CN113270200A

CN113270200A - Abnormal patient identification method based on artificial intelligence and related equipment

Info

Publication number: CN113270200A
Application number: CN202110567586.XA
Authority: CN
Inventors: 唐蕊; 蒋雪涵; 孙行智
Original assignee: Ping An Technology Shenzhen Co Ltd
Current assignee: Ping An Technology Shenzhen Co Ltd
Priority date: 2021-05-24
Filing date: 2021-05-24
Publication date: 2021-08-17
Anticipated expiration: 2041-05-24
Also published as: CN113270200B

Abstract

The application is applied to the field of digital medical treatment, relates to the technical field of artificial intelligence, and discloses an abnormal patient identification method based on artificial intelligence and related equipment, wherein the method comprises the following steps: inputting each patient embedding vector set into a target patient behavior learning model respectively to obtain patient behavior vectors corresponding to the patient embedding vector sets respectively, wherein the target patient behavior learning model comprises: an input layer, a 12-layer encoder, an output layer; respectively carrying out vector extraction of a single charging item on each patient behavior vector according to the position of the identifier to obtain behavior vectors to be clustered, which correspond to a plurality of patient embedded vector sets respectively; clustering all behavior vectors to be clustered by adopting a DBSCAN clustering algorithm to obtain a plurality of patient behavior vector clustering sets, detecting abnormal patients according to the plurality of patient behavior vector clustering sets, and determining the abnormal patient sets. The behavior of the patient is deeply excavated, and the accuracy of identifying the abnormal patient is improved.

Description

Abnormal patient identification method based on artificial intelligence and related equipment

Technical Field

The present application relates to the field of artificial intelligence technology, and in particular, to an abnormal patient identification method based on artificial intelligence and related equipment.

Background

Medical quality control refers to the identification, analysis, assessment and treatment of existing and potential risks during medical activities, planned and organized reduction and elimination of risks, and reduction of adverse effects and economic losses due to risk events.

In the field of medical quality control, identification of abnormal patients is one of the important tasks. The existing abnormal patient identification method mainly analyzes basic information (such as age, sex, height, weight and the like) of a patient and performs abnormal identification on the patient from statistical values, however, the method is more basic, the capacity of performing abnormal patient identification is limited, and the accuracy of abnormal patient identification is reduced.

Disclosure of Invention

The application mainly aims to provide an abnormal patient identification method, an abnormal patient identification device, abnormal patient identification equipment and a storage medium based on artificial intelligence, and aims to solve the technical problems that in the prior art, the abnormal patient identification method mainly analyzes basic information of a patient, the abnormal patient identification capability is limited from a statistic value, and the accuracy of abnormal patient identification is reduced.

In order to achieve the above object, the present application proposes an abnormal patient identification method based on artificial intelligence, the method comprising:

obtaining a plurality of sets of patient embedding vectors, the sets of patient embedding vectors comprising: a toll item name embedding vector, a toll item type embedding vector and a toll item charge embedding vector corresponding to a plurality of toll items of a patient;

inputting each patient embedding vector set into a target patient behavior learning model for patient behavior learning to obtain patient behavior vectors corresponding to the plurality of patient embedding vector sets, wherein the target patient behavior learning model comprises: an input layer, a 12-layer encoder, an output layer;

acquiring an identifier position, and respectively performing vector extraction on each patient behavior vector according to the identifier position to obtain behavior vectors to be clustered, which correspond to the patient embedding vector sets respectively;

clustering all the behavior vectors to be clustered by adopting a DBSCAN clustering algorithm to obtain a plurality of patient behavior vector clustering sets, and detecting abnormal patients according to the plurality of patient behavior vector clustering sets to obtain abnormal patient behavior vector sets;

and determining abnormal patients according to the abnormal patient behavior vector set to obtain an abnormal patient set.

Further, before the step of inputting each patient embedded vector set into the target patient behavior learning model for patient behavior learning to obtain the patient behavior vectors corresponding to the patient embedded vector sets, the method further includes:

obtaining a plurality of patient training samples, the patient training samples comprising: patient embedding vector sample data, a patient behavior calibration vector and a total cost classification calibration vector;

extracting one patient training sample from the plurality of patient training samples as a target patient training sample;

determining the hidden charging items according to the training samples of the target patients by adopting a random algorithm and a preset proportion to obtain a hidden charging item set;

embedding vector hiding corresponding to each charging item in the hidden charging item set aiming at the patient embedded vector sample data of the target patient training sample to obtain hidden patient embedded vector sample data;

according to the patient behavior calibration vector of the target patient training sample, extracting a calibration vector corresponding to each charging item in the hidden charging item set to obtain a target calibration vector;

adopting the embedded vectors of all the non-hidden toll projects to carry out the patient behavior learning of the hidden toll projects, and carrying out the patient behavior learning according to the hidden patient embedded vector sample data and the patient behavior learning model to be trained to obtain a first training vector;

adopting sample embedded vectors of all the toll items with the same type as the concealed toll items to carry out patient behavior learning of the concealed toll items, carrying out patient behavior learning according to the concealed patient embedded vector sample data and the patient behavior learning model to be trained, and determining a second training vector;

performing total cost classification prediction according to the identifier position, the hidden patient embedded vector sample data, the patient behavior learning model to be trained, the full connection layer and the softmax layer to obtain a total cost classification probability training vector;

calculating the target calibration vector, the first training vector, the second training vector, the total cost classification probability training vector and the total cost classification calibration vector of the target patient training sample by inputting a loss function to obtain a target loss value, updating parameters of the patient behavior learning model to be trained, the full connection layer and the softmax layer according to the target loss value, using the updated patient behavior learning model to be trained for calculating the first training vector and the second training vector next time, and using the updated patient behavior learning model to be trained, the full connection layer and the softmax layer for calculating the total cost classification probability training vector next time;

and repeatedly executing the step of extracting one patient training sample from the plurality of patient training samples as a target patient training sample until the target loss value reaches a first convergence condition or the iteration number reaches a second convergence condition, and determining the patient behavior learning model to be trained, of which the target loss value reaches the first convergence condition or the iteration number reaches the second convergence condition, as the target patient behavior learning model.

Further, the method for learning the patient behavior of the toll collection project by using the non-hidden embedded vectors of all the toll collection projects comprises the steps of learning the patient behavior according to the hidden patient embedded vector sample data and the patient behavior learning model to be trained to obtain a first training vector, learning the patient behavior of the toll collection project by using the sample embedded vectors of all the toll collection projects with the same type as the hidden toll collection projects, learning the patient behavior according to the hidden patient embedded vector sample data and the patient behavior learning model to be trained, determining a second training vector, and performing total cost classification prediction according to the identifier position, the hidden patient embedded vector sample data, the patient behavior learning model to be trained, the full connection layer and the softmax layer, the step of obtaining the total cost classification probability training vector comprises the following steps:

inputting the hidden patient embedded vector sample data into the patient behavior learning model to be trained for patient behavior learning to obtain a patient behavior training vector to be analyzed;

extracting a training vector corresponding to each charging item in the hidden charging item set from the patient behavior training vectors to be analyzed to obtain a first training vector;

extracting one charging item from the hidden charging item set as a charging item to be learned;

embedding vector hiding of all the toll items with different types from the toll items to be learned is carried out on the hidden patient embedded vector sample data, and patient embedded vector sample data to be analyzed is obtained;

inputting the patient embedding vector sample data to be analyzed into the patient behavior learning model to be trained for patient behavior learning to obtain a similar prediction patient behavior training vector;

extracting a training vector corresponding to the toll collection project to be learned from the similar prediction patient behavior training vectors to obtain a training vector to be calculated corresponding to the toll collection project to be learned;

repeatedly executing the step of extracting one charging item from the hidden charging item set as a charging item to be learned until the training vector to be calculated corresponding to all the charging items in the hidden charging item set is determined;

taking all the training vectors to be calculated as the second training vectors;

extracting training vectors corresponding to the positions of the identification symbols from the patient behavior training vectors to be analyzed to obtain identification symbol training vectors;

and sequentially inputting the identification symbol training vector into the full-connection layer and the softmax layer to carry out total cost classification prediction to obtain the total cost classification probability training vector.

