CN109636061A - Training method, device, equipment and the storage medium of medical insurance Fraud Prediction network - Google Patents
Training method, device, equipment and the storage medium of medical insurance Fraud Prediction network Download PDFInfo
- Publication number
- CN109636061A CN109636061A CN201811589288.5A CN201811589288A CN109636061A CN 109636061 A CN109636061 A CN 109636061A CN 201811589288 A CN201811589288 A CN 201811589288A CN 109636061 A CN109636061 A CN 109636061A
- Authority
- CN
- China
- Prior art keywords
- patient
- doctor
- training
- medical insurance
- matrix
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q30/00—Commerce
- G06Q30/018—Certifying business or products
- G06Q30/0185—Product, service or business identity fraud
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q40/00—Finance; Insurance; Tax strategies; Processing of corporate or income taxes
- G06Q40/08—Insurance
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Strategic Management (AREA)
- General Physics & Mathematics (AREA)
- Economics (AREA)
- Marketing (AREA)
- Finance (AREA)
- General Business, Economics & Management (AREA)
- Development Economics (AREA)
- Accounting & Taxation (AREA)
- Entrepreneurship & Innovation (AREA)
- Human Resources & Organizations (AREA)
- Biomedical Technology (AREA)
- Computing Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Game Theory and Decision Science (AREA)
- Health & Medical Sciences (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Tourism & Hospitality (AREA)
- Technology Law (AREA)
- Medical Treatment And Welfare Office Work (AREA)
Abstract
The present invention is applicable in field of computer technology, provide a kind of training method of medical insurance Fraud Prediction network, device, equipment and storage medium, this method comprises: concentrating the diagnosis information of clinical samples according to training sample, obtain patient characteristic matrix, and the doctor-patient relationship between patient and doctor is analyzed according to diagnosis information, obtain the corresponding doctor-patient relationship adjacency matrix of doctor-patient relationship, according to patient characteristic matrix and doctor-patient relationship adjacency matrix, prediction training is carried out to the medical insurance Fraud Prediction network being made of figure convolutional neural networks and variation self-encoding encoder, until reaching preset frequency of training, then terminate prediction training, to which the feature to relevant patient can be learnt well by figure convolutional neural networks, and the expression information hidden in patient characteristic is excavated using variation self-encoding encoder algorithm, pass through the medical insurance Fraud Prediction to realize Network is focused to find out fraud crowd and the fraud maximum crowd of suspicion degree in medical data, improves medical insurance Fraud Prediction accuracy rate.
Description
Technical field
The invention belongs to field of computer technology more particularly to a kind of training method of medical insurance Fraud Prediction network, device,
Equipment and storage medium.
Background technique
With the transition of science and technology, medical treatment, economy, humanity etc., modern's work strain is caused, it is associated therewith to be
Live it is irregular, overrun, amount of exercise is insufficient, many ciril diseases are then taken advantage of a weak point, and it is increasing to cure difficulty, " state
Border health care " magazine once delivered grading of the relevant department to Global Health and medical conditions, and Japan is in most healthy countries and regions
Middle ranking the 5th, TaiWan, China arranges the 14th, and China's Mainland arranges the 20th, and therewith, medical treatment & health and medical expenses problem are got over
It is more concerned by people, and medical security socialization is the inexorable trend of social development.
Currently, medical insurance has become the main expenditure of many countries, reported within 2012 according to British Insurance Association, daily
15 potential medical frauds can occur per hour.The generation of medical insurance fraud causes huge punching to China's medical insurance fund
It hits, 10,000,000,000 yuan of economic loss will be caused every year.Effective detection means is only limited to artificial screening at present, however, artificial sieve
It looks into and needs artificially to formulate relevant fraud rule, rule is more inflexible, is difficult to cope with multifarious fraudulent mean；Secondly, in number
According to explosion time generation, diagnosis of the number in terms of necessarily can be all generated per minute, and the efficiency of artificial detection is extremely low；Furthermore artificial screening needs
Want the professional knowledge of more domain experts.With the development of information technology, related researcher is with machine learning algorithm to doctor
It protects fraud and has done numerous studies, mainly used support vector machines (the Support Vector in monitor mode
Machine, SVM), logistic regression, neural network scheduling algorithm, the part in unsupervised mode peels off the factor (Local Outlier
Factor, LOF), noisy density clustering method (the Density-Based Spatial Clustering of of tool
Applications with Noise, DBSCAN) scheduling algorithm, however, these algorithms are needed using a large amount of fraud sample (just
Sample) and non-fraud sample (negative sample) be used to train, and in reality cheat sample quantity be not able to satisfy trained requirement,
It is thus impossible to obtain preferable training result.
Summary of the invention
The purpose of the present invention is to provide a kind of training method of medical insurance Fraud Prediction network, device, equipment and storages to be situated between
Matter, it is intended to solve the low efficiency for causing to cheat medical insurance patient's prediction due to the prior art, and the problem that predictablity rate is not high.
On the one hand, the present invention provides a kind of training method of medical insurance Fraud Prediction network, the method includes following steps
It is rapid:
Preset quantity clinical samples are extracted from preset hospital information system, the clinical samples include having marked to take advantage of
The medical insurance fraud patient of swindleness label and the unknown patient for not marking label；
The diagnosis information that each clinical samples are concentrated according to the training sample being made of the clinical samples obtains the instruction
Practice the patient characteristic matrix of sample set, and the doctor-patient relationship between patient and doctor analyzed according to the diagnosis information,
Obtain the corresponding doctor-patient relationship adjacency matrix of the doctor-patient relationship；
According to the patient characteristic matrix and the doctor-patient relationship adjacency matrix, to the medical insurance Fraud Prediction net constructed in advance
Network carries out prediction training, until reaching preset frequency of training, then terminates the prediction training, passes through trained institute to realize
It states medical insurance Fraud Prediction network and detects the suspicious patient for being accused of medical insurance fraud in unknown patient, wherein the medical insurance fraud is pre-
Survey grid network is made of preset figure convolutional neural networks and variation self-encoding encoder, and the figure convolutional neural networks include 2 implicit
Layer.
On the other hand, the present invention provides a kind of training device of medical insurance Fraud Prediction network, described device includes:
Clinical samples extracting unit, for extracting preset quantity clinical samples, institute from preset hospital information system
Stating clinical samples includes the medical insurance fraud patient for having marked fraud label and the unknown patient for not marking label；
Matrix acquiring unit, for the diagnosis information of the clinical samples according to the training sample set, described in acquisition
The patient characteristic matrix of training sample set, and the doctor-patient relationship between patient and doctor is divided according to the diagnosis information
Analysis, obtains the corresponding doctor-patient relationship adjacency matrix of the doctor-patient relationship；And
Network training unit is used for according to the patient characteristic matrix and the doctor-patient relationship adjacency matrix, to preparatory structure
The medical insurance Fraud Prediction network built carries out prediction training, until reaching preset frequency of training, then terminates the prediction training, with
Realize the suspicious patient for detecting to be accused of medical insurance fraud in unknown patient by the trained medical insurance Fraud Prediction network,
In, the medical insurance Fraud Prediction network is made of preset figure convolutional neural networks and variation self-encoding encoder, the picture scroll product mind
It include 2 hidden layers through network.
On the other hand, the present invention also provides a kind of calculating equipment, including memory, processor and it is stored in described deposit
In reservoir and the computer program that can run on the processor, the processor are realized such as when executing the computer program
Step described in the training method of above-mentioned medical insurance Fraud Prediction network.
