Method for learning and device for reviewing insurance review claim statement on basis of deep neural network

1. A method for learning an insurance claim statement on a basis of a deep neural network, the method comprising:

receiving, by an insurance claim statement reviewing device, an input of a plurality of insurance review claim statements containing at least one of:

first information including a plurality of general items related to general details of the insurance claim statements;

second information including a plurality of diagnosed patient injury/disease items related to an injury or a disease of a diagnosed patient;

third information including a plurality of treatment history items related to a treatment history; and

fourth information including a plurality of prescription issuance items related to prescription issuance;

classifying, by the insurance claim statement reviewing device, the general items, the diagnosed patient injury/disease items, the treatment history items, or the prescription items included in the first to fourth information in the insurance claim statements into categorical data or numerical data;

converting, by the insurance claim statement reviewing device, the categorical data or the numerical data so as to be input to a deep neural network learning unit;

inputting, by the insurance claim statement reviewing device, the converted data to the deep neural network learning unit in the insurance claim statement reviewing device;

training, by the insurance claim statement reviewing device, the deep neural network using training data of the converted data; and

verifying, by the insurance claim statement reviewing device, the deep neural network using verification data of the converted data,

wherein the verifying comprises:

calculating, by the insurance claim statement reviewing device, possibility of reconciliation of the insurance review claim statements and deriving a first result;

receiving, by the insurance claim statement reviewing device, an actual reconciliation result of the insurance claim statements from an expert review information provider in the insurance claim statement reviewing device;

calculating, by the insurance claim statement reviewing device, an error between the first result and the actual reconciliation result;

updating, by the insurance claim statement reviewing device, a neural network weight based on the calculated error to reduce the error;

suppressing, by the insurance claim statement reviewing device, overfitting of the updated neural network based on the verification data; and

when the calculated error is less than or equal to a predetermined value, terminating, by the insurance claim statement reviewing device, the learning method.

2. The method of claim 1 , wherein, in the converting, when the data constituting the general items, the diagnosed patient injury/disease items, the treatment history items, or the prescription issuance items corresponds to a specific category as the categorical data, a value for identifying the specific category is input to an element mapped to the specific category,

wherein the categorical data for each of the items is configured in an array having as many elements as the number of possible categories,

wherein each of the elements is mapped to the specific category.

3. The method of claim 1 , wherein, when the data constituting the general items, the diagnosed patient injury/disease items, the treatment history items, or the prescription issuance items are specific numerical values as the numerical data, the converting comprises:

normalizing the numerical values considering a maximum, minimum, average, and standard deviation of each of the numerical values.

4. The method of claim 1 , wherein the first information has a one-to-many relationship with the second to fourth information,

wherein the inputting comprises:

inputting, by the insurance claim statement reviewing device, the converted second to fourth information to a recurrent neural network learning unit.

5. The method of claim 1 , wherein the receiving comprises:

dividing, by the insurance claim statement reviewing device, the input of the insurance review claim statements into a predetermined size.