Further, the step of calculating the target calibration vector, the first training vector, the second training vector, the total cost classification probability training vector, and the total cost classification calibration vector of the target patient training sample by inputting the loss function to obtain a target loss value includes:

inputting the target calibration vector and the first training vector into a first cross entropy loss function for calculation to obtain a first loss value;

inputting the target calibration vector and the second training vector into a second cross entropy loss function for calculation to obtain a second loss value;

inputting the total cost classification probability training vector and the total cost classification calibration vector of the target patient training sample into a third cross entropy loss function for calculation to obtain a third loss value;

and adding and calculating the first loss value, the second loss value and the third loss value to obtain the target loss value.

Further, the step of inputting each patient embedded vector set into a target patient behavior learning model for patient behavior learning to obtain patient behavior vectors corresponding to the plurality of patient embedded vector sets includes:

acquiring a patient embedding vector set from the plurality of patient embedding vector sets to obtain a target patient embedding vector set;

inputting the set of target patient embedding vectors into the input layer of the target patient behavior learning model;

performing vector addition calculation on the charging item name embedding vector, the charging item type embedding vector and the charging item cost embedding vector of the target patient embedding vector set by adopting the input layer of the target patient behavior learning model to obtain a vector to be spliced;

sequentially splicing the identification symbols and the vectors to be spliced by adopting the input layer of the target patient behavior learning model to obtain target vectors to be analyzed;

coding and learning the target vector to be analyzed by adopting the 12-layer coder of the target patient behavior learning model to obtain a target coding vector;

vector output is carried out on the target coding vector by adopting the output layer of the target patient behavior learning model, and the patient behavior vector corresponding to the target patient embedding vector set is obtained;

and repeating the step of obtaining a patient embedding vector set from the plurality of patient embedding vector sets to obtain a target patient embedding vector set until the patient behavior vectors corresponding to the plurality of patient embedding vector sets are determined.

Further, the step of clustering all the behavior vectors to be clustered by using a DBSCAN clustering algorithm to obtain a plurality of patient behavior vector cluster sets, and performing abnormal patient detection according to the plurality of patient behavior vector cluster sets to obtain abnormal patient behavior vector sets includes:

acquiring preset clustering parameter radius and a preset minimum vector number required for forming a high-density area;

based on the preset minimum vector number and the Euclidean distance algorithm, clustering all the behavior vectors to be clustered by adopting the DBSCAN clustering algorithm to obtain a plurality of patient behavior vector clustering sets;

respectively establishing a clustering contour for each patient behavior vector clustering set by adopting the preset clustering parameter radius to obtain target clustering contours corresponding to the patient behavior vector clustering sets;

and based on the target cluster contour corresponding to each of the plurality of patient behavior vector cluster sets, performing abnormal patient detection on the plurality of patient behavior vector cluster sets to obtain the abnormal patient behavior vector set.

Further, the step of performing abnormal patient detection on the plurality of patient behavior vector cluster sets based on the target cluster contours corresponding to the plurality of patient behavior vector cluster sets, to obtain the abnormal patient behavior vector set, includes:

extracting one behavior vector to be clustered from the plurality of patient behavior vector clustering sets to serve as an object to be detected;

extracting one target clustering contour from the target clustering contours corresponding to the plurality of patient behavior vector clustering sets respectively to serve as a clustering contour to be detected;

judging whether the density of all core points of the patient behavior vector clustering set corresponding to the object to be detected and the cluster contour to be detected is reachable or not, and obtaining a density reachable result corresponding to the cluster contour to be detected;

repeatedly executing the step of extracting one target cluster contour from the target cluster contours corresponding to the plurality of patient behavior vector cluster sets respectively, and taking the target cluster contour as a cluster contour to be detected until the density reachable result corresponding to each target cluster contour is determined;

when all the density reachable results are density unreachable, determining the behavior vector to be clustered corresponding to the object to be detected as an abnormal patient behavior vector;

repeatedly executing the step of extracting one behavior vector to be clustered from the plurality of patient behavior vector clustering sets as an object to be detected until the extraction of all the behavior vectors to be clustered in the plurality of patient behavior vector clustering sets is completed;

and taking all the abnormal patient behavior vectors as the abnormal patient behavior vector set.

The present application further proposes an abnormal patient identification device based on artificial intelligence, the device comprising:

an embedded vector acquisition module to acquire a plurality of patient embedded vector sets, the patient embedded vector sets comprising: a toll item name embedding vector, a toll item type embedding vector and a toll item charge embedding vector corresponding to a plurality of toll items of a patient;

a patient behavior learning module, configured to input each patient embedded vector set into a target patient behavior learning model for patient behavior learning, so as to obtain patient behavior vectors corresponding to the patient embedded vector sets, where the target patient behavior learning model includes: an input layer, a 12-layer encoder, an output layer;

the vector extraction module is used for acquiring the positions of the identification symbols, and respectively carrying out vector extraction on each patient behavior vector according to the positions of the identification symbols to obtain behavior vectors to be clustered, which correspond to the patient embedding vector sets respectively;

the abnormal patient detection module is used for clustering all the behavior vectors to be clustered by adopting a DBSCAN clustering algorithm to obtain a plurality of patient behavior vector clustering sets, and performing abnormal patient detection according to the plurality of patient behavior vector clustering sets to obtain abnormal patient behavior vector sets;

and the abnormal patient set determining module is used for determining abnormal patients according to the abnormal patient behavior vector set to obtain an abnormal patient set.

The present application further proposes a computer device comprising a memory and a processor, the memory storing a computer program, the processor implementing the steps of any of the above methods when executing the computer program.

The present application also proposes a computer-readable storage medium having stored thereon a computer program which, when being executed by a processor, carries out the steps of the method of any of the above.

The abnormal patient identification method, the abnormal patient identification device, the abnormal patient identification equipment and the abnormal patient identification storage medium based on the artificial intelligence carry out patient behavior learning by inputting each patient embedding vector set into a target patient behavior learning model respectively to obtain patient behavior vectors corresponding to a plurality of patient embedding vector sets, and the patient embedding vector sets comprise: the method comprises the steps of embedding charge item name vectors, charge item type embedded vectors and charge item cost embedded vectors corresponding to a plurality of charge items of a patient, extracting vectors of single charge items of each patient behavior vector according to identification symbol positions to obtain behavior vectors to be clustered of a plurality of patient embedded vector sets, clustering all the behavior vectors to be clustered by adopting a DBSCAN clustering algorithm to obtain a plurality of patient behavior vector cluster sets, carrying out abnormal patient detection according to the plurality of patient behavior vector cluster sets to obtain abnormal patient behavior vector sets, carrying out abnormal patient determination according to the abnormal patient behavior vector sets to obtain abnormal patient sets, carrying out abnormal identification on basic information of the patients relatively, embedding the vectors, charge item type embedded vectors and charge item cost embedded vectors according to the charge item names, charge item types, and charge item cost, The charge project cost embedding vector and the DBSCAN clustering algorithm are used for identifying abnormal patients, so that deep mining of the behaviors of the patients is realized, and the accuracy of identifying the abnormal patients is improved.