On the other hand, the present invention also provides a kind of computer readable storage medium, the computer readable storage mediums
It is stored with computer program, the training such as above-mentioned medical insurance Fraud Prediction network is realized when the computer program is executed by processor
Step described in method.
The present invention concentrates the diagnosis information of each clinical samples according to training sample, obtains the patient characteristic of training sample set
Matrix, and the doctor-patient relationship between patient and doctor is analyzed according to diagnosis information, obtain the corresponding doctors and patients of doctor-patient relationship
Relation adjacent matrix, according to patient characteristic matrix and doctor-patient relationship adjacency matrix, to self-editing by figure convolutional neural networks and variation
The medical insurance Fraud Prediction network of code device composition carries out prediction training, until reaching preset frequency of training, then terminates prediction training,
To realize the suspicious patient for detecting to be accused of medical insurance fraud in unknown patient by the medical insurance Fraud Prediction network, to pass through figure
Convolutional neural networks can learn the feature to relevant patient well, and excavate patient characteristic using variation self-encoding encoder algorithm
In hide expression information, with realize medical data be focused to find out fraud crowd and fraud the maximum crowd of suspicion degree, mention
High medical insurance Fraud Prediction accuracy rate.
Detailed description of the invention
Fig. 1 is the implementation flow chart of the training method for the medical insurance Fraud Prediction network that the embodiment of the present invention one provides；
Fig. 2 is the doctor-patient relationship pessimistic concurrency control schematic diagram that the embodiment of the present invention one provides；
Fig. 3 is the schematic diagram for the medical insurance Fraud Prediction network that the embodiment of the present invention one provides；
Fig. 4 is the structural schematic diagram of the training device of medical insurance Fraud Prediction network provided by Embodiment 2 of the present invention；
Fig. 5 is the preferred structure schematic diagram of the training device of medical insurance Fraud Prediction network provided by Embodiment 2 of the present invention；
And
Fig. 6 is the structural schematic diagram for the calculating equipment that the embodiment of the present invention three provides.
Specific embodiment
In order to make the objectives, technical solutions, and advantages of the present invention clearer, with reference to the accompanying drawings and embodiments, right
The present invention is further elaborated.It should be appreciated that the specific embodiments described herein are merely illustrative of the present invention, and
It is not used in the restriction present invention.
Specific implementation of the invention is described in detail below in conjunction with specific embodiment:
Embodiment one:
Fig. 1 shows the implementation process of the training method of the medical insurance Fraud Prediction network of the offer of the embodiment of the present invention one, is
Convenient for explanation, only parts related to embodiments of the present invention are shown, details are as follows:
In step s101, preset quantity clinical samples are extracted from preset hospital information system.
The embodiment of the present invention is suitable for medical computing processing platform, equipment or system, for example, personal computer, server
Deng.All numbers stored in preset hospital information system (Hospital Information System, HIS) database
According to data pick-up attribute is determined in attribute, according to determining data pick-up attribute (for example, medical odd numbers, medical certificate number, Bing Renxing
Not, age, medical insurance deduct fees, at one's own expense, purchase medicine quantity, drug name, medicinal number of days, medical insurance type, doctor ID, drug unit price, just
Examine department etc.), preset quantity clinical samples are extracted from HIS database using SQL statement, which includes that medical insurance is taken advantage of
Cheat patient and unknown patient, wherein medical insurance fraud patient has passed through the mark for manually having carried out fraud label, i.e. the medical insurance is cheated
The sample of patient is tape label sample, and the sample of unknown patient is the unlabeled exemplars without carrying out label for labelling, i.e., to be predicted
Sample, label include fraud label (being expressed as 1) and non-fraud label (being expressed as 0).
When extracting preset quantity clinical samples from hospital information system, it is preferable that the clinical samples that will be extracted
It is pre-processed, obtains pretreated clinical samples, to reduce the noise in clinical samples, improve the available of clinical samples
Property.
When the clinical samples extracted are pre-processed, it is preferable that realized by following step to clinical samples
Pretreatment:
1) missing values processing is carried out to clinical samples and specifically judges that each key data attribute is corresponding in clinical samples
Whether numerical value lacks, if missing, calculates the approximation of the numerical value of missing, using Lagrange's interpolation to pass through the approximation
Value carries out data filling to the corresponding numerical value of the data attribute；
2) outlier processing is carried out to the clinical samples obtained after missing values processing and specifically passes through interquartile-range IQR
The method of (Interquartile Range, IQR) judges that whether the corresponding numerical value of each data attribute is less than first in clinical samples
Exceptional value (QL- 1.5IQR) or greater than the abnormal (Q of the second exceptional valueU+ 1.5IQR), abnormal numerical value if it exists, then by abnormal numerical value
It is deleted from the clinical samples, or takes the average value of the regime values of corresponding data attribute in several other clinical samples, it will
Abnormal numerical value replaces with the average value, or the exception numerical value is considered as missing values, carries out data using Lagrange's interpolation
Fitting, to fill up the exception numerical value, wherein IQR=F-1(0.75)-F-1(0.25), lower quartile QL=F-1(0.25) it indicates
In clinical samples after the corresponding all ascending arrangements of numerical value of each data attribute the 25%th numerical value, upper quartile QU=F-1
(0.75) the after the corresponding all ascending arrangements of numerical value of each data attribute the 75%th numerical value is indicated in clinical samples；
3) clinical samples obtained after outlier processing are returned by just too standardization (Z-score standardization) method
One change processing.
1)-step 3) realizes the pretreatment to clinical samples through the above steps, to reduce in obtained clinical samples
Noise, improve the availability and reliability of clinical samples.
In step s 102, the diagnosis information of each clinical samples is concentrated according to the training sample being made of clinical samples,
The patient characteristic matrix of training sample set is obtained, and the doctor-patient relationship between patient and doctor is divided according to diagnosis information
Analysis obtains the corresponding doctor-patient relationship adjacency matrix of doctor-patient relationship.
In embodiments of the present invention, all clinical samples extracted are formed into training sample set, is concentrated from training sample
It is extracted in all diagnosis informations of each clinical samples corresponding just with key data attribute (i.e. the main feature) chosen in advance
Information is examined, the patient characteristic matrix X of N × C dimension is made up of the diagnosis information extracted, and according to all of each clinical samples
Diagnosis information analyzes the doctor-patient relationship between patient and doctor, obtains doctor-patient relationship pessimistic concurrency control, and then according to the doctors and patients
Discussion on relation obtains the doctor-patient relationship adjacency matrix A of the corresponding N × N-dimensional of doctor-patient relationship, wherein and N is the number of clinical samples,
C is the number of main data attribute.
As illustratively, Fig. 2 shows a doctor-patient relationship pessimistic concurrency controls, wherein P is patient's set, and M is doctor's set, Vij
For the side right weight of the doctor-patient relationship pessimistic concurrency control, i-th of patient P is indicatediAccess j-th of doctor MjNumber.
