attribute data of each entity and second-category attribute data of each association relationship, wherein the first-category attribute data is used to describe data related to drug purchase behavior of each entity, and the second-category attribute data is used to describe data related to drug purchase behavior jointly generated by a plurality of entities having an association relationship; The scoring of each of the objects to be detected is determined according to the functional relationship between the plurality of scoring elements.

4. The method according to claim 1, characterized in that Determining target physicians and target medical institutions that have committed medical insurance fraud according to the to-be-detected group and the knowledge graph includes: Generate a fully connected graph according to the knowledge graph, wherein the nodes in the fully connected graph are all target entities representing the insured person bound to the target medical insurance card, and there is a connection relationship between any two nodes in the fully connected graph; Determine a target object in the to-be-detected group according to the fully connected graph, wherein the target object is an to-be-detected object that has committed medical insurance fraud; The target physician and the target medical institution are determined according to the target object and the knowledge graph.

5. The method according to claim 4, characterized in that Generating a fully connected graph according to the knowledge graph includes: Determine the target entity in the knowledge graph, and determine multiple consumption moments corresponding to each of the target medical insurance cards; Connecting a plurality of the target entities corresponding to a plurality of the consumption moments whose differences are less than a preset value to generate an association graph; A target adjacent node is determined among multiple adjacent nodes of each of the target entities, and multiple target adjacent nodes are connected to generate the fully connected graph, wherein the target adjacent node is an adjacent node that has a connection relationship with each node in the association graph.

6. The method according to claim 4, characterized in that Determining a target object in the to-be-detected population according to the fully connected graph includes: Acquire the identity information of each of the insured persons, and determine the to-be-filtered objects among the target objects according to the identity information, wherein the to-be-filtered objects are the target objects that have a target social relationship with the insured persons, and the target social relationship includes at least: blood relationship; For other objects to be detected in the group to be detected except the object to be filtered, determining an embedding vector of each of the other objects to be detected according to the knowledge graph and the preset path, wherein the embedding vector is used to indicate an association relationship between the other objects to be detected and each entity in the preset path in the knowledge graph, and the preset path includes a plurality of the entities and a plurality of the association relationships; The target object is obtained by clustering the other objects to be detected according to the distances of the multiple embedded vectors, wherein the number of the target objects is multiple.

7. The method according to claim 4, characterized in that Determining the target physician and the target medical institution according to the target object and the knowledge graph includes: Determine in the knowledge graph a medical insurance card to be tested that is associated with the target object, and determine in the knowledge graph a physician to be tested that is associated with the medical insurance card to be tested; Determine the first page ranking PageRank score of each of the doctors to be tested according to the number of the doctors to be tested and the association relationship between the multiple doctors to be tested recorded in the knowledge graph; Determine the physician to be detected corresponding to the first PageRank score greater than a first preset score value as the target physician; The target medical institution is determined according to the target physician and the knowledge graph.

8. The method according to claim 4, characterized in that Determining the target medical institution according to the target physician and the knowledge graph includes: Determining in the knowledge graph the medical institution to be detected that has an association relationship with the target physician; Determine a second PageRank score for each of the medical institutions to be detected according to the number of the medical institutions to be detected and the association relationship between the plurality of medical institutions to be detected recorded in the knowledge graph; The medical institution to be detected corresponding to the second PageRank score that is greater than a second preset score value is determined as the target medical institution.

9. The method according to claim 1, characterized in that: After determining the target physician and the target medical institution, the method further includes: A target chain is determined in the knowledge graph, wherein the target chain is an edge in the knowledge graph that connects in series the entity corresponding to the object to be detected that has committed medical insurance fraud, the entity representing the target physician, the entity representing the target medical institution, and the entity representing the target insured person, and the target insured person is an insured person associated with the object to be detected that has committed medical insurance fraud in the knowledge graph.

10. A device for detecting an object that has committed medical insurance fraud, characterized in that: include: A first acquisition module is used to acquire medical insurance settlement data generated within a preset period, wherein the medical insurance settlement data at least includes: a window identifier and a medical insurance card that generates the consumption behavior, wherein the window identifier is used to indicate the window that generates the consumption behavior; A second acquisition module is used to acquire a first type of facial image associated with the window identifier, and acquire a second type of facial image associated with the medical insurance card, wherein the first type of facial image is a facial image collected by the monitoring device of the consumption window within the preset period, and the second type of facial image is a facial image of a binding object of the medical insurance card; A first determination module is used to determine the first type of face image different from the second type of face image as an object to be detected, and obtain target medical insurance settlement data associated with the object to be detected, wherein the target medical insurance settlement data at least includes: a target medical insurance card associated with the consumption behavior of the object to be detected, and multidimensional data associated with the target medical insurance card; A generation module, used to generate a knowledge graph according to the target medical insurance settlement data, wherein the knowledge graph is generated by determining a plurality of the objects to be detected, a plurality of the target medical insurance cards, and the multidimensional data as a plurality of entities, and connecting the plurality of entities having an associated relationship; A second determination module is used to score each of the objects to be detected according to the knowledge graph to obtain a score value of each of the objects to be detected, and determine a plurality of the objects to be detected corresponding to a plurality of the score values ​​in a preset score range as a group to be detected, wherein the group to be detected is a group suspected of medical insurance fraud; The third determination module is used to determine the target physicians and target medical institutions that have committed medical insurance fraud based on the group to be detected and the knowledge graph.

Images