Drawings

FIG. 1 is a schematic flow chart of an artificial intelligence based abnormal patient identification method according to an embodiment of the present application;

FIG. 2 is a block diagram of an artificial intelligence based abnormal patient identification device according to an embodiment of the present application;

fig. 3 is a block diagram illustrating a structure of a computer device according to an embodiment of the present application.

The objectives, features, and advantages of the present application will be further described with reference to the accompanying drawings.

Detailed Description

In order to make the objects, technical solutions and advantages of the present application more apparent, the present application is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the present application and are not intended to limit the present application.

In order to solve the technical problems that in the prior art, an abnormal patient identification method mainly analyzes basic information of a patient, the abnormal patient identification capability is limited from a statistic value, and the accuracy of abnormal patient identification is reduced, the abnormal patient identification method based on artificial intelligence is provided. The abnormal patient identification method based on artificial intelligence extracts a name embedding vector, a type embedding vector and a cost embedding vector from all charging items of each patient, then model patient behavior learning is obtained by training based on a user-defined input layer, a 12-layer encoder of Bert and an output layer of Bert, and then extracting a vector corresponding to the position of the identification symbol from the patient behavior learning result, clustering and abnormal patient detection are carried out on the extracted vector by adopting a DBSCAN clustering algorithm to obtain a behavior vector corresponding to an abnormal patient, the basic information of the patient is relatively analyzed to carry out abnormal identification, and the abnormal patient is identified based on the name embedded vector, the type embedded vector, the cost embedded vector and the DBSCAN clustering algorithm of the charging project, so that deep mining on the behavior of the patient is realized, and the accuracy of identifying the abnormal patient is improved.

Referring to fig. 1, an embodiment of the present application provides an artificial intelligence based abnormal patient identification method, including:

s1: obtaining a plurality of sets of patient embedding vectors, the sets of patient embedding vectors comprising: a toll item name embedding vector, a toll item type embedding vector and a toll item charge embedding vector corresponding to a plurality of toll items of a patient;

s2: inputting each patient embedding vector set into a target patient behavior learning model for patient behavior learning to obtain patient behavior vectors corresponding to the plurality of patient embedding vector sets, wherein the target patient behavior learning model comprises: an input layer, a 12-layer encoder, an output layer;

s3: acquiring an identifier position, and respectively performing vector extraction on each patient behavior vector according to the identifier position to obtain behavior vectors to be clustered, which correspond to the patient embedding vector sets respectively;

s4: clustering all the behavior vectors to be clustered by adopting a DBSCAN clustering algorithm to obtain a plurality of patient behavior vector clustering sets, and detecting abnormal patients according to the plurality of patient behavior vector clustering sets to obtain abnormal patient behavior vector sets;

s5: and determining abnormal patients according to the abnormal patient behavior vector set to obtain an abnormal patient set.

In this embodiment, each patient embedding vector set is input into the target patient behavior learning model to perform patient behavior learning, so as to obtain patient behavior vectors corresponding to the plurality of patient embedding vector sets, where the patient embedding vector set includes: the method comprises the steps of embedding charge item name vectors, charge item type embedded vectors and charge item cost embedded vectors corresponding to a plurality of charge items of a patient, extracting vectors of single charge items of each patient behavior vector according to identification symbol positions to obtain behavior vectors to be clustered of a plurality of patient embedded vector sets, clustering all the behavior vectors to be clustered by adopting a DBSCAN clustering algorithm to obtain a plurality of patient behavior vector cluster sets, carrying out abnormal patient detection according to the plurality of patient behavior vector cluster sets to obtain abnormal patient behavior vector sets, carrying out abnormal patient determination according to the abnormal patient behavior vector sets to obtain abnormal patient sets, carrying out abnormal identification on basic information of the patients relatively, embedding the vectors, charge item type embedded vectors and charge item cost embedded vectors according to the charge item names, charge item types, and charge item cost, The charge project cost embedding vector and the DBSCAN clustering algorithm are used for identifying abnormal patients, so that deep mining of the behaviors of the patients is realized, and the accuracy of identifying the abnormal patients is improved.

For S1, a plurality of sets of patient embedding vectors may be obtained from the database, or a plurality of sets of patient embedding vectors input by the user, or a plurality of sets of patient embedding vectors sent by the third-party application system.

The patient embedding vector set is an embedding vector extracted from the charging items obtained by one or more visits of a patient. The set of patient embedded vectors includes a toll project name embedded vector, a toll project type embedded vector, and a toll project cost embedded vector.

The charging items include: name of charge item, type of charge item, cost. The charging item names include, but are not limited to: x-ray examination, X-ray fluoroscopy examination, general fluoroscopy, X-ray photography, digital photography, B-ultrasonic routine examination, single-organ B-ultrasonic examination, general color Doppler ultrasonic examination, and color Doppler ultrasonic routine examination.

The same patient identifier is corresponding to the same toll project identifier of the same patient in the same position of the toll project name embedding vector, the toll project type embedding vector and the toll project cost embedding vector of the same patient embedding vector set. For example, the toll project name embedded vector a11, the toll project type embedded vector a12, and the toll project cost embedded vector a13 in the patient embedded vector set a, the 3 rd vector element in the toll project name embedded vector a11, the 3 rd vector element in the toll project type embedded vector a12, and the 3 rd vector element in the toll project cost embedded vector a13 all correspond to the toll project identifier B11, which is not limited in this example.

The charging item identifier refers to an item ID, such as B11, B12, B13, B14, and B15, which is not limited by the examples herein.

The charging item name embedding vector is an embedding vector composed of coding vectors of charging item names corresponding to a plurality of charging item identifications respectively. The coding vector for the toll project name may be looked up using the toll project name in a toll project name and coding vector dictionary. The toll item name and code vector dictionary comprises: the name of the charging project and the coding vector correspond to each other one by one.

The charging item type embedded vector is an embedded vector formed by a plurality of coding vectors of charging item types respectively corresponding to the charging item identifications. The coding vector for the toll item type can be looked up in a toll item type and coding vector dictionary using the toll item type. The toll item type and encoding vector dictionary comprises: the charging item type and the coding vector correspond to each other one by one.

The charge item charge embedding vector is an embedding vector composed of a plurality of coding vectors for identifying charge item charges corresponding to the charge items. The coding vector for the toll project fee may be looked up in the toll project fee segmentation and coding vector dictionary using the toll project fee. For example, the cost is divided into 100 intervals, and each interval corresponds to one code vector, which is not limited in this example. The toll project cost segmentation and coding vector dictionary comprises: the cost interval corresponds to the coding vector one by one. The cost interval includes: starting point cost, end point cost.