Before the patient characteristic matrix for obtaining training sample set, it is preferable that from all data attributes of clinical samples
Key data attribute is chosen, table 1 shows the main feature selected, thus by extracting the spy being of practical significance to algorithm
Sign, improves the predictablity rate for the medical insurance Fraud Prediction network that subsequent training obtains.
| Attribute-name | Attribute meaning |
| Ylzh | Medical certificate number |
| Jzhm | Medical number |
| Xb | Gender |
| Age | Age |
| Brxz | Medical insurance classification |
| Max_ypdj | Average drug unit price maximum every time |
| Min_ypdj | Average drug unit price minimum every time |
| Yyts | Average medication number of days maximum every time |
| Sum_sl | It is average to buy medicine quantity every time |
| Sum_zf | Averagely every time at one's own expense |
| Sum_zje | It is average to see a doctor the amount of money every time |
| Kbcs | See a doctor number |
| Zsum_zje | History sees a doctor total amount |
1 main feature of table
In step s 103, according to patient characteristic matrix and doctor-patient relationship adjacency matrix, the medical insurance constructed in advance is cheated
Prediction network carries out prediction training, until reaching preset frequency of training, then terminates prediction training, to realize by trained
Medical insurance Fraud Prediction network detects the suspicious patient for being accused of medical insurance fraud in unknown patient.
In embodiments of the present invention, according to patient characteristic matrix and doctor-patient relationship adjacency matrix, to it is constructing in advance, by pre-
If figure convolutional neural networks (Graph Convolution Neural Network, GCNN) and variation self-encoding encoder
The medical insurance Fraud Prediction network of (Variational Auto-Encoder, VAE) composition carries out prediction training, until reaching default
Frequency of training, then terminate prediction training, with complete concentrate the prediction of unknown patient to classify training sample, that is, predict unknown
Patient is that patient or non-fraud patient are cheated in medical insurance, and realize and detect medical treatment by trained medical insurance Fraud Prediction network
It is accused of the suspicious patient of medical insurance fraud in other unknown patients of data set, wherein figure convolutional neural networks include 2 implicit
Layer, variation self-encoding encoder are connected with second hidden layer in figure convolutional neural networks hidden layer.
In embodiments of the present invention, it is preferable that the propagation formula of every layer of figure convolutional neural networks isIt is imitated to improve the study for carrying out network characterisation study to figure convolutional neural networks
Fruit.Wherein, H(l)Indicate the output of l layers of figure convolutional neural networks,X is patient characteristic square
Battle array, A are doctor-patient relationship adjacency matrix, InFor unit matrix,It is doctor-patient relationship adjacency matrix from ring matrix,ForDegree
Matrix, W(l)For the weight matrix of figure convolutional neural networks l layer network, σ () is sigmoid activation primitive.
Before carrying out prediction training to medical insurance Fraud Prediction network, it is preferable that the posteriority point of building variation self-encoding encoder
Cloth functionSubsequent medical insurance Fraud Prediction network is trained to improve
Training effect.WhereinZ is from Posterior distrbutionp function q (Z | X, A, y)
Sample obtained sampling matrix, μiFor the corresponding mean value of i-th of clinical samples, σiFor the corresponding standard side of i-th of clinical samples
Difference, ziFor from μiAnd σiGaussian ProfileMiddle to sample obtained sampled data, y is training sample set
In the corresponding tag set of each clinical samples, tag set includes the corresponding fraud label of medical insurance fraud patient in training sample set
(i.e. y=1) and the corresponding Unknown Label of unknown patient, yiFor the corresponding label of the i-th clinical samples, the corresponding patient of label is cheated
Sample is positive sample.
It is another preferably, construct variation self-encoding encoder generation model
To improve the subsequent training effect being trained to medical insurance Fraud Prediction network.Wherein,For side prediction model, side corresponded between patient and patient, patient and doctor
Between relationship, indicate by the side prediction model to predict relationship A between i-th of node and j-th of nodeijBelong to classification 1
The probability of (there is medical insurance fraud relationship),For node prediction model, section
The corresponding clinical samples of point, indicate to predict the corresponding label y of i-th of node by the node prediction modeliBelong to 1 (class of classification
Other 1 i.e. medical insurance cheat patient categories) probability, ziAnd zjI-th and j sampled data in respectively sampling matrix Z, N is patient's sample
This number, σ () are sigmoid activation primitive, and θ is the logistic regression factor to be trained, and b is bias term to be trained.
When carrying out predicting to train to the medical insurance Fraud Prediction network constructed in advance, it is preferable that by maximizing preparatory structure
Evidence lower bound objective function (Evidence Lower Bound Objective, ELBO) build, variation, to figure convolutional Neural
Network and variation self-encoding encoder carry out joint training, and the formula of the evidence lower bound objective function indicates are as follows: L=Eq(Z|X,A,y)
[logp (y, A | Z)]-KL [q (Z | X, A, y) | | p (Z)], to improve the training effect of medical insurance Fraud Prediction network.Wherein, KL
[q (Z | X, A, y) | | p (Z)] is KL divergence, for calculating the Posterior distrbutionp function q constructed in advance (Z | X, A, y) and prior distribution
The diversity factor of function p (Z), Eq(Z|X,A,y)[logp (y, A | Z)] be the corresponding label y of clinical samples that is concentrated to training sample into
The desired value of row fraud Tag Estimation, and between the corresponding patient of doctor-patient relationship adjacency matrix A and patient, patient and doctor it
Between the desired value predicted of relationship, p (y, A | Z) is the generation model of variation self-encoding encoder, and X is patient characteristic matrix, and A is
Doctor-patient relationship adjacency matrix, Z are the sampling matrix that sampling obtains from Posterior distrbutionp function q (Z | X, A, y).
It is further preferred that realizing that the maximization to ELBO function solves by following step:
1) it is by ELBO functional transformation according to the generation model of variation self-encoding encoderWherein,
Eq(Z|X,A,y)[logp (y | Z)] it is the desired value that medical insurance Fraud Prediction is carried out to node y, value is the bigger the better, Eq(Z|X,A,y)
[logp (A | Z)] relationship between the corresponding patient of doctor-patient relationship adjacency matrix A and patient, between patient and doctor is carried out in advance
The desired value of survey, value are the bigger the better, and KL [q (Z | X, A, y) | | p (Z)] is for calculating Posterior distrbutionp function q (Z | X, A, y)
With the diversity factor of prior density function p (Z), it is worth the smaller the better；
2) E is measured by the way of calculating intersection entropy lossq(Z|X,A,y)[logp (y | Z)] and Eq(Z|X,A,y)[logp(A|
Z desired value)], obtains
, whereinM is
The number of medical insurance fraud patient in training sample set；
3) KL divergence is transformed to
Wherein, J is the dimension of sampling matrix Z, μjThe component of j-th of dimension of mean value, σ are corresponded to for clinical samplesjIt is corresponding for clinical samples
The component of j-th of dimension of standard variance.
To through the above steps 1) -3) improve convergence speed of the algorithm.
When carrying out joint training to figure convolutional neural networks and variation self-encoding encoder, it is preferable that according to preset mean value
The corresponding patient's hidden feature of patient characteristic matrix learnt by picture scroll product neural network is set variation by formula
The mean value of self-encoding encoder, and the doctor that will be learnt by picture scroll product neural network according to preset standard variance logarithmic formula
Suffer from the logarithm that the corresponding patient's relationship hidden feature of relation adjacent matrix is set as the standard variance of variation self-encoding encoder, to be
The training of medical insurance Fraud Prediction network provides binding character, improves the training effect of medical insurance Fraud Prediction network.