For example, the charging items of the patient M are identified by B11, B12, B13, B14, and B15, wherein B11 (X-ray examination for charging item name, L1 for charging item type, and F1 for cost), B12 (B-super-routine examination for charging item name, L2 for charging item type, and F2 for cost), B13 (L-single-organ B-super-examination for charging item name, L2 for charging item type, and F3 for cost), B14 (color doppler ultrasound routine examination for charging item name, L-3 for charging item type, and F4 for charging item name), B15 (digital photography for charging item name, L-4 for charging item type, and F5 for cost), the charging item name embedding vectors of the patient embedding vector set of the patient M are obtained by encoding the charging vector obtained by coding the charging vector found in the dictionary 11 for charging vector found by the B11, The embedded vectors obtained by sequentially splicing B12, B13, B14, and B15, the toll project type embedded vector of the patient embedded vector set of the patient M is an embedded vector obtained by sequentially finding the coding vector in the toll project type and coding vector dictionary by L1, L2, L2, L3, and L4, and splicing the found coding vector in the sequence of B11, B12, B13, B14, and B15, and the toll project cost embedded vector of the patient embedded vector set of the patient M is an embedded vector obtained by sequentially finding the coding vector in the toll project cost segmentation and coding vector dictionary by F1, F2, F3, F4, and F5, and splicing the found coding vector in the sequence of B11, B12, B13, B14, and B15, which is not specifically limited herein.

For S2, inputting each of the patient-embedded vector sets into an input layer of the target patient behavior learning model for patient behavior learning, and outputting a patient behavior vector for each of the patient-embedded vector sets into an output layer of the target patient behavior learning model of the input layer.

The input layer includes: the device comprises an embedded vector addition module and a vector splicing module. And the embedded vector addition module is used for carrying out vector addition on the charging item name embedded vector, the charging item type embedded vector and the charging item cost embedded vector of the same patient embedded vector set to obtain a one-dimensional vector, and taking the obtained one-dimensional vector as a vector to be spliced. And the vector splicing module is used for carrying out vector splicing according to the identification symbol and the vector to be spliced and inputting the spliced vector into the 12-layer encoder.

The 12-layer Encoder is a 12-layer Encoder using a bert (bidirectional Encoder reproduction from transforms) model.

And the output layer adopts the output layer of the Bert model and is used for outputting the coding result of the 12-layer coder.

The patient behavior vector sequentially comprises an identification symbol and a learned vector representation corresponding to each charging item identification. For example, the charging items of the patient M are identified by B11, B12, B13, B14, and B15, the patient-embedded vector set corresponding to the patient M is subjected to patient behavior learning through the input layer of the target patient behavior learning model, and the obtained patient behavior vector is [ O0O 1O 2O 3O 4O 5], that is, O0 is a learned vector representation corresponding to the identifier, O1 is a learned vector representation corresponding to B11, O2 is a learned vector representation corresponding to B12, O3 is a learned vector representation corresponding to B13, O4 is a learned vector representation corresponding to B14, and O5 is a learned vector representation corresponding to B15, which is not specifically limited in this example.

For S3, the identifier position may be obtained from the database, or may be an identifier position sent by a third-party application system, or may be written in a program file implementing the present application; and respectively carrying out vector extraction on each patient behavior vector according to the position of the identifier, and taking each extracted vector as a behavior vector to be clustered. For example, the patient behavior vector is [ O0O 1O 2O 3O 4O 5], and according to the identifier position, vector extraction is performed on each patient behavior vector, and the obtained behavior vector to be clustered is [ O0], which is not specifically limited in this example. Obtaining the position of the identifier, and respectively carrying out vector extraction on each patient behavior vector according to the position of the identifier to obtain behavior vectors to be clustered, which correspond to the patient embedding vector sets respectively

The marker position refers to the position data of the first marker in the vector of the input layer input 12-layer encoder.

The identifier position can be obtained from a database, can also be the identifier position input by a user, can also be the identifier position sent by a third-party application system, and can also be written into a program file for realizing the application.

For S4, it can be understood that, by using a DBSCAN clustering algorithm, all the behavior vectors to be clustered are clustered to obtain a plurality of patient behavior vector cluster sets, and abnormal patient detection is performed according to the plurality of patient behavior vector cluster sets, and a specific implementation manner of obtaining an abnormal patient behavior vector set may also be selected from the prior art, which is not specifically limited herein.

Optionally, a DBSCAN clustering algorithm (a noise-based density clustering method) and a preset minimum vector number are adopted, all the behavior vectors to be clustered are automatically clustered into a plurality of clusters, and each obtained cluster is used as a patient behavior vector clustering set; then presetting clustering parameter radiuses to respectively find out a clustering contour of each patient behavior vector clustering set; and carrying out abnormal patient detection on each behavior vector to be clustered according to the clustering contour to obtain an abnormal patient behavior vector set.

For S5: and taking the patient corresponding to each behavior vector to be clustered in the abnormal patient behavior vector set as an abnormal patient, and taking all abnormal patients as an abnormal patient set.

In an embodiment, before the step of inputting each of the patient-embedded vector sets into a target patient behavior learning model for patient behavior learning to obtain the patient behavior vectors corresponding to the patient-embedded vector sets, the method further includes:

s021: obtaining a plurality of patient training samples, the patient training samples comprising: patient embedding vector sample data, a patient behavior calibration vector and a total cost classification calibration vector;

s022: extracting one patient training sample from the plurality of patient training samples as a target patient training sample;

s023: determining the hidden charging items according to the training samples of the target patients by adopting a random algorithm and a preset proportion to obtain a hidden charging item set;

s024: embedding vector hiding corresponding to each charging item in the hidden charging item set aiming at the patient embedded vector sample data of the target patient training sample to obtain hidden patient embedded vector sample data;

s025: according to the patient behavior calibration vector of the target patient training sample, extracting a calibration vector corresponding to each charging item in the hidden charging item set to obtain a target calibration vector;

s026: adopting the embedded vectors of all the non-hidden toll projects to carry out the patient behavior learning of the hidden toll projects, and carrying out the patient behavior learning according to the hidden patient embedded vector sample data and the patient behavior learning model to be trained to obtain a first training vector;

s027: adopting sample embedded vectors of all the toll items with the same type as the concealed toll items to carry out patient behavior learning of the concealed toll items, carrying out patient behavior learning according to the concealed patient embedded vector sample data and the patient behavior learning model to be trained, and determining a second training vector;

s028: performing total cost classification prediction according to the identifier position, the hidden patient embedded vector sample data, the patient behavior learning model to be trained, the full connection layer and the softmax layer to obtain a total cost classification probability training vector;

s029: calculating the target calibration vector, the first training vector, the second training vector, the total cost classification probability training vector and the total cost classification calibration vector of the target patient training sample by inputting a loss function to obtain a target loss value, updating parameters of the patient behavior learning model to be trained, the full connection layer and the softmax layer according to the target loss value, using the updated patient behavior learning model to be trained for calculating the first training vector and the second training vector next time, and using the updated patient behavior learning model to be trained, the full connection layer and the softmax layer for calculating the total cost classification probability training vector next time;

s0210: and repeatedly executing the step of extracting one patient training sample from the plurality of patient training samples as a target patient training sample until the target loss value reaches a first convergence condition or the iteration number reaches a second convergence condition, and determining the patient behavior learning model to be trained, of which the target loss value reaches the first convergence condition or the iteration number reaches the second convergence condition, as the target patient behavior learning model.