It is further preferred that mean value formula is μ=GCNμ(H(1),A；Wμ), standard variance logarithmic formula is log σ=GCNσ
(H(1),A；Wσ), to further increase the training effect of medical insurance Fraud Prediction network, wherein
μ is the mean value of variation self-encoding encoder, and log σ is the logarithm of the standard variance of variation self-encoding encoder, H(1)For figure convolutional neural networks
Hidden layer in the first hidden layer output, W0Connected between the first hidden layer and the input layer of figure convolutional neural networks
One weight matrix, WμTo be used to carry out net list according to patient characteristic matrix in the second hidden layer in hidden layer and the first hidden layer
Levy the second weight matrix connected between the network node of study, WσTo be used in the second hidden layer and the first hidden layer according to doctor
Suffer from the third weight matrix connected between the network node of relation adjacent matrix progress network characterisation study, GCN () is picture scroll
The figure convolution operation of product neural network.
In embodiments of the present invention, it is preferable that during carrying out prediction training to medical insurance Fraud Prediction network, use
Adam optimizes (Adam Optimization) method to update all parameters in medical insurance Fraud Prediction network, and parameter includes figure
The connection weight matrix W of network layer in convolutional neural networks0、WμAnd WσWith logistic regression factor θ in variation self-encoding encoder, partially
A b is set, to improve the training effect and training speed of medical insurance Fraud Prediction network.
As illustratively, Fig. 3 shows the schematic diagram of a medical insurance Fraud Prediction network, by patient characteristic matrix X and doctors and patients
Relation adjacent matrix A is input in figure convolutional neural networks, carries out network by the layer-by-layer propagation formula of figure convolutional neural networks
Representative learning, the individual features vector learnt are input in variation self-encoding encoder, are completed by maximizing ELBO function to figure
The joint training of convolutional neural networks and variation self-encoding encoder, and during training, to network in figure convolutional neural networks
The connection weight matrix W of layer0(W0The weight matrix connected between the first hidden layer and input layer), Wμ(WμFor the second hidden layer
The weight square connected between the network node in the first hidden layer for carrying out network characterisation study according to patient characteristic matrix X
Battle array) and Wσ(WσTo be used to carry out network characterisation according to doctor-patient relationship adjacency matrix A in the second hidden layer and the first hidden layer
The weight matrix connected between the network node of habit) and variation self-encoding encoder in logistic regression factor θ, bias term b be updated,
Until reaching frequency of training, to realize that predicting unknown patient by the node prediction model in figure convolutional neural networks is accused of curing
Protect fraud probability, and by the side prediction model in figure convolutional neural networks predict with medical insurance cheat patient it is related
Suspicious crowd.
In the embodiment of the present invention, the diagnosis information of each clinical samples is concentrated according to training sample, obtains training sample set
Patient characteristic matrix, and the doctor-patient relationship between patient and doctor is analyzed according to diagnosis information, obtains doctor-patient relationship
Corresponding doctor-patient relationship adjacency matrix, according to patient characteristic matrix and doctor-patient relationship adjacency matrix, to by figure convolutional neural networks
Prediction training is carried out with the medical insurance Fraud Prediction network of variation self-encoding encoder composition, until reaching preset frequency of training, is then tied
Beam prediction training, to realize the suspicious trouble for detecting to be accused of medical insurance fraud in unknown patient by the medical insurance Fraud Prediction network
Person to can learn the feature to relevant patient well by figure convolutional neural networks, and is calculated using variation self-encoding encoder
Method excavates the expression information hidden in patient characteristic, is focused to find out fraud crowd and fraud suspicion degree in medical data to realize
Maximum crowd effectively facilitates the detection of clique's crime, improves medical insurance Fraud Prediction accuracy rate.
Embodiment two:
Fig. 4 shows the structure of the training device of medical insurance Fraud Prediction network provided by Embodiment 2 of the present invention, in order to just
In explanation, only parts related to embodiments of the present invention are shown, including:
Clinical samples extracting unit 41, for extracting preset quantity clinical samples from preset hospital information system；
Matrix acquiring unit 42, for concentrating the medical of each clinical samples according to by the training sample that clinical samples form
Information, obtains the patient characteristic matrix of training sample set, and according to diagnosis information to the doctor-patient relationship between patient and doctor into
Row analysis obtains the corresponding doctor-patient relationship adjacency matrix of doctor-patient relationship；And
Network training unit 43 is used for according to patient characteristic matrix and doctor-patient relationship adjacency matrix, to the doctor constructed in advance
It protects Fraud Prediction network and carries out prediction training, until reaching preset frequency of training, then terminate prediction training, pass through instruction to realize
The medical insurance Fraud Prediction network perfected detects the suspicious patient for being accused of medical insurance fraud in unknown patient.
As shown in Figure 5, it is preferable that network training unit 43 includes:
Network training subelement 431, it is right for evidence lower bound objective function constructed in advance by maximization, variation
Figure convolutional neural networks and variation self-encoding encoder carry out joint training, and the formula of evidence lower bound objective function indicates are as follows: L=
Eq(Z|X,A,y)[logp (y, A | Z)]-KL [q (Z | X, A, y) | | p (Z)], wherein KL [q (Z | X, A, y) | | p (Z)] it is KL divergence,
For calculating the diversity factor of the Posterior distrbutionp function q constructed in advance (Z | X, A, y) and prior density function p (Z), Eq(Z|X,A,y)
[logp (y, A | Z)] it is the desired value for carrying out cheating Tag Estimation to the corresponding label y of clinical samples that training sample is concentrated, and
To the desired value that the relationship between the corresponding patient of doctor-patient relationship adjacency matrix A and patient, between patient and doctor is predicted,
P (y, A | Z) is the generation model of variation self-encoding encoder, and X is patient characteristic matrix, and A is doctor-patient relationship adjacency matrix, and Z is from rear
The sampling matrix that sampling obtains in distribution function q (Z | X, A, y) is tested, y is that training sample concentrates the corresponding label of each clinical samples
Set, tag set include in training sample set the corresponding fraud label (i.e. y=1) of medical insurance fraud patient it is corresponding with unknown patient
Unknown Label；And
Parameter updating unit 432, for medical insurance Fraud Prediction network carry out prediction training during, using Adam
Optimization method updates all parameters in medical insurance Fraud Prediction network.
Wherein, network training subelement 431 includes:
Mean variance setting unit 4311, for picture scroll product neural network will to be passed through according to preset mean value formula
The corresponding patient's hidden feature of the patient characteristic matrix practised is set as the mean value of variation self-encoding encoder, and according to preset standard
Variance logarithmic formula is hidden by the corresponding patient's relationship of doctor-patient relationship adjacency matrix learnt by picture scroll product neural network
The logarithm of the standard variance of variation self-encoding encoder is set as containing feature.
In embodiments of the present invention, each unit of the training device of medical insurance Fraud Prediction network can be by corresponding hardware or soft
Part unit realizes that each unit can be independent soft and hardware unit, also can integrate as a soft and hardware unit, does not have to herein
To limit the present invention.Specifically, the embodiment of each unit can refer to the description of previous embodiment one, and details are not described herein.
Embodiment three:
Fig. 6 shows the structure of the calculating equipment of the offer of the embodiment of the present invention three, for ease of description, illustrates only and this
The relevant part of inventive embodiments.
The calculating equipment 6 of the embodiment of the present invention includes processor 60, memory 61 and is stored in memory 61 and can
The computer program 62 run on processor 60.The processor 60 realizes that above-mentioned medical insurance fraud is pre- when executing computer program 62
Step in the training method embodiment of survey grid network, such as step S101 to S103 shown in FIG. 1.Alternatively, processor 60 executes
The function of each unit in above-mentioned each Installation practice, such as the function of unit 41 to 43 shown in Fig. 4 are realized when computer program 62.
In embodiments of the present invention, the diagnosis information of each clinical samples is concentrated according to training sample, obtains training sample
The patient characteristic matrix of collection, and the doctor-patient relationship between patient and doctor is analyzed according to diagnosis information, it obtains doctors and patients and closes
It is corresponding doctor-patient relationship adjacency matrix, according to patient characteristic matrix and doctor-patient relationship adjacency matrix, to by picture scroll product nerve net
The medical insurance Fraud Prediction network of network and variation self-encoding encoder composition carries out prediction training, until reaching preset frequency of training, then
Terminate prediction training, to realize the suspicious trouble for detecting to be accused of medical insurance fraud in unknown patient by the medical insurance Fraud Prediction network
Person to can learn the feature to relevant patient well by figure convolutional neural networks, and is calculated using variation self-encoding encoder
Method excavates the expression information hidden in patient characteristic, is focused to find out fraud crowd and fraud suspicion degree in medical data to realize
Maximum crowd improves medical insurance Fraud Prediction accuracy rate.
The calculating equipment of the embodiment of the present invention can be personal computer, server.Processor 60 is held in the calculating equipment 6
The step of realizing when realizing the training method of medical insurance Fraud Prediction network when row computer program 62 can refer to preceding method implementation
The description of example, details are not described herein.
Example IV:
In embodiments of the present invention, a kind of computer readable storage medium is provided, which deposits
Computer program is contained, which realizes that the training method of above-mentioned medical insurance Fraud Prediction network is real when being executed by processor
The step in example is applied, for example, step S101 to S103 shown in FIG. 1.Alternatively, realization when the computer program is executed by processor
The function of each unit in above-mentioned each Installation practice, such as the function of unit 41 to 43 shown in Fig. 4.
In embodiments of the present invention, the diagnosis information of each clinical samples is concentrated according to training sample, obtains training sample
The patient characteristic matrix of collection, and the doctor-patient relationship between patient and doctor is analyzed according to diagnosis information, it obtains doctors and patients and closes
It is corresponding doctor-patient relationship adjacency matrix, according to patient characteristic matrix and doctor-patient relationship adjacency matrix, to by picture scroll product nerve net
The medical insurance Fraud Prediction network of network and variation self-encoding encoder composition carries out prediction training, until reaching preset frequency of training, then
Terminate prediction training, to realize the suspicious trouble for detecting to be accused of medical insurance fraud in unknown patient by the medical insurance Fraud Prediction network
Person to can learn the feature to relevant patient well by figure convolutional neural networks, and is calculated using variation self-encoding encoder
Method excavates the expression information hidden in patient characteristic, is focused to find out fraud crowd and fraud suspicion degree in medical data to realize
Maximum crowd improves medical insurance Fraud Prediction accuracy rate.
The computer readable storage medium of the embodiment of the present invention may include can carry computer program code any
Entity or device, recording medium, for example, the memories such as ROM/RAM, disk, CD, flash memory.
The foregoing is merely illustrative of the preferred embodiments of the present invention, is not intended to limit the invention, all in essence of the invention
Made any modifications, equivalent replacements, and improvements etc., should all be included in the protection scope of the present invention within mind and principle.
Claims (10)
1. a kind of training method of medical insurance Fraud Prediction network, which is characterized in that the method includes the following steps:
Preset quantity clinical samples are extracted from preset hospital information system, the clinical samples include having marked fraud mark
The medical insurance fraud patient of label and the unknown patient for not marking label；
The diagnosis information of each clinical samples is concentrated according to the training sample being made of the clinical samples, obtains the trained sample
The patient characteristic matrix of this collection, and the doctor-patient relationship between patient and doctor is analyzed according to the diagnosis information, it obtains
The corresponding doctor-patient relationship adjacency matrix of the doctor-patient relationship；
According to the patient characteristic matrix and the doctor-patient relationship adjacency matrix, to the medical insurance Fraud Prediction network constructed in advance into
Row prediction training then terminates the prediction training, passes through the trained doctor to realize until reaching preset frequency of training
It protects Fraud Prediction network and detects the suspicious patient for being accused of medical insurance fraud in unknown patient, wherein the medical insurance Fraud Prediction net
Network is made of preset figure convolutional neural networks and variation self-encoding encoder, and the figure convolutional neural networks include 2 hidden layers.
2. the method as described in claim 1, which is characterized in that every layer of figure convolutional neural networks of the propagation formula isWherein, H(l)Indicate l layers of figure convolutional neural networks of the output, H(0)=
X,X is the patient characteristic matrix, and A is the doctor-patient relationship adjacency matrix, InFor unit matrix,It is described
Doctor-patient relationship adjacency matrix from ring matrix,It is describedDegree matrix, W(l)For described l layers of net of figure convolutional neural networks
The weight matrix of network, σ () are sigmoid activation primitive.
3. the method as described in claim 1, which is characterized in that carry out prediction instruction to the medical insurance Fraud Prediction network constructed in advance
Experienced step, comprising:
By maximizing evidence lower bound objective function construct in advance, variation, to the figure convolutional neural networks and the change
Self-encoding encoder is divided to carry out joint training, the formula of the evidence lower bound objective function indicates are as follows: L=Eq(Z|X,A,y)[logp(y,A|
Z)]-KL [q (Z | X, A, y) | | p (Z)], wherein KL [q (Z | X, A, y) | | p (Z)] is KL divergence, is constructed in advance for calculating
The diversity factor of Posterior distrbutionp function q (Z | X, A, y) and prior density function p (Z), Eq(Z|X,A,y)[logp (y, A | Z)] for institute
The corresponding label y of the clinical samples for stating training sample concentration carries out the desired value of fraud Tag Estimation, and to the doctors and patients
The desired value that relationship between the corresponding patient of relation adjacent matrix A and patient, between patient and doctor is predicted, p (y, A |
It Z is) the generation model of the variation self-encoding encoder, X is the patient characteristic matrix, and A is the doctor-patient relationship adjacency matrix, Z
For the sampling matrix that sampling obtains from the Posterior distrbutionp function q (Z | X, A, y), y is each described in the training sample set
The corresponding tag set of clinical samples, the tag set include that the fraud patient of medical insurance described in the training sample set is corresponding
Cheat label and the corresponding Unknown Label of the unknown patient.
4. method as claimed in claim 3, which is characterized in that the figure convolutional neural networks and the variation self-encoding encoder
The step of carrying out joint training, comprising:
The patient characteristic matrix pair that will be learnt by picture scroll product neural network according to preset mean value formula
The patient's hidden feature answered is set as the mean value of the variation self-encoding encoder, and will be through according to preset standard variance logarithmic formula
The corresponding patient's relationship hidden feature of the doctor-patient relationship adjacency matrix that the picture scroll product neural network learns is crossed to set
It is set to the logarithm of the standard variance of the variation self-encoding encoder.