The embodiment realizes model training based on three training methods, provides a model for representing the patient behaviors by combining three embedded vectors with different meanings, and provides a basis for clustering and abnormal patient detection by subsequently adopting the patient behaviors learned from the toll collection project.

For S021, a plurality of patient training samples may be obtained from the database, or a plurality of patient training samples input by the user, or a plurality of patient training samples sent by the third-party application system.

Each patient training sample includes a patient embedded vector sample data, a patient behavior calibration vector, and a total cost classification calibration vector. The patient behavior calibration vector is a calibration result of patient behavior learning for patient embedding vector sample data. The total cost classification calibration vector is a total cost classification calibration result of embedding vector sample data into a patient. For example, if the total cost label is high, medium, or low, the total cost classification calibration vector may be any one of [ 100 ], [ 010 ], [ 001 ], where [ 100 ] indicates that the total cost is high, [ 010 ] indicates that the total cost is medium, and [ 001 ] indicates that the total cost is low, which is not limited in this example.

Patient embedding vector sample data includes: a toll project name sample embedding vector, a toll project type sample embedding vector, a toll project cost sample embedding vector. The toll item name sample embedded vector is an embedded vector of toll item names. The toll item type sample embedded vector is an embedded vector of toll item types. The charge item cost sample embedded vector is an embedded vector of charge item costs.

Each patient embedded vector sample data comprises a charging item name sample embedded vector, a charging item type sample embedded vector and a charging item cost sample embedded vector.

For S022, one patient training sample is sequentially extracted from the plurality of patient training samples, and the extracted patient training sample is used as a target patient training sample.

And S023, multiplying a preset proportion by the number of the charging item identifications corresponding to the target patient training sample to obtain the number of the hidden items. And then, determining the charging item identifications of the charging items to be hidden according to all the charging item identifications corresponding to the target patient training sample by adopting a random algorithm and the quantity of the hidden items, and taking the charging items corresponding to all the determined charging item identifications as a hidden charging item set.

For S024, in a charging item name sample embedding vector of the patient embedding vector sample data of the target patient training sample, replacing an embedding vector corresponding to each charging item in the hidden charging item set by using a mask character to obtain a hidden charging item name sample embedding vector; in a charging item type sample embedding vector of the patient embedding vector sample data of the target patient training sample, replacing an embedding vector corresponding to each charging item in the hidden charging item set by using a mask symbol to obtain a hidden charging item type sample embedding vector; in a charge item cost sample embedding vector of the patient embedding vector sample data of the target patient training sample, replacing an embedding vector corresponding to each charge item in the hidden charge item set by using a mask symbol to obtain a hidden charge item cost sample embedding vector; and embedding the hidden charge item name sample into a vector, the hidden charge item type sample into the vector and the hidden charge item cost sample into the vector to serve as the hidden patient embedded vector sample data.

Optionally, the Mask is set to Mask.

For S025, according to the patient behavior calibration vector of the target patient training sample, extracting the calibration vector corresponding to each charging item in the hidden charging item set, and taking all the extracted data as the target calibration vector.

And S026, performing patient behavior learning according to the hidden patient embedded vector sample data and the patient behavior learning model to be trained, then extracting a training vector corresponding to each charging item in the hidden charging item set from vectors learned by patient behavior, and taking all extracted data as first training vectors. That is, only the embedded vectors of all of the toll items that are not hidden in the hidden patient embedded vector sample data are data available for learning.

For S027, for each charging item in the hidden charging item set, finding out sample embedded vectors of the same type for all hidden charging items from the hidden patient embedded vector sample data, then performing patient behavior learning on the hidden charging item by using the found sample embedded vectors, obtaining homogeneous prediction patient behavior training vectors corresponding to the hidden charging items, and taking all homogeneous prediction patient behavior training vectors as second training vectors.

And S028, performing patient behavior learning according to the hidden patient embedded vector sample data and the patient behavior learning model to be trained, then extracting a vector corresponding to the identifier position from the vector learned by the patient behavior, performing total cost classification prediction on the extracted vector through a full connection layer and a softmax layer, and taking the prediction result as a total cost classification probability training vector.

Wherein, all vector elements in the total cost classification probability training vector are added to be 1. That is, each vector element in the total cost classification probability training vector is a number from 0 to 1, and may include 0 or 1.

For example, if the total cost label is high, medium, or low, the total cost classification probability training vector may be [ 0.80.150.05 ], that is, the probability of total cost being high is 0.8, the probability of total cost being medium is 0.15, and the probability of total cost being low is 0.05, which is not limited in this example.

And for S029, inputting the target calibration vector, the first training vector, the second training vector, the total cost classification probability training vector and the total cost classification calibration vector of the target patient training sample into a loss function for calculation, and taking the calculated result as a target loss value.

It can be understood that the method for updating the parameters of the patient behavior learning model to be trained, the full-link layer and the softmax layer according to the target loss value may be selected from the prior art, and details are not repeated herein.

For S0210, the steps S022 to S0210 are repeatedly performed until the target loss value reaches a first convergence condition or the number of iterations reaches a second convergence condition.

The first convergence condition means that the magnitudes of target loss values calculated two adjacent times satisfy a lipschitz condition (lipschitz continuity condition).

The number of iterations reaching the second convergence condition refers to the number of times that the patient behavior learning model to be trained is used for calculating the first training vector and the second training vector, that is, the number of iterations is increased by 1 after calculation.

In one embodiment, the above method for learning patient behavior of a toll project hidden by using an unhidden embedded vector of all toll projects performs patient behavior learning according to the hidden patient embedded vector sample data and a patient behavior learning model to be trained to obtain a first training vector, the method for learning patient behavior of a toll project hidden by using a hidden sample embedded vector of all toll projects of the same type as the hidden toll project performs patient behavior learning according to the hidden patient embedded vector sample data and the patient behavior learning model to be trained, determines a second training vector, and performs total cost classification prediction according to the identifier position, the hidden patient embedded vector sample data, the patient behavior learning model to be trained, a full connection layer and a softmax layer, the step of obtaining the total cost classification probability training vector comprises the following steps:

s0221: inputting the hidden patient embedded vector sample data into the patient behavior learning model to be trained for patient behavior learning to obtain a patient behavior training vector to be analyzed;

s0222: extracting a training vector corresponding to each charging item in the hidden charging item set from the patient behavior training vectors to be analyzed to obtain a first training vector;

s0223: extracting one charging item from the hidden charging item set as a charging item to be learned;

s0224: embedding vector hiding of all the toll items with different types from the toll items to be learned is carried out on the hidden patient embedded vector sample data, and patient embedded vector sample data to be analyzed is obtained;

s0225: inputting the patient embedding vector sample data to be analyzed into the patient behavior learning model to be trained for patient behavior learning to obtain a similar prediction patient behavior training vector;

s0226: extracting a training vector corresponding to the toll collection project to be learned from the similar prediction patient behavior training vectors to obtain a training vector to be calculated corresponding to the toll collection project to be learned;

s0227: repeatedly executing the step of extracting one charging item from the hidden charging item set as a charging item to be learned until the training vector to be calculated corresponding to all the charging items in the hidden charging item set is determined;

s0228: taking all the training vectors to be calculated as the second training vectors;

s0229: extracting training vectors corresponding to the positions of the identification symbols from the patient behavior training vectors to be analyzed to obtain identification symbol training vectors;

s02210: and sequentially inputting the identification symbol training vector into the full-connection layer and the softmax layer to carry out total cost classification prediction to obtain the total cost classification probability training vector.