5. method as claimed in claim 4, which is characterized in that the mean value formula is μ=GCNμ(H(1),A；Wμ), the mark
Quasi- variance logarithmic formula is log σ=GCNσ(H(1),A；Wσ), whereinμ is the variation
The mean value of self-encoding encoder, log σ are the logarithm of the standard variance of the variation self-encoding encoder, H(1)For the figure convolutional neural networks
The hidden layer in the first hidden layer output, W0For the input layer of first hidden layer and the figure convolutional neural networks
Between the first weight matrix for connecting, WμTo be used in the second hidden layer in the hidden layer and first hidden layer according to institute
State the second weight matrix connected between the network node of patient characteristic matrix progress network characterisation study, WσIt is hidden for described second
Containing the network node for being used to carry out network characterisation study in layer and first hidden layer according to the doctor-patient relationship adjacency matrix
Between the third weight matrix that connects, GCN () is the figure convolution operation of the figure convolutional neural networks.
6. the method as described in claim 1, which is characterized in that carry out prediction instruction to the medical insurance Fraud Prediction network constructed in advance
Experienced step, further includes:
During carrying out prediction training to the medical insurance Fraud Prediction network, using Adam's optimization method to update
State all parameters in medical insurance Fraud Prediction network.
7. a kind of training device of medical insurance Fraud Prediction network, which is characterized in that described device includes:
Clinical samples extracting unit, for extracting preset quantity clinical samples, the trouble from preset hospital information system
Person's sample includes the medical insurance fraud patient for having marked fraud label and the unknown patient for not marking label；
Matrix acquiring unit obtains the training for the diagnosis information of the clinical samples according to the training sample set
The patient characteristic matrix of sample set, and the doctor-patient relationship between patient and doctor is analyzed according to the diagnosis information, it obtains
Obtain the corresponding doctor-patient relationship adjacency matrix of the doctor-patient relationship；And
Network training unit is used for according to the patient characteristic matrix and the doctor-patient relationship adjacency matrix, to what is constructed in advance
Medical insurance Fraud Prediction network carries out prediction training, until reaching preset frequency of training, then terminates the prediction training, to realize
The suspicious patient for being accused of medical insurance fraud in unknown patient is detected by the trained medical insurance Fraud Prediction network, wherein
The medical insurance Fraud Prediction network is made of preset figure convolutional neural networks and variation self-encoding encoder, the picture scroll product nerve net
Network includes 2 hidden layers.
8. device as claimed in claim 7, which is characterized in that the network training unit includes:
Network training subelement, for evidence lower bound objective function constructed in advance by maximization, variation, to the picture scroll
Product neural network and the variation self-encoding encoder carry out joint training, and the formula of the evidence lower bound objective function indicates are as follows: L=
Eq(Z|X,A,y)[logp (y, A | Z)]-KL [q (Z | X, A, y) | | p (Z)], wherein KL [q (Z | X, A, y) | | p (Z)] it is KL divergence,
For calculating the diversity factor of the Posterior distrbutionp function q constructed in advance (Z | X, A, y) and prior density function p (Z), Eq(Z|X,A,y)
[logp (y, A | Z)] it is that fraud Tag Estimation is carried out to the corresponding label y of the clinical samples that the training sample is concentrated
Desired value, and the relationship between the corresponding patient of the doctor-patient relationship adjacency matrix A and patient, between patient and doctor is carried out
The desired value of prediction, p (y, A | Z) are the generation model of the variation self-encoding encoder, and X is the patient characteristic matrix, and A is described
Doctor-patient relationship adjacency matrix, Z are the sampling matrix that sampling obtains from the Posterior distrbutionp function q (Z | X, A, y), and y is described
The corresponding tag set of each clinical samples described in training sample set, the tag set includes described in the training sample set
The corresponding fraud label of patient and the corresponding Unknown Label of the unknown patient are cheated in medical insurance.
9. a kind of calculating equipment, including memory, processor and storage are in the memory and can be on the processor
The computer program of operation, which is characterized in that the processor realizes such as claim 1 to 6 when executing the computer program
The step of any one the method.
10. a kind of computer readable storage medium, the computer-readable recording medium storage has computer program, and feature exists
In when the computer program is executed by processor the step of any one of such as claim 1 to 6 of realization the method.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811589288.5A CN109636061B (en) | 2018-12-25 | 2018-12-25 | Training method, device and equipment for medical insurance fraud prediction network and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811589288.5A CN109636061B (en) | 2018-12-25 | 2018-12-25 | Training method, device and equipment for medical insurance fraud prediction network and storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109636061A true CN109636061A (en) | 2019-04-16 |
| CN109636061B CN109636061B (en) | 2023-04-18 |
Family
ID=66077287
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811589288.5A Active CN109636061B (en) | 2018-12-25 | 2018-12-25 | Training method, device and equipment for medical insurance fraud prediction network and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109636061B (en) |
Cited By (34)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110163261A (en) * | 2019-04-28 | 2019-08-23 | 平安科技（深圳）有限公司 | Unbalanced data disaggregated model training method, device, equipment and storage medium |
| CN110188176A (en) * | 2019-04-30 | 2019-08-30 | 深圳大学 | Deep learning neural network and training, prediction technique, system, equipment, medium |
| CN110210330A (en) * | 2019-05-13 | 2019-09-06 | 清华大学 | Electromagnetic signal recognition methods and device based on Tacit Knowledge structure figures convolutional network |
| CN110245787A (en) * | 2019-05-24 | 2019-09-17 | 阿里巴巴集团控股有限公司 | A kind of target group's prediction technique, device and equipment |
| CN110263227A (en) * | 2019-05-15 | 2019-09-20 | 阿里巴巴集团控股有限公司 | Clique based on figure neural network finds method and system |
| CN110490132A (en) * | 2019-08-16 | 2019-11-22 | 京东城市(北京)数字科技有限公司 | Data processing method and device |
| CN110688536A (en) * | 2019-09-25 | 2020-01-14 | 中国建设银行股份有限公司 | Label prediction method, device, equipment and storage medium |
| CN110781406A (en) * | 2019-10-14 | 2020-02-11 | 西安交通大学 | Social network user multi-attribute inference method based on variational automatic encoder |
| CN110796497A (en) * | 2019-10-31 | 2020-02-14 | 支付宝(杭州)信息技术有限公司 | Method and device for detecting abnormal operation behaviors |
| CN110827159A (en) * | 2019-11-11 | 2020-02-21 | 上海交通大学 | Financial medical insurance fraud early warning method, device and terminal based on relational graph |
| CN111126255A (en) * | 2019-12-23 | 2020-05-08 | 华中科技大学 | Numerical control machine tool cutter wear value prediction method based on deep learning regression algorithm |
| CN111427926A (en) * | 2020-03-23 | 2020-07-17 | 平安医疗健康管理股份有限公司 | Abnormal medical insurance group identification method and device, computer equipment and storage medium |
| CN111488924A (en) * | 2020-04-07 | 2020-08-04 | 中国农业科学院农业信息研究所 | Multivariate time sequence data clustering method |
| CN111581962A (en) * | 2020-05-14 | 2020-08-25 | 福州大学 | Text representation method based on subject word vector and hybrid neural network |
| CN111612636A (en) * | 2020-04-29 | 2020-09-01 | 山东大学 | Abnormal medical insurance data detection system and method based on dual clustering algorithm |
| CN111785014A (en) * | 2020-05-26 | 2020-10-16 | 浙江工业大学 | Road network traffic data restoration method based on DTW-RGCN |
| CN111798324A (en) * | 2020-07-16 | 2020-10-20 | 浙江工业大学 | Medical insurance fraud discovery method based on dynamic hospitalizing behavior alignment |
| CN111899114A (en) * | 2020-06-23 | 2020-11-06 | 山东大学 | Doctor-seeking fraud detection method and system based on multi-view double clustering |
| CN112200684A (en) * | 2020-09-15 | 2021-01-08 | 深圳大学 | Method, system and storage medium for detecting medical insurance fraud |
| CN112331230A (en) * | 2020-11-17 | 2021-02-05 | 平安科技（深圳）有限公司 | Method and device for identifying fraudulent conduct, computer equipment and storage medium |
| CN112435133A (en) * | 2020-11-18 | 2021-03-02 | 厦门理工学院 | Medical insurance combined fraud detection method, device and equipment based on graph analysis |
| CN112801804A (en) * | 2021-01-15 | 2021-05-14 | 浙江大学山东工业技术研究院 | Method for detecting map neural network of abnormal medical insurance data |
| CN112991075A (en) * | 2021-02-04 | 2021-06-18 | 浙江大学山东工业技术研究院 | Package type medicine purchasing abnormity detection method based on FP-growth and graph network |
| CN113139586A (en) * | 2021-03-31 | 2021-07-20 | 同济大学 | Model training method, device abnormality diagnosis method, electronic device, and medium |
| CN113179250A (en) * | 2021-03-26 | 2021-07-27 | 北京六方云信息技术有限公司 | Web unknown threat detection method and system |
| CN113538126A (en) * | 2021-07-16 | 2021-10-22 | 中国工商银行股份有限公司 | Fraud risk prediction method and device based on GCN |
| CN113554515A (en) * | 2021-06-26 | 2021-10-26 | 陈思佳 | Internet financial control method, system, device and medium |
| CN113628057A (en) * | 2021-08-16 | 2021-11-09 | 湖南工商大学 | Patient identification method with cheat-save behavior based on improved generation of countermeasure network |
| CN113642672A (en) * | 2021-08-31 | 2021-11-12 | 平安医疗健康管理股份有限公司 | Feature processing method and device of medical insurance data, computer equipment and storage medium |
| CN113657548A (en) * | 2021-08-31 | 2021-11-16 | 平安医疗健康管理股份有限公司 | Medical insurance abnormity detection method and device, computer equipment and storage medium |
| CN114822691A (en) * | 2022-04-13 | 2022-07-29 | 上海人工智能创新中心 | Clinical event prediction device based on graph convolution neural network |
| CN115456805A (en) * | 2022-11-14 | 2022-12-09 | 华信咨询设计研究院有限公司 | Medical insurance anti-fraud method and system based on machine learning |
| CN116150341A (en) * | 2023-04-23 | 2023-05-23 | 之江实验室 | Method for detecting claim event, computer device and storage medium |
| CN116755397A (en) * | 2023-05-26 | 2023-09-15 | 北京航空航天大学 | Multi-machine collaborative task scheduling method based on graph convolution strategy gradient |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20180284735A1 (en) * | 2016-05-09 | 2018-10-04 | StrongForce IoT Portfolio 2016, LLC | Methods and systems for industrial internet of things data collection in a network sensitive upstream oil and gas environment |
-
2018
- 2018-12-25 CN CN201811589288.5A patent/CN109636061B/en active Active
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20180284735A1 (en) * | 2016-05-09 | 2018-10-04 | StrongForce IoT Portfolio 2016, LLC | Methods and systems for industrial internet of things data collection in a network sensitive upstream oil and gas environment |
Cited By (52)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110163261A (en) * | 2019-04-28 | 2019-08-23 | 平安科技（深圳）有限公司 | Unbalanced data disaggregated model training method, device, equipment and storage medium |
| WO2020220544A1 (en) * | 2019-04-28 | 2020-11-05 | 平安科技（深圳）有限公司 | Unbalanced data classification model training method and apparatus, and device and storage medium |
| CN110188176A (en) * | 2019-04-30 | 2019-08-30 | 深圳大学 | Deep learning neural network and training, prediction technique, system, equipment, medium |
| CN110188176B (en) * | 2019-04-30 | 2022-12-23 | 深圳大学 | Deep learning neural network, and training and predicting method, system, device and medium |
| CN110210330A (en) * | 2019-05-13 | 2019-09-06 | 清华大学 | Electromagnetic signal recognition methods and device based on Tacit Knowledge structure figures convolutional network |
| CN110263227A (en) * | 2019-05-15 | 2019-09-20 | 阿里巴巴集团控股有限公司 | Clique based on figure neural network finds method and system |
| CN110245787A (en) * | 2019-05-24 | 2019-09-17 | 阿里巴巴集团控股有限公司 | A kind of target group's prediction technique, device and equipment |
| CN110245787B (en) * | 2019-05-24 | 2023-11-17 | 创新先进技术有限公司 | Target group prediction method, device and equipment |
| CN110490132B (en) * | 2019-08-16 | 2020-09-29 | 京东城市(北京)数字科技有限公司 | Data processing method and device |
| CN110490132A (en) * | 2019-08-16 | 2019-11-22 | 京东城市(北京)数字科技有限公司 | Data processing method and device |
| CN110688536A (en) * | 2019-09-25 | 2020-01-14 | 中国建设银行股份有限公司 | Label prediction method, device, equipment and storage medium |
| CN110781406B (en) * | 2019-10-14 | 2021-07-06 | 西安交通大学 | Social network user multi-attribute inference method based on variational automatic encoder |
| CN110781406A (en) * | 2019-10-14 | 2020-02-11 | 西安交通大学 | Social network user multi-attribute inference method based on variational automatic encoder |
| CN110796497A (en) * | 2019-10-31 | 2020-02-14 | 支付宝(杭州)信息技术有限公司 | Method and device for detecting abnormal operation behaviors |
| CN110827159A (en) * | 2019-11-11 | 2020-02-21 | 上海交通大学 | Financial medical insurance fraud early warning method, device and terminal based on relational graph |
| CN110827159B (en) * | 2019-11-11 | 2023-11-03 | 上海交通大学 | Financial medical insurance fraud early warning method, device and terminal based on relation diagram |
| CN111126255A (en) * | 2019-12-23 | 2020-05-08 | 华中科技大学 | Numerical control machine tool cutter wear value prediction method based on deep learning regression algorithm |
| CN111427926A (en) * | 2020-03-23 | 2020-07-17 | 平安医疗健康管理股份有限公司 | Abnormal medical insurance group identification method and device, computer equipment and storage medium |
| CN111427926B (en) * | 2020-03-23 | 2023-02-03 | 平安医疗健康管理股份有限公司 | Abnormal medical insurance group identification method and device, computer equipment and storage medium |
| CN111488924A (en) * | 2020-04-07 | 2020-08-04 | 中国农业科学院农业信息研究所 | Multivariate time sequence data clustering method |
| CN111488924B (en) * | 2020-04-07 | 2024-04-26 | 中国农业科学院农业信息研究所 | Multivariable time sequence data clustering method |