The embodiment trains the model based on three training methods, so that the trained model can better learn the behavior of the patient, and the accuracy of identifying the abnormal patient is improved.

For S0221, the hidden patient embedded vector sample data is input into the patient behavior learning model to be trained to perform patient behavior learning, and the learned data is used as the patient behavior training vector to be analyzed, so that the hidden patient behavior learning of the toll collection items is performed by adopting the embedded vectors of all the toll collection items which are not hidden.

And for S0222, sequentially extracting training vectors corresponding to each charging item in the hidden charging item set from the training vectors of the patient behaviors to be analyzed, and taking all extracted data as the first training vector.

For S0223, one of the toll items is sequentially extracted from the hidden toll item set, and the extracted toll item is taken as a toll item to be learned.

For S0224, for the hidden patient embedded vector sample data, performing embedded vector hiding of all the toll items different from the type of the toll item to be learned, thereby retaining the embedded vectors of all the toll items same as the type of the toll item to be learned. That is, the types of the charging items corresponding to the data which is not replaced by the mask in the patient embedded vector sample data to be analyzed are the same.

For S0225, the patient embedded vector sample data to be analyzed is input into the patient behavior learning model to be trained for patient behavior learning, and the learned vector is used as a similar prediction patient behavior training vector, so that the hidden patient behavior learning of the toll collection project is realized by adopting the sample embedded vectors of all the toll collection projects with the same type as the hidden toll collection project.

And S0226, extracting the training vector corresponding to the toll collection item to be learned from the similar prediction patient behavior training vector, and taking the extracted data as the training vector to be calculated corresponding to the toll collection item to be learned. That is, the training vector to be calculated is the learning result corresponding to the tolling item to be learned, which is learned by the patient behavior of the tolling item concealed by using the sample embedding vector of all the tolling items of the same type as the concealed tolling item to be learned.

For S0227, repeatedly performing steps S0223 to S0227 until determining the training vector to be calculated corresponding to each of all the charging items in the hidden charging item set.

And S0229, extracting training vectors corresponding to the positions of the identification symbols from the patient behavior training vectors to be analyzed, and taking the extracted data as the identification symbol training vectors.

For S02210, the method for sequentially inputting the identifier training vectors into the fully-connected layer and the softmax layer to perform the total cost classification prediction may be selected from the prior art, and is not repeated here.

In an embodiment, the step of calculating the target calibration vector, the first training vector, the second training vector, the total cost classification probability training vector, and the total cost classification calibration vector of the target patient training sample by inputting the loss function to obtain the target loss value includes:

s0291: inputting the target calibration vector and the first training vector into a first cross entropy loss function for calculation to obtain a first loss value;

s0292: inputting the target calibration vector and the second training vector into a second cross entropy loss function for calculation to obtain a second loss value;

s0293: inputting the total cost classification probability training vector and the total cost classification calibration vector of the target patient training sample into a third cross entropy loss function for calculation to obtain a third loss value;

s0294: and adding and calculating the first loss value, the second loss value and the third loss value to obtain the target loss value.

In the embodiment, an independent loss function is set for each training method, and then the loss values of the three training methods are added, so that the three training methods are trained to be optimal at the same time, and the patient behavior learning is facilitated.

For S0291, the method for calculating by inputting the target calibration vector and the first training vector into the first cross entropy loss function is selected from the prior art, and is not described herein again.

For S0292, the method for calculating by inputting the target calibration vector and the second training vector into the second cross entropy loss function is selected from the prior art, and is not described herein again.

For S0293, the method for calculating by inputting the total cost classification probability training vector and the total cost classification calibration vector of the target patient training sample into a third cross entropy loss function is selected from the prior art, and details are not repeated herein.

In an embodiment, the step of inputting each of the patient-embedded vector sets into a target patient behavior learning model for patient behavior learning to obtain the patient behavior vectors corresponding to the patient-embedded vector sets includes:

s21: acquiring a patient embedding vector set from the plurality of patient embedding vector sets to obtain a target patient embedding vector set;

s22: inputting the set of target patient embedding vectors into the input layer of the target patient behavior learning model;

s23: performing vector addition calculation on the charging item name embedding vector, the charging item type embedding vector and the charging item cost embedding vector of the target patient embedding vector set by adopting the input layer of the target patient behavior learning model to obtain a vector to be spliced;

s24: sequentially splicing the identification symbols and the vectors to be spliced by adopting the input layer of the target patient behavior learning model to obtain target vectors to be analyzed;

s25: coding and learning the target vector to be analyzed by adopting the 12-layer coder of the target patient behavior learning model to obtain a target coding vector;

s26: vector output is carried out on the target coding vector by adopting the output layer of the target patient behavior learning model, and the patient behavior vector corresponding to the target patient embedding vector set is obtained;

s27: and repeating the step of obtaining a patient embedding vector set from the plurality of patient embedding vector sets to obtain a target patient embedding vector set until the patient behavior vectors corresponding to the plurality of patient embedding vector sets are determined.

The user-defined input layer is adopted to replace the original embedded layer of the Bert model, so that the information of the toll project can be well learned, the behavior learning of the patient can be better performed, and the accuracy of abnormal patient identification can be improved.

For S21, a set of patient embedding vectors is obtained from the plurality of sets of patient embedding vectors, resulting in a set of target patient embedding vectors.

For S22, the set of target patient embedding vectors is input to an embedding vector addition module of the input layer of the target patient behavior learning model.

For S23, an embedding vector addition module of the input layer of the target patient behavior learning model is adopted to perform vector addition calculation on the toll project name embedding vector, the toll project type embedding vector, and the toll project cost embedding vector of the target patient embedding vector set to obtain a one-dimensional vector, and the one-dimensional vector is used as a vector to be spliced.

And S24, sequentially splicing the identification symbols and the vectors to be spliced by adopting the vector splicing module of the input layer of the target patient behavior learning model to obtain one-dimensional vectors, and taking the one-dimensional vectors obtained by splicing as the vectors to be analyzed.

Optionally, the identification symbol is set to CLS. The identification symbol, i.e. the mark corresponding to the mark bit in Bert, is used for subsequent classification.

And S25, coding and learning the target vector to be analyzed by adopting the 12-layer coder of the target patient behavior learning model, and taking the learned data as a target coding vector.

For S26, the output layer of the target patient behavior learning model is used to perform vector output on the target encoding vector, and the output vector is used as the patient behavior vector corresponding to the target patient embedding vector set.

For S27, repeating steps S21-S27 until the patient behavior vectors corresponding to each of the plurality of sets of patient-embedded vectors are determined.

In an embodiment, the step of clustering all the behavior vectors to be clustered by using the DBSCAN clustering algorithm to obtain a plurality of patient behavior vector cluster sets, and performing abnormal patient detection according to the plurality of patient behavior vector cluster sets to obtain an abnormal patient behavior vector set includes:

s41: acquiring preset clustering parameter radius and a preset minimum vector number required for forming a high-density area;

s42: based on the preset minimum vector number and the Euclidean distance algorithm, clustering all the behavior vectors to be clustered by adopting the DBSCAN clustering algorithm to obtain a plurality of patient behavior vector clustering sets;

s43: respectively establishing a clustering contour for each patient behavior vector clustering set by adopting the preset clustering parameter radius to obtain target clustering contours corresponding to the patient behavior vector clustering sets;

s44: and based on the target cluster contour corresponding to each of the plurality of patient behavior vector cluster sets, performing abnormal patient detection on the plurality of patient behavior vector cluster sets to obtain the abnormal patient behavior vector set.

According to the embodiment, abnormal patient identification is performed based on the DBSCAN clustering algorithm, and clusters in any shapes are found from the data to be identified with noise, so that the accuracy of abnormal patient detection is improved.

For S41, the preset clustering parameter radius and the preset minimum number of vectors needed to form the high-density region may be obtained from the database, or the preset clustering parameter radius and the preset minimum number of vectors needed to form the high-density region may be input by the user, or the preset clustering parameter radius and the preset minimum number of vectors needed to form the high-density region may be sent by the third-party application system, or the preset clustering parameter radius and the preset minimum number of vectors needed to form the high-density region may be written in the program file implementing the present application.

For step S42, based on the preset minimum vector number and the euclidean distance algorithm, the DBSCAN clustering algorithm is adopted to cluster all the behavior vectors to be clustered, and each clustered set obtained by clustering is used as a patient behavior vector clustered set. Based on the preset minimum vector number and the euclidean distance algorithm, the method for clustering all the behavior vectors to be clustered by using the DBSCAN clustering algorithm may be selected from the prior art, and is not described herein again.

For S43, the method for establishing the clustering contour for each patient behavior vector clustering set by using the preset clustering parameter radius may be selected from the prior art, and will not be described herein again.

For S44, filtering each patient behavior vector cluster set based on the target cluster contour corresponding to each of the plurality of patient behavior vector cluster sets, performing abnormal patient detection according to the filtered patient behavior vector cluster sets, determining whether the detected behavior vector to be clustered and the core point corresponding to the filtered patient behavior vector cluster set are density-reachable, and determining that the patient corresponding to the detected behavior vector to be clustered is an abnormal patient if the detected behavior vector to be clustered and all the filtered patient behavior vector cluster sets do not satisfy density-reachable, and at this time, taking the detected behavior vector to be clustered as an abnormal patient behavior vector; and taking all abnormal patient behavior vectors as the abnormal patient behavior vector set.

In an embodiment, the step of performing abnormal patient detection on the plurality of patient behavior vector cluster sets based on the target cluster contours corresponding to the plurality of patient behavior vector cluster sets respectively to obtain the abnormal patient behavior vector set includes:

s441: extracting one behavior vector to be clustered from the plurality of patient behavior vector clustering sets to serve as an object to be detected;

s442: extracting one target clustering contour from the target clustering contours corresponding to the plurality of patient behavior vector clustering sets respectively to serve as a clustering contour to be detected;

s443: judging whether the density of all core points of the patient behavior vector clustering set corresponding to the object to be detected and the cluster contour to be detected is reachable or not, and obtaining a density reachable result corresponding to the cluster contour to be detected;

s444: repeatedly executing the step of extracting one target cluster contour from the target cluster contours corresponding to the plurality of patient behavior vector cluster sets respectively, and taking the target cluster contour as a cluster contour to be detected until the density reachable result corresponding to each target cluster contour is determined;

s445: when all the density reachable results are density unreachable, determining the behavior vector to be clustered corresponding to the object to be detected as an abnormal patient behavior vector;

s446: repeatedly executing the step of extracting one behavior vector to be clustered from the plurality of patient behavior vector clustering sets as an object to be detected until the extraction of all the behavior vectors to be clustered in the plurality of patient behavior vector clustering sets is completed;

s447: and taking all the abnormal patient behavior vectors as the abnormal patient behavior vector set.

According to the method and the device, the abnormal patient detection is performed on the plurality of patient behavior vector cluster sets based on the target cluster outlines corresponding to the plurality of patient behavior vector cluster sets, so that the speed of determining the abnormal patient behavior vector sets is improved.

For step S441, one behavior vector to be clustered is extracted from the multiple patient behavior vector clustering sets, and the extracted behavior vector to be clustered is used as an object to be detected.

For step S442, one target clustering contour is extracted from the target clustering contours corresponding to the plurality of patient behavior vector clustering sets, and the extracted target clustering contour is used as a to-be-detected clustering contour.

For S443, the core point of the patient behavior vector cluster set refers to a core object of the patient behavior vector cluster set determined by the DBSCAN clustering algorithm.

When the Euclidean distances between the object to be detected and all core points of the patient behavior vector clustering set corresponding to the cluster contour to be detected are larger than the preset clustering parameter radius, determining that the density reachable result corresponding to the cluster contour to be detected is density unreachable, and otherwise determining that the density reachable result corresponding to the cluster contour to be detected is density reachable.

For S444, repeating steps S442 to S444 until the density achievable result corresponding to each of all the target cluster contours is determined.

For S445, when all the density reachable results are that the density is not reachable, it means that the patient corresponding to the object to be detected is abnormal, and at this time, it may be determined that the behavior vector to be clustered corresponding to the object to be detected is an abnormal patient behavior vector.

For S446, repeatedly executing steps S441 to S446 until the extraction of all the behavior vectors to be clustered in the plurality of patient behavior vector clustering sets is completed.

Referring to fig. 2, the present application also proposes an artificial intelligence based abnormal patient identification apparatus, the apparatus comprising:

an embedded vector acquisition module 100 configured to acquire a plurality of patient embedded vector sets, the patient embedded vector sets comprising: a toll item name embedding vector, a toll item type embedding vector and a toll item charge embedding vector corresponding to a plurality of toll items of a patient;

a patient behavior learning module 200, configured to input each patient embedded vector set into a target patient behavior learning model for patient behavior learning, so as to obtain patient behavior vectors corresponding to the patient embedded vector sets, where the target patient behavior learning model includes: an input layer, a 12-layer encoder, an output layer;

the vector extraction module 300 is configured to obtain an identifier position, and perform vector extraction on each patient behavior vector according to the identifier position to obtain behavior vectors to be clustered, which correspond to the patient embedding vector sets respectively;

the abnormal patient detection module 400 is configured to cluster all the behavior vectors to be clustered by using a DBSCAN clustering algorithm to obtain a plurality of patient behavior vector cluster sets, and perform abnormal patient detection according to the plurality of patient behavior vector cluster sets to obtain an abnormal patient behavior vector set;

and an abnormal patient set determining module 500, configured to determine abnormal patients according to the abnormal patient behavior vector set, so as to obtain an abnormal patient set.

Referring to fig. 3, a computer device, which may be a server and whose internal structure may be as shown in fig. 3, is also provided in the embodiment of the present application. The computer device includes a processor, a memory, a network interface, and a database connected by a system bus. Wherein the computer designed processor is used to provide computational and control capabilities. The memory of the computer device comprises a nonvolatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used for storing data such as abnormal patient identification methods based on artificial intelligence. The network interface of the computer device is used for communicating with an external terminal through a network connection. The computer program is executed by a processor to implement an artificial intelligence based abnormal patient identification method. The abnormal patient identification method based on artificial intelligence comprises the following steps: obtaining a plurality of sets of patient embedding vectors, the sets of patient embedding vectors comprising: a toll item name embedding vector, a toll item type embedding vector and a toll item charge embedding vector corresponding to a plurality of toll items of a patient; inputting each patient embedding vector set into a target patient behavior learning model for patient behavior learning to obtain patient behavior vectors corresponding to the plurality of patient embedding vector sets, wherein the target patient behavior learning model comprises: an input layer, a 12-layer encoder, an output layer; acquiring an identifier position, and respectively performing vector extraction on each patient behavior vector according to the identifier position to obtain behavior vectors to be clustered, which correspond to the patient embedding vector sets respectively; clustering all the behavior vectors to be clustered by adopting a DBSCAN clustering algorithm to obtain a plurality of patient behavior vector clustering sets, and detecting abnormal patients according to the plurality of patient behavior vector clustering sets to obtain abnormal patient behavior vector sets; and determining abnormal patients according to the abnormal patient behavior vector set to obtain an abnormal patient set.

An embodiment of the present application further provides a computer-readable storage medium having stored thereon a computer program which, when executed by a processor, implements an artificial intelligence based abnormal patient identification method, including the steps of: obtaining a plurality of sets of patient embedding vectors, the sets of patient embedding vectors comprising: a toll item name embedding vector, a toll item type embedding vector and a toll item charge embedding vector corresponding to a plurality of toll items of a patient; inputting each patient embedding vector set into a target patient behavior learning model for patient behavior learning to obtain patient behavior vectors corresponding to the plurality of patient embedding vector sets, wherein the target patient behavior learning model comprises: an input layer, a 12-layer encoder, an output layer; acquiring an identifier position, and respectively performing vector extraction on each patient behavior vector according to the identifier position to obtain behavior vectors to be clustered, which correspond to the patient embedding vector sets respectively; clustering all the behavior vectors to be clustered by adopting a DBSCAN clustering algorithm to obtain a plurality of patient behavior vector clustering sets, and detecting abnormal patients according to the plurality of patient behavior vector clustering sets to obtain abnormal patient behavior vector sets; and determining abnormal patients according to the abnormal patient behavior vector set to obtain an abnormal patient set.

In the above abnormal patient identification method based on artificial intelligence, patient behavior learning is performed by inputting each patient embedded vector set into the target patient behavior learning model, and patient behavior vectors corresponding to the plurality of patient embedded vector sets are obtained, where the patient embedded vector set includes: the method comprises the steps of embedding charge item name vectors, charge item type embedded vectors and charge item cost embedded vectors corresponding to a plurality of charge items of a patient, extracting vectors of single charge items of each patient behavior vector according to identification symbol positions to obtain behavior vectors to be clustered of a plurality of patient embedded vector sets, clustering all the behavior vectors to be clustered by adopting a DBSCAN clustering algorithm to obtain a plurality of patient behavior vector cluster sets, carrying out abnormal patient detection according to the plurality of patient behavior vector cluster sets to obtain abnormal patient behavior vector sets, carrying out abnormal patient determination according to the abnormal patient behavior vector sets to obtain abnormal patient sets, carrying out abnormal identification on basic information of the patients relatively, embedding the vectors, charge item type embedded vectors and charge item cost embedded vectors according to the charge item names, charge item types, and charge item cost, The charge project cost embedding vector and the DBSCAN clustering algorithm are used for identifying abnormal patients, so that deep mining of the behaviors of the patients is realized, and the accuracy of identifying the abnormal patients is improved.

It will be understood by those skilled in the art that all or part of the processes of the methods of the embodiments described above can be implemented by hardware instructions of a computer program, which can be stored in a non-volatile computer-readable storage medium, and when executed, can include the processes of the embodiments of the methods described above. Any reference to memory, storage, database, or other medium provided herein and used in the examples may include non-volatile and/or volatile memory. Non-volatile memory can include read-only memory (ROM), Programmable ROM (PROM), Electrically Programmable ROM (EPROM), Electrically Erasable Programmable ROM (EEPROM), or flash memory. Volatile memory can include Random Access Memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms such as Static RAM (SRAM), Dynamic RAM (DRAM), Synchronous DRAM (SDRAM), double-rate SDRAM (SSRSDRAM), Enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), Rambus Direct RAM (RDRAM), direct bus dynamic RAM (DRDRAM), and bus dynamic RAM (RDRAM).

It should be noted that, in this document, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, apparatus, article, or method that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, apparatus, article, or method. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other like elements in a process, apparatus, article, or method that includes the element.

The above description is only a preferred embodiment of the present application, and not intended to limit the scope of the present application, and all modifications of equivalent structures and equivalent processes, which are made by the contents of the specification and the drawings of the present application, or which are directly or indirectly applied to other related technical fields, are also included in the scope of the present application.

Claims

1. An artificial intelligence based abnormal patient identification method, the method comprising:

2. The artificial intelligence based abnormal patient identification method according to claim 1, wherein before the step of inputting each of the patient-embedded vector sets into a target patient behavior learning model for patient behavior learning to obtain the patient behavior vectors corresponding to the patient-embedded vector sets, the method further comprises:

3. The artificial intelligence based abnormal patient identification method according to claim 2, wherein the method for learning the patient behavior of the hidden tolls using the embedded vectors of all the unhidden tolls performs the patient behavior learning according to the hidden patient embedded vector sample data and the patient behavior learning model to be trained to obtain a first training vector, the method for learning the patient behavior of the hidden tolls using the sample embedded vectors of all the tolls of the same type as the hidden tolls performs the patient behavior learning according to the hidden patient embedded vector sample data and the patient behavior learning model to be trained to determine a second training vector, and the method for learning the patient behavior according to the identifier position, the hidden patient embedded vector sample data, the hidden patient behavior learning model to be trained, The method comprises the following steps of carrying out total cost classification prediction on a patient behavior learning model to be trained, a full connection layer and a softmax layer to obtain a total cost classification probability training vector, wherein the steps comprise:

4. The artificial intelligence based abnormal patient identification method of claim 2, wherein said step of calculating said target calibration vector, said first training vector, said second training vector, said total cost classification probability training vector, said total cost classification calibration vector of said target patient training sample into a loss function to obtain a target loss value comprises:

5. The artificial intelligence based abnormal patient identification method according to claim 1, wherein the step of inputting each of the patient-embedded vector sets into a target patient behavior learning model for patient behavior learning to obtain the patient behavior vectors corresponding to the patient-embedded vector sets comprises:

6. The artificial intelligence based abnormal patient identification method according to claim 1, wherein the step of clustering all the behavior vectors to be clustered by using a DBSCAN clustering algorithm to obtain a plurality of patient behavior vector cluster sets, and performing abnormal patient detection according to the plurality of patient behavior vector cluster sets to obtain an abnormal patient behavior vector set comprises:

7. The artificial intelligence based abnormal patient identification method according to claim 6, wherein the step of performing abnormal patient detection on the plurality of patient behavior vector cluster sets based on the target cluster contours corresponding to the plurality of patient behavior vector cluster sets, to obtain the abnormal patient behavior vector set comprises:

8. An artificial intelligence based abnormal patient identification device, the device comprising:

9. A computer device comprising a memory and a processor, the memory storing a computer program, wherein the processor implements the steps of the method of any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium, on which a computer program is stored, which, when being executed by a processor, carries out the steps of the method of any one of claims 1 to 7.