| CN111612636A (en) * | 2020-04-29 | 2020-09-01 | 山东大学 | Abnormal medical insurance data detection system and method based on dual clustering algorithm |
| CN111581962B (en) * | 2020-05-14 | 2023-02-21 | 福州大学 | Text representation method based on subject word vector and hybrid neural network |
| CN111581962A (en) * | 2020-05-14 | 2020-08-25 | 福州大学 | Text representation method based on subject word vector and hybrid neural network |
| CN111785014A (en) * | 2020-05-26 | 2020-10-16 | 浙江工业大学 | Road network traffic data restoration method based on DTW-RGCN |
| CN111899114A (en) * | 2020-06-23 | 2020-11-06 | 山东大学 | Doctor-seeking fraud detection method and system based on multi-view double clustering |
| CN111798324B (en) * | 2020-07-16 | 2024-04-02 | 浙江工业大学 | Medical insurance fraud discovery method based on dynamic medical practice alignment |
| CN111798324A (en) * | 2020-07-16 | 2020-10-20 | 浙江工业大学 | Medical insurance fraud discovery method based on dynamic hospitalizing behavior alignment |
| CN112200684B (en) * | 2020-09-15 | 2024-05-07 | 深圳大学 | Method, system and storage medium for detecting medical insurance fraud |
| WO2022057057A1 (en) * | 2020-09-15 | 2022-03-24 | 深圳大学 | Method for detecting medicare fraud, and system and storage medium |
| CN112200684A (en) * | 2020-09-15 | 2021-01-08 | 深圳大学 | Method, system and storage medium for detecting medical insurance fraud |
| CN112331230A (en) * | 2020-11-17 | 2021-02-05 | 平安科技（深圳）有限公司 | Method and device for identifying fraudulent conduct, computer equipment and storage medium |
| CN112435133A (en) * | 2020-11-18 | 2021-03-02 | 厦门理工学院 | Medical insurance combined fraud detection method, device and equipment based on graph analysis |
| CN112801804A (en) * | 2021-01-15 | 2021-05-14 | 浙江大学山东工业技术研究院 | Method for detecting map neural network of abnormal medical insurance data |
| CN112991075A (en) * | 2021-02-04 | 2021-06-18 | 浙江大学山东工业技术研究院 | Package type medicine purchasing abnormity detection method based on FP-growth and graph network |
| CN113179250A (en) * | 2021-03-26 | 2021-07-27 | 北京六方云信息技术有限公司 | Web unknown threat detection method and system |
| CN113179250B (en) * | 2021-03-26 | 2022-05-17 | 北京六方云信息技术有限公司 | Method and system for detecting unknown web threats |
| CN113139586A (en) * | 2021-03-31 | 2021-07-20 | 同济大学 | Model training method, device abnormality diagnosis method, electronic device, and medium |
| CN113139586B (en) * | 2021-03-31 | 2022-09-23 | 同济大学 | Model training method, device abnormality diagnosis method, electronic device, and medium |
| CN113554515A (en) * | 2021-06-26 | 2021-10-26 | 陈思佳 | Internet financial control method, system, device and medium |
| CN113538126A (en) * | 2021-07-16 | 2021-10-22 | 中国工商银行股份有限公司 | Fraud risk prediction method and device based on GCN |
| CN113628057B (en) * | 2021-08-16 | 2022-04-19 | 湖南工商大学 | Patient identification method with cheat-save behavior based on improved generation of countermeasure network |
| CN113628057A (en) * | 2021-08-16 | 2021-11-09 | 湖南工商大学 | Patient identification method with cheat-save behavior based on improved generation of countermeasure network |
| CN113642672B (en) * | 2021-08-31 | 2024-05-14 | 平安医疗健康管理股份有限公司 | Feature processing method and device of medical insurance data, computer equipment and storage medium |
| CN113657548A (en) * | 2021-08-31 | 2021-11-16 | 平安医疗健康管理股份有限公司 | Medical insurance abnormity detection method and device, computer equipment and storage medium |
| CN113642672A (en) * | 2021-08-31 | 2021-11-12 | 平安医疗健康管理股份有限公司 | Feature processing method and device of medical insurance data, computer equipment and storage medium |
| CN114822691A (en) * | 2022-04-13 | 2022-07-29 | 上海人工智能创新中心 | Clinical event prediction device based on graph convolution neural network |
| CN114822691B (en) * | 2022-04-13 | 2024-06-11 | 上海人工智能创新中心 | Clinical event prediction device based on graph convolution neural network |
| CN115456805A (en) * | 2022-11-14 | 2022-12-09 | 华信咨询设计研究院有限公司 | Medical insurance anti-fraud method and system based on machine learning |
| CN116150341A (en) * | 2023-04-23 | 2023-05-23 | 之江实验室 | Method for detecting claim event, computer device and storage medium |
| CN116755397B (en) * | 2023-05-26 | 2024-01-23 | 北京航空航天大学 | Multi-machine collaborative task scheduling method based on graph convolution strategy gradient |
| CN116755397A (en) * | 2023-05-26 | 2023-09-15 | 北京航空航天大学 | Multi-machine collaborative task scheduling method based on graph convolution strategy gradient |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109636061B (en) | 2023-04-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109636061A (en) | Training method, device, equipment and the storage medium of medical insurance Fraud Prediction network | |
| ȚĂRANU | Data mining in healthcare: decision making and precision. | |
| CN111242948B (en) | Image processing method, image processing device, model training method, model training device, image processing equipment and storage medium | |
| Abugabah et al. | Data mining techniques for analyzing healthcare conditions of urban space-person lung using meta-heuristic optimized neural networks | |
| CN106529110A (en) | Classification method and equipment of user data | |
| CN109585020A (en) | A kind of model that disease risks are predicted with convolutional neural networks | |
| Dobay et al. | Potential use of deep learning techniques for postmortem imaging | |
| Wankhade et al. | A novel hybrid deep learning method for early detection of lung cancer using neural networks | |
| DE112021004559T5 (en) | SYSTEM FOR ROBUST PREDICTION OF ERGONOMIC TIME SERIES IN DIALYSIS PATIENT RECORDS | |
| CN116741411A (en) | Intelligent health science popularization recommendation method and system based on medical big data analysis | |
| CN107145715A (en) | A kind of clinical medical intelligent discriminating gear based on election algorithm | |
| CN109871869A (en) | A kind of Lung neoplasm classification method and its device | |
| Mahbub et al. | Detect bangladeshi mango leaf diseases using lightweight convolutional neural network | |
| Chao | Estimating project overheads rate in bidding: DSS approach using neural networks | |
| Wei et al. | (Retracted) Image analysis and pattern recognition method of three-dimensional process in physical education teaching based on big data | |
| WO2022057057A1 (en) | Method for detecting medicare fraud, and system and storage medium | |
| Cao et al. | 3D convolutional neural networks fusion model for lung nodule detection onclinical CT scans | |
| CN115762721A (en) | Medical image quality control method and system based on computer vision technology | |
| Dedhia et al. | SCouT: synthetic counterfactuals via spatiotemporal transformers for actionable healthcare | |
| Yamin et al. | Leveraging Retinal Fundus Images with Deep Learning for Diabetic Retinopathy Grading and Classification. | |
| Dey et al. | Deep Greedy Network: a tool for medical diagnosis on exiguous dataset of COVID-19 | |
| An et al. | A reinforcement learning guided adaptive cost-sensitive feature acquisition method | |
| Wadhonkar et al. | Artificial neural network approach for classification of heart disease dataset | |
| CN111986034A (en) | Medical insurance group fraud monitoring method, system and storage medium | |
| Dinara et al. | Fine-tuning the hyperparameters of pre-trained models for solving multiclass classification problems |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |