JP7257452B2 - PET INSURANCE PAYMENT ASSESSMENT SYSTEM, METHOD AND PROGRAM - Google Patents

Description

Application of Article 30, Paragraph 2 of the Patent Law Posting date on the website April 23, 2021 Website address https://deliveru. jp/

TECHNICAL FIELD The present invention relates to a technology related to payment assessment of pet insurance money.

In recent years, more and more households have pets. Along with this, pet insurance for covering medical expenses for pets is also rapidly spreading (see, for example, Patent Documents 1 to 3). The number of pet insurance claims is also on the rise.

By the way, the items of medical treatment performed on people are basically in accordance with the medical fees announced by the Ministry of Health, Labor and Welfare. For this reason, the items described in medical specifications received by patients from medical institutions are generally standardized at each medical institution. Therefore, it can be said that the work of confirming the medical treatment items performed in the work of assessing the payment of insurance benefits relating to medical treatment of a person can be performed relatively without hindrance. In addition, in the case of insurance for people such as life insurance, since the payment amount is fixed in advance, it can be said that the assessment work is easy in this respect as well.

On the other hand, the items of medical care performed on pets vary widely depending on the type of pet and the medical institution. For this reason, the items to be described in the medical statement received by the pet breeder from the medical institution vary depending on the type of pet and the medical institution. Therefore, it is more difficult to check the medical treatment items in the insurance payment assessment work for pet medical treatment than in the insurance payment assessment for human medical treatment. In addition, it is not easy to automate assessment due to the wide variety of items described in medical specifications. Furthermore, since pet insurance is non-life insurance, and the insurance money is calculated by multiplying the actual medical expenses by the compensation rate, it is not easy to calculate the insurance money.

Japanese Patent No. 6735228 JP-A-2006-40158

SUMMARY OF THE INVENTION The present invention has been made in view of the above problems, and an object to be solved is to make it possible to accurately and quickly assess payment of pet insurance claims.

In order to solve the above-mentioned problems, in the present invention, an assessor displays a screen in which a correction reception display for accepting individual correction of the judgment result is associated with each medical treatment item name for the judgment result obtained using the learning model of machine learning. When the correction acceptance display on the screen for assessors accepts individual corrections, the content of the corrections is reflected in the learning model.

More specifically, the present invention is a pet insurance payment assessment system for assessing payment of pet insurance, wherein the name of the medical treatment item described in the medical specification and the A medical care detail information acquisition unit that acquires the amount of money, information data of the assessment target including the medical care item name described in the medical care specification, and payment assessment as to whether or not the assessment target is subject to payment for each medical care item name. A machine learning unit that performs machine learning by associating the assessment results at the time of the assessment, and a medical care item name acquired from the medical details to be newly assessed by the medical record information acquisition unit, is input to the learning model obtained by the machine learning unit, an assessment processing unit for outputting a judgment result as to whether or not the medical treatment item name is subject to payment on a screen for an assessor, and the assessment processing unit acquires from the medical details to be newly assessed by the medical details information acquiring unit. A screen for the assessor, which includes the name of the medical care item and the amount of money, the judgment result shown in association with each medical care item name, and a correction acceptance display for accepting individual correction of the judgment result corresponding to each medical care item name. , and the machine learning unit reflects the content of the correction to the learning model when the correction reception display on the assessor screen receives the individual correction.

Here, a pet is an animal raised by a person, and examples thereof include dogs, cats, birds, and other various animals. A medical institution that provides medical care for pets is an institution that can provide medical care for pets. In addition, pet insurance is money paid to compensate for all or part of the expenses required for medical treatment of pets. Items substantially equivalent to money are also included.

With the above pet insurance payment assessment system, even for pet insurance, which has a wide variety of medical care items compared to human insurance, assessments using machine learning models can significantly reduce the processing time involved in assessments. and a rapid assessment can be carried out. In addition, since the assessor can immediately correct the assessment result using the learning model of machine learning by displaying the correction acceptance display on the assessor screen, the assessor can correct errors in the assessment using the learning model. Furthermore, since the corrections are reflected in the learning model, it is possible to automatically improve the accuracy of the assessment using the learning model. Therefore, it is possible to accurately and quickly assess the payment of pet insurance money.

Note that the medical care statement information acquisition unit may read the name of the medical care item and the amount of money described in the medical care statement from an image of the medical care statement issued by a medical institution that treats pets. According to this, since the information described in the medical bill is automatically converted into data, it is possible to assess pet insurance payments more accurately and quickly.

Also, the assessment processing unit may output an assessor screen that further includes an image of the medical bill. According to this, the assessor can easily check the medical specification, so that mistakes, omissions, etc. can be easily found. Therefore, it is possible to assess payment of pet insurance money more accurately and quickly.

In addition, the assessment processing unit may include a non-reflecting operation acceptance display that does not reflect the content of the correction in the learning model on the screen for the assessor. According to this, it is also possible not to reflect the correction content in the learning model. Therefore, for example, if the assessor is inexperienced, the knowledge of the inexperienced assessor can be reflected in the learning model by operating the non-reflecting operation acceptance display so that the correction contents are not reflected in the learning model. can be prevented.

The present invention can also be viewed from a method aspect. The present invention is, for example, a pet insurance payment assessment method for assessing payment of pet insurance, wherein the names of medical care items and amounts described in the medical care details are obtained from the medical care details issued by a medical institution that treats pets. When the process of obtaining medical care details information to be acquired, the data of the information to be assessed including the names of medical care items described in the medical care details, and the payment assessment of whether or not each medical care item name is subject to payment for the items to be assessed The machine learning process for machine learning by associating with the assessment results of , and the medical care item name acquired from the new medical care details to be assessed by the medical care details information acquisition process, is input to the learning model obtained in the machine learning process, and the medical care an assessment processing step of outputting a determination result as to whether or not the item name is subject to payment to an assessor screen; A screen for an assessor is a screen containing the medical care item name and amount, the judgment result shown in association with each medical care item name, and a correction reception display for accepting individual correction of the judgment result corresponding to each medical care item name. In the machine learning process, when the correction acceptance display on the assessor screen accepts the individual correction, the correction content may be reflected in the learning model.

Moreover, the present invention can also be understood from the aspect of a program. The present invention provides, for example, a pet insurance payment assessment program for assessing payment of pet insurance claims, in which a computer obtains a medical care item name described in a medical care specification from a medical care specification issued by a medical institution that provides medical care for pets. and a medical bill information acquisition step for obtaining medical bills and amounts, data of information to be assessed including the names of medical treatment items described in the medical bills, and payment assessment as to whether or not each medical item to be assessed is subject to payment. In the machine learning process, machine learning is performed by associating the assessment results with the assessment results obtained at the time of assessment, and the medical care item names acquired from the medical details to be newly assessed in the medical care details information acquisition process are input to the learning model obtained in the machine learning process. , and an assessment processing step of outputting a determination result as to whether or not the medical care item name is subject to payment to an assessor screen. The assessor displays a screen containing the medical care item name and amount obtained from the In the machine learning process, when the correction acceptance display on the assessor screen accepts an individual correction, the contents of the correction may be reflected in the learning model.

ADVANTAGE OF THE INVENTION According to this invention, payment assessment of pet insurance money can be implemented correctly and rapidly.

FIG. 1 is a diagram showing an overview of the entire system for payment of pet insurance claims. FIG. 2 is an example of a functional block diagram implemented in the server. FIG. 3 is a diagram showing an example of an overall process related to insurance appraisal. FIG. 4 is a diagram showing a flow chart of processing that is mainly executed in a terminal when reception processing is performed by a receptionist. FIG. 5 is a diagram showing an example of a bill and statement issued by an animal hospital. FIG. 6 is a diagram showing an example of billing information stored in the billing information database. FIG. 7 is a diagram showing a flowchart of processing that is mainly executed in a terminal when an assessor performs assessment work. FIG. 8 is a diagram showing an example of customer information stored in the customer information database. FIG. 9 is a diagram showing an example of payment information stored in the payment information database. FIG. 10 is an example of a screen showing assessment results.

Embodiments will be described below. The embodiments shown below are merely examples, and do not limit the technical scope of the present disclosure to the following aspects.

<System configuration>
FIG. 1 is a diagram showing an overview of the entire system for payment of pet insurance claims. In pet insurance, an insurance company 3 compensates a user 1 who has a pet for the medical expenses required for treatment of the pet at an animal hospital 2. - 特許庁The user 1 is an insured person who can receive the insurance money, and is also a policyholder when the user 1 himself/herself contracts pet insurance. A user 1 subscribing to the pet insurance of an insurance company 3 pays the expenses at the animal hospital 2 when the pet is treated at the animal hospital 2. - 特許庁Then, the user 1 receives the bill/specification 6 . In the bill/statement 6, in addition to the expenses paid to the animal hospital 2, details such as medical treatment items are written.
This bill/statement 6 is prepared by the animal hospital 2 independently. If the user 1 who received the bill/statement 6 wishes to use insurance, he/she can request insurance coverage by sending the bill/statement 6 and other billing documents to the insurance company 3 by mail or electronic document. I do. When the user 1 requests insurance coverage, the insurance company 3 confirms and assesses the claim documents, and pays the insurance money.

An insurance company 3 has a receptionist 4 and an assessor 5 who check and assess claim documents. The receptionist 4 is mainly responsible for receiving documents mailed from the user 1 . In addition, the assessor 5 manages assessment operations based on the information in the documents received by the receptionist 4 . The receptionist 4 may be an employee of the insurance company 3 or may be an employee of an agency business entrusted by the insurance company 3 .

The insurance company 3 has a server 10, a terminal 20, and a terminal 30 that can communicate with each other via the network N. The server 10 is a core system server that handles insurance information provided by the insurance company 3 . The terminal 20 is a computer on which the receptionist 4 can perform reception work, and is equipped with a scanner for scanning documents and acquiring image data, a display capable of displaying a screen, and the like. In addition, the terminal 30 is a computer with which the person in charge of assessment 5 can carry out assessment work. Each of the server 10, the terminal 20, and the terminal 30 includes a CPU (Central Processing Unit), memory, storage, communication interface, input/output device, etc. By executing a computer program developed in the memory, Implements various functional units.

FIG. 2 is an example of a functional block diagram implemented in the server 10. As shown in FIG. When the computer program is executed by the server 10, an information acquisition unit 11, a machine learning unit 12, and an assessment processing unit 14 are implemented as shown in FIG. The information acquisition unit 11 (an example of the “medical record information acquisition unit” referred to in the present application) acquires information from the customer information database 15 and the billing information database 16 stored in the storage of the server 10, and the machine learning unit 12 and the function of sending to the assessment processing unit 14. The machine learning unit 12 has a function of generating a learning model 13 based on information sent from the information acquisition unit 11 . The assessment processing unit 14 manages assessment processing such as AI (Artificial Intelligence) assessment using the learning model 13 based on the information sent from the information acquisition unit 11 . The assessment result is stored in the payment information database 17. FIG. The customer information database 15, the billing information database 16, and the payment information database 17 are types of databases that constitute a core system used by the insurance company 3. FIG.

<Overview of assessment process>
FIG. 3 is a diagram showing an example of an overall process related to insurance appraisal. In FIG. 3, an outline of processing in an assessment system implemented by the server 10 in cooperation with the terminal 20 and the terminal 30 is shown as an "embodiment", and an outline of a conventional process in which assessment is performed manually is shown as a "comparative example". In FIG. 3, the length of each step in the vertical direction of the page expresses the length of the elapsed time required for processing.

As can be seen from the outline of processing shown on the left side of FIG. 3, in this embodiment, first, the billing documents sent by the user 1 to the insurance company 3 are accepted (S1). Then, the receptionist 4 reads the billing document sent by the user 1 to the insurance company 3 with the scanner of the terminal 20 and converts the billing document into an image (S2). When the terminal 20 acquires the image of the billing document with a scanner, it performs OCR (Optical Character Recognition) processing, and automatically converts the character information written on the billing document into data using AI or the like (S3). . Since the data of character information automatically converted into data by the terminal 20 may contain reading errors, etc., the receptionist 4 operates the terminal 20 and compares the data with the image of the billing document. is corrected (S4).

After the receptionist 4 completes the data conversion of the billing documents at the terminal 20 and the information of the billing documents is stored in the billing information database 16, automatic payment assessment using AI is performed in the server 10 (S5). . In this payment assessment, the assessment processing section 14 uses the learning model 13 to assess based on the information in the customer information database 15 and the billing information database 16 acquired by the information acquisition section 11 . The assessment result of the payment assessment using AI is stored in the payment information database 17 of the core system (S6). The assessment result data stored in the payment information database 17 is checked and corrected by the assessor 5 (S7). When the processing within the insurance company 3 is completed, payment is made to the user 1, and a series of processing is completed (S8).

On the other hand, as can be seen from the outline of processing shown on the right side of FIG. 3, in the comparative example, receipt of billing documents (Sh1) and imaging (Sh2) are performed in the same manner as in S1 to S2 described above. However, in the comparative example, the receptionist 4 visually reads all the character information written on the invoice and manually inputs it into the terminal 20 (Sh3). Then, the input data is stored in the billing information database 16 (Sh4). The assessor 5 refers to the information stored in the billing information database 16 and performs manual payment assessment without using AI (Sh5). When the processing within the insurance company 3 is completed, payment is made to the user 1, and a series of processing is completed (Sh6).

As can be seen by comparing the processing outline of the embodiment shown on the left side of the page in FIG. 3 and the processing outline of the comparative example shown on the right side of the page in FIG. is expected to be significantly reduced. For example, in the embodiment, document information input, which was performed manually in the comparative example, is performed almost automatically, so the number of man-hours related to information input can be greatly reduced.

<Processing flow>
Details of processing in each step of the present embodiment will be described below.

FIG. 4 is a diagram showing a flowchart of processing mainly executed by the terminal 20 when the receptionist 4 performs the reception processing. The details of the processing executed during the reception processing will be described below with reference to the flowchart of FIG. 4 .

The receptionist 4 first sets the billing documents on the scanner when performing billing document receiving processing. When the receptionist 4 sets the billing document on the scanner and performs an operation to start reading, the terminal 20 activates the scanner and acquires the image data of the billing document (S101). The scanner may be a device that reads a document in a fixed state, or a device that reads the document while it is being moved by a document feeder. The scanner incorporates a light source and a solid-state imaging device. Then, reflected light when the original is irradiated from the light source is imaged on the solid-state imaging device, and a raster image reading signal is obtained from the solid-state imaging device.

After acquiring the image data in step S101, the terminal 20 reads character information by OCR processing (S102). In OCR processing, various techniques are used in combination in order to improve character recognition accuracy. For example, image processing for removing fine noise dots may be used as preprocessing. Alternatively, a technique called "word matching" may be used in which predicted candidate characters are compared with a predefined word list to find the closest character string.

Further, the billing documents read in this embodiment include various bills and specifications 6 issued by each animal hospital 2 according to its own style. FIG. 5 is a diagram showing an example of a bill/statement 6 issued by the animal hospital 2. As shown in FIG. As shown in FIG. 5, a bill/statement 6 issued by a specific animal hospital 2 has a table in which columns of "medical care item", "unit price", "quantity", and "price" are arranged in order from the left side. is provided. The total settlement amount is described on the upper side of the table. However, in the invoice/description 6 issued by another animal hospital 2, there may be a table in which the unit price, quantity, and amount are listed in a different order, or the total settlement amount may be provided in a different position. There is In addition, unlike medical care for humans, medical care for pets is not performed according to the medical care items in accordance with the medical fees set by the Ministry of Health, Labor and Welfare. also vary widely depending on the veterinary hospital 2.

Therefore, in the OCR processing of the present embodiment, in order to identify the character string area such as the issuer of the invoice/description 6 and the amount of money, for the billing document to be read, the format of the invoice/description 6 is specified. In addition to known methods, AI is also utilized in layout analysis for recognizing structures. Utilization of AI may be realized by the terminal 20 alone or in cooperation with the server 10 . In OCR processing, a character area extracted by layout analysis is extracted, character extraction analysis processing is performed to decompose the characters in the character area one by one, and recognition processing is performed to recognize what character the image data is. In recognition processing, for example, normalization processing that enlarges/reduces character images to a certain size, characteristics of character images (vertical, horizontal, diagonal lines, length of each line, positional relationship between lines, etc.) Extraction processing, matching processing for searching for character codes that match the feature values of character images, and automatic correction processing using dictionaries (word dictionaries, etc.) for each character code and/or character string extracted by the matching processing. processing is included.

After automatically reading the character information in step S102, the terminal 20 accepts correction by the receptionist 4 (S103). The terminal 20 displays, for example, the image of the billing document obtained in the process of step S101 and the text of the character information read in the process of step S102 on the same screen so that the receptionist 4 can easily compare the two. The terminal 20 provides a GUI (Graphical User Interface) for the receptionist 4 to manually change the text by operating the keyboard and mouse (pointing device).

After the terminal 20 starts receiving corrections in step S103, when corrections are made by the receptionist 4 (S104), the data developed in the memory of the terminal 20 is updated (S105). . Further, when the correction or confirmation by the receptionist 4 is completed, the terminal 20 transmits the data to the server 10 as billing information (S106). In the server 10 , the billing information transmitted from the terminal 20 is stored in the billing information database 16 .

FIG. 6 is a diagram showing an example of billing information stored in the billing information database 16. As shown in FIG. By performing the processes of steps S101 to S106, billing information data as shown in FIG. 6 is accumulated in the billing information database 16. FIG. That is, by performing the processing of steps S101 to S106, the bill number and medical treatment items of user 1 described in the billing document correspond to the bill number issued for each claim for insurance coverage from user 1. , unit price, quantity, amount, etc. are accumulated. As can be seen from FIG. 6, the medical treatment items, quantities, and amounts described in the bill/specification 6 of the billing documents sent from the user 1 to the insurance company 3 vary from bill to bill. In addition, there are various kinds of pets covered by the insurance that the user 1 uses. Therefore, in assessing insurance claims for pets, unlike assessing insurance claims for human medical care, even if the names of medical treatment items differ for each claim, the contents are the same, and It is necessary to comprehensively take into consideration the type of pet, etc., for cases with different contents even if the name of the medical treatment item is the same. Therefore, in the present embodiment, the insurance payment assessment work is performed as follows.

FIG. 7 is a diagram showing a flowchart of processing mainly executed by the terminal 30 during the assessment work of the assessor 5. As shown in FIG. The details of the processing executed during the assessment work will be described below with reference to the flowchart of FIG.

In the server 10, while the person in charge of assessment 5 is operating the terminal 30 in order to carry out insurance application assessment work, or at an appropriate timing before starting operation of the terminal 30 (for example, late at night when batch processing is performed). etc.), assessment using AI is automatically performed. That is, when the server 10 detects that billing information has been newly stored in the billing information database 16, the server 10 causes the information obtaining unit 11 to execute data reception processing for obtaining data of the billing information from the billing information database 16 (S201). ). In addition, the server 10 causes the information acquiring section 11 to acquire the customer information corresponding to the billing information from the customer information database 15 (S202). Then, the server 10 causes the assessment processing unit 14 to perform AI assessment processing using the learning model 13 (S203). The result of assessment processing by the assessment processing unit 14 is stored in the payment information database 17 .

When performing the assessment work, the assessor 5 operates the terminal 30 to display a screen for assessment work. The screen displayed on the terminal 30 may be a screen generated by a computer program installed on the terminal 30, or may be generated by the server 10 by executing a computer program installed on the server 10 and displayed on the network. It may be a screen displayed on the display of the terminal 30 via N. By this operation, the terminal 30 displays the assessment result information stored in the payment information database 17 in the process of step S203 on the display (S204).

FIG. 8 is a diagram showing an example of customer information stored in the customer information database 15. As shown in FIG. The customer information database 15 stores customer information data as shown in FIG. The customer information stored in the customer information database 15 is information accumulated when the pet insurance contract used by the user 1 is exchanged with the insurance company 3 . As can be seen from FIG. 8, the customer information includes, for example, the number of the policy as proof of the insurance contract, the name of the user 1 such as the policyholder and the insured, the type and breed of the pet covered by the contract, the animal Includes names, etc. The customer information also includes the address of the user 1, information on the account for payment of insurance premiums, information on the content of insurance coverage, and various other information.

FIG. 9 is a diagram showing an example of payment information stored in the payment information database 17. As shown in FIG. The payment information stored in the payment information database 17 is composed of customer information in the customer information database 15, billing information in the billing information database 16, and assessment result information obtained in step S203. That is, as can be seen from FIG. 9, the payment information stored in the payment information database 17 includes policy numbers, medical treatment items, unit prices, and quantities, each corresponding to a claim number issued for each insurance application claim. , the amount, and information on whether or not it is covered by insurance. At the stage when the process of step S203 is executed, the information on whether or not it is insured stored in the payment information database 17 is only a temporary assessment result using AI. It will be fixed.

Since such customer information is stored in the customer information database 15 and such payment information is stored in the payment information database 17, the assessment result screen displayed on the display of the terminal 30 in step S204 is: For example, the following information is displayed.

FIG. 10 is an example of a screen showing assessment results. When this step S204 is executed, the display of the terminal 30 displays, for example, a screen as shown in FIG. That is, on the display of the terminal 30, customer information such as the policy number and the name of the user 1, the image of the invoice/description 6 whose image data was acquired in the processing of step S101, the image read by the OCR processing and corrected as appropriate, etc. Claimed information, information on whether or not the product is covered by insurance, which is the result of assessment using AI, and automatically calculated amount of insurance money are displayed. For this reason, the assessor 5 confirms information such as the type of pet covered by insurance and the image of the bill/statement 6 to judge and confirm whether or not each medical treatment item is covered by insurance. It is possible to easily verify the validity of assessment results using AI. In addition, since the calculation of the insurance amount is also automatically performed, the burden on the assessor 5 can be reduced as compared with the case of manual calculation.

As can be seen from FIG. 10, the assessment result screen displayed on the display of the terminal 30 includes a "correction" button (referred to as "correction acceptance display" in the present application) for correcting the assessment result using AI. is an example) is prepared for each medical care item. Therefore, when the assessor 5 judges that the assessment result using AI is not valid, the assessor 5 can correct the assessment result as to whether or not it is covered by insurance. Correction work can be easily performed by operating the keyboard or pointing device of the terminal 30 .

In addition, as can be seen from FIG. 10, on the assessment result screen, a check box (an example of the "non-reflection operation acceptance display" referred to in the present application) is prepared for the item name "AI learning". The person in charge of assessment 5 can select whether or not to reflect the corrected content of the assessment result in the learning model 13 and perform automatic re-learning of the AI. Therefore, the server 10 or the terminal 30 determines whether or not the checkbox is checked, in other words, whether or not the correction work performed by the assessor 5 is to be reflected in the learning model (S205). . If the "AI learning" check box is checked, an affirmative determination is made in step S205, and the contents of correction of the assessment result by the assessor 5 are reflected in the learning model 13 (S206). In other words, the content of correction of the assessment result by the assessor 5 is taken into the machine learning unit 12 as learning data of the learning model 13 and used to correct the learning model 13 . If the "AI learning" check box is not checked, a negative determination is made in step S205, and the processing in step S206 is omitted. For example, if the assessor 5 does not have much experience, the knowledge of the inexperienced assessor 5 is reflected in the learning model by unchecking the "AI learning" check box. can be prevented.

When the assessor 5 completes the correction of the assessment result, the content of the assessment result corrected by the assessor 5 is reflected in the payment information stored in the payment information database 17 . Then, the payment information stored in the payment information database 17 is confirmed by confirmation work by the superior of the person in charge of assessment 5 or an appropriate department (S207). The insurance company 3 pays the insurance money to the user 1 when the payment information is confirmed. When the user 1 receives the insurance money from the insurance company 3, a series of processing is completed.

In the screen shown in FIG. 10, if the image of the invoice/description 6 is displayed in a place where the person in charge of assessment 5 can easily see it, it is possible to It is easy to find leaks and the like. However, the image of the invoice/description 6 may be omitted from the screen of FIG. 10, or may be displayed when the call button is pressed.

<Details of processing using AI>
The content of assessment processing using AI will be described in detail below.

Pet insurance payment assessments basically follow the steps below. That is, first, taking into consideration the contract information, the injury or disease information, the breed and age of the pet, etc., a comprehensive judgment is made as to whether or not the bill can be paid. Next, each medical treatment item described in the bill/specification 6 is classified as to whether or not it is subject to payment. Next, the total amount of treatment costs for the medical treatment items to be paid is calculated, and the assessment results such as the amount of insurance money to be paid are determined.

However, as described above, there are many types of pets, so there are many factors to consider when assessing payment for pet insurance. For example, even for the same injury or disease, payment decisions differ depending on the breed, age of onset, and cause. In addition, even for the same medical care item, the judgment differs depending on which injury or disease is being treated. Thus, there is a lot of information to consider in payment assessment. In addition, when assessing payment for pet insurance, it is necessary to determine whether payment is possible for each medical treatment item. The bill/statement 6 issued by the veterinary hospital 2 is characterized by many items other than treatment costs (items not covered by insurance, such as prevention and supplements). Therefore, it is necessary to check the medical care items on the bill/specification 6 and determine whether each medical care item is responsible or not. In addition, since veterinary medicine is a private practice, there are a wide variety of forms such as bills and statements 6 issued by the veterinary hospital 2 . Therefore, there are a wide variety of forms and names of medical care items to be used, and the location of information necessary for assessment differs for each bill/specification 6, making patterning difficult. In addition, each animal hospital 2 may have a different display form of consumption tax (exclusive/inclusive) of the consumption tax, or may have its own member discount or the like.

Therefore, in the assessment performed using AI in this embodiment, the problem of "various types of specifications", which is a problem unique to pet insurance, is solved as follows.

That is, as shown in FIG. 5, the invoice/description 6 has ruled lines and the like in addition to characters. In addition, there are various typefaces such as Mincho typeface and Gothic typeface for characters. In the case of characters with thin lines such as the Mincho typeface, each odd structure may be recognized as a different character due to the gaps that exist between the odd structures that make up the character. Also, if there is a ruled line near a character, the ruled line may be erroneously recognized as part of the character. In addition, if the color of the characters and the background color are similar, the characters may be partially missing and recognized. Further, even a slight change in the layout included in the form may change the OCR result. Therefore, when creating an AI learning model to be used in the processing of step S102 described above, images of bills and statements 6 of various variations issued by each animal hospital 2 are read as data for learning. When the bills/descriptions 6 of each animal hospital 2 are read as learning data, a wide variety of bills/descriptions 6 are collected, and the collected bills/descriptions 6 are sorted according to layout patterns and classified into pattern categories. to implement. Then, it is read as learning data, and the layout determination accuracy and character recognition rate are verified. If there is an error in judging the layout, the learning model is corrected and an attempt is made to improve the learning model. In addition, if there is an error in character recognition, correct the character and try to improve the learning model. The accuracy of character recognition can also be improved by marking specific characters (¥ mark or alphanumeric characters) such as policy numbers and medical treatment amounts, for example.

In the present embodiment, the AI created in this way with the learning model is also used in the OCR processing in step S102, thereby automatically reading bills and specifications 6 in a wide variety of formats.

In addition, in the assessment performed using AI in this embodiment, the problem peculiar to pet insurance, ``there are many judgment materials to be considered in the payment assessment,'' is solved as follows.

In other words, when creating the AI learning model used in the process of step S203 described above, the items used in the actual payment assessment by the assessor are clarified, and the judgment materials necessary for the assessment are extracted. Then, in generating the AI learning model, this judgment material is used as a key, and in the feedback of payment assessment results, if the initial judgment result of the assessor and AI is different, which item of the judgment material was different. Automatic learning is enabled by the processing of step S206 and the like.

For example, if "anal gland procedure" was performed for the purpose of treating "anal adenitis", it would be covered by insurance, but if you were examined for the purpose of treating a disease other than "anal adenitis", " If an anal gland procedure was performed, it would not be covered by insurance. Payment assessment requires information such as diagnosis name, date of onset, medical item name, treatment date, veterinary hospital name, contract details, pet individual information, notification details, and the like. The "diagnosis name" is effective, for example, in confirming whether or not the disease falls under an exempted disease. In addition, the "day of onset" is effective in determining, for example, whether the onset falls within the waiting period, the onset before the insurance start period, or the onset outside the insurance period. Also, the "medical care item name" is effective, for example, in determining whether or not the item is subject to payment. Also, the "date of treatment" is effective in determining, for example, whether the treatment is within the contract period or whether the patient has a pre-existing disease. Also, the "animal hospital name" is effective, for example, in determining non-payable items specific to the animal hospital. In addition, "contract information" such as insurance type, special contract, contract period, etc., is effective for determining, for example, identification of insurance type, content of compensation, presence or absence of special contract, pre-existing disease, onset during waiting period, and the like. In addition, "individual information" is effective, for example, in making payment decisions based on breed-specific diseases and age of onset. In addition, the “notified content” such as a specific disease or notified injury is effective in determining whether the claim content does not fall under the treatment costs for an unsecured disease or a pre-existing disease.

Therefore, in the generation of the AI learning model used in the process of step S203, it is based on information such as insurance contract clauses, assessment results of past payment results, and medical treatment items obtained from affiliated animal hospitals. Set the decision rules programmatically. Then, a learning model is generated by actually reading past payment examples, performing an AI assessment, and providing feedback on the assessment results. When past payment examples are read, not only the assessment results but also the memo information created by the assessor is read. Then, it verifies the results of the assessment using the generated learning model, re-learns another payment case with the same conditions for the case of incorrect answers, and tries to correct the learning model. In addition, for special medical care items, program settings are also used as appropriate instead of AI.

In addition, in the present embodiment, in calculating the amount of insurance payouts in the assessment process using AI, the consumption tax display format (tax excluded/tax included) differs for each animal hospital 2, and unique member discounts, etc. In order to be able to calculate an appropriate amount of payment even if there is an invoice/description 6, the information is processed as follows. That is, the consumption tax rate and the discount rate are calculated by comparing the subtotal of the cost of each medical treatment item and the total treatment cost. Regarding discounts, if there is a description that marks the discount (for example, 〇 or ★), it will be recognized that there is a discount, and the program and AI will be implemented so that calculations that take the discount into account are performed. Set training data. In this way, in this embodiment, both the program and the AI are devised to improve the accuracy.

In the present embodiment, by using both the AI learning model and the program setting in this way, in the assessment process in step S203, it is possible to solve the problem peculiar to pet insurance, ``there are many judgment factors to be considered in the payment assessment.'' makes it possible to respond.

<effect>
According to the present embodiment, even in pet insurance, which has a wide variety of medical care items compared to human insurance, the assessment using AI significantly reduces the processing time for assessment, enabling prompt assessment. can be done. In addition, in this embodiment, the GUI that allows the assessor 5 to immediately modify the AI assessment results is used, so the AI assessment can be used as a tool for the primary assessment. That is, according to the present embodiment, since the GUI is based on the premise that the person in charge of assessment 5 makes the final judgment of the assessment, the person in charge of assessment 5 can correct an error in the AI assessment. In addition, according to this embodiment, when the assessor 5 corrects an error in the AI assessment, the learning model of the AI is also improved at the same time. can be automatically improved.

In addition, it takes a considerable amount of money and a period of time to make the hired person acquire practical knowledge and become the assessor 5 . However, according to the present embodiment, it is possible to greatly reduce the processing time related to the assessment, so it is suitable even when the number of personnel for the assessor 5 cannot be secured or when the training of the assessor 5 is not easy. is.

Moreover, even if the person in charge of assessment 5 is an excellent person, individual differences and mistakes are inherent in the work performed by people. However, according to this embodiment, the results of the AI assessment are provided to the assessor 5 as information for the first screening, so individual differences and mistakes of each assessor 5 are suppressed as much as possible. and an accurate assessment can be made.

In addition, information on medical treatment items, which was difficult when manually entering data into documents, can be automatically imported as data in this embodiment. It is possible to grasp changes in trends related to applications for application of insurance, etc., and contribute to the development of new insurance that various users 1 want.

<Modification>
In the above embodiment, the receptionist 4 automatically reads the invoice/description 6 using OCR or the like and converts it into data. A form in which the person 4 reads it visually and manually inputs it to the terminal 20 may be used.

In the above embodiment, the receptionist 4 and the assessor 5 work separately, but the operations performed by the receptionist 4 and the assessor 5 may be performed by the same person. Alternatively, it may be shared by three or more people.

In the above embodiment, the invoice/description 6 and the like are paper documents, but the invoice/description 6 is electronic data sent directly from the veterinary hospital 2 to the insurance company 3; Electronic data indirectly sent from the hospital 2 to the insurance company 3 via the user 1 may be used.

<Computer-readable recording medium>
A program that causes a computer or other machine or device (hereinafter referred to as a computer or the like) to implement any of the functions described above can be recorded in a computer-readable recording medium. By causing a computer or the like to read and execute the program of this recording medium, the function can be provided.

Here, a computer-readable recording medium is a recording medium that stores information such as data and programs by electrical, magnetic, optical, mechanical, or chemical action and can be read by a computer, etc. Say. Examples of such recording media that can be removed from a computer or the like include flexible discs, magneto-optical discs, CD-ROMs, CD-R/Ws, DVDs, Blu-ray discs (Blu-ray is a registered trademark), DAT, and 8mm tapes. , memory cards such as flash memory. In addition, there are a hard disk, a ROM (read only memory), and the like as a recording medium fixed to a computer or the like.

N... Network 1... User 2... Animal hospital 3... Insurance company 4... Receptionist 5... Assessor 6... Invoice and statement 10... Server 11... Information acquisition unit 12 Machine learning unit 13 Learning model 14 Assessment processing unit 15 Customer information database 16 Billing information database 17 Payment information database 20 Terminal 30 Terminal

Claims (6)

A pet insurance payment assessment system for assessing payment of pet insurance,
a medical care statement information acquisition unit that acquires the names of medical care items and amounts described in the medical care statement from the medical care statement issued by a medical institution that treats pets;
A machine that associates the data of information on the subject of assessment, including the names of medical care items described in the medical bills, with the results of the payment assessment of whether or not the subject of the assessment is subject to payment for each name of the medical care item. a machine learning unit for learning,
The medical care item name acquired from the medical care details to be newly assessed by the medical care item information acquisition unit is input to the learning model obtained by the machine learning unit, and the determination result of whether or not the medical care item name is subject to payment. and an assessment processing unit that outputs to the assessor screen,
The assessment processing unit acquires the medical care item name and amount acquired from the medical care details to be newly assessed by the medical care item information acquisition unit, the determination result shown in association with each medical care item name, and each medical care item name. outputting a screen including a correction acceptance display for accepting an individual correction of the corresponding determination result as the screen for the assessor;
When the correction acceptance display on the assessor screen accepts the individual correction, the machine learning unit reflects the correction content in the learning model.
Pet insurance payment assessment system. The medical care statement information acquisition unit reads medical care item names and amounts described in the medical care statement from an image of a medical care statement issued by a medical institution that treats pets.
The pet insurance payment assessment system according to claim 1. The assessment processing unit outputs the assessor screen that further includes an image of the medical statement.
The pet insurance payment assessment system according to claim 2. The assessment processing unit causes the assessor screen to include a non-reflecting operation acceptance display that does not reflect the correction content in the learning model,
The pet insurance payment assessment system according to any one of claims 1 to 3. A pet insurance payment assessment method for assessing payment of pet insurance, comprising:
a medical bill information obtaining step of obtaining medical treatment item names and amounts described in the medical bill from a medical bill issued by a medical institution that treats pets;
A machine that associates the data of information on the subject of assessment, including the names of medical care items described in the medical bills, with the results of the payment assessment of whether or not the subject of the assessment is subject to payment for each name of the medical care item. a machine learning process for learning;
The medical care item name acquired from the medical care details to be newly assessed by the medical care details information acquisition step is input to the learning model obtained in the machine learning step, and the determination result of whether or not the medical care item name is subject to payment. and an assessment processing step of outputting to the assessor screen,
In the assessment processing step, the medical care item name and amount acquired from the medical care details to be newly assessed by the medical care details information acquisition step, the determination result shown in association with each medical care item name, and each medical care item name outputting a screen including a correction acceptance display for accepting an individual correction of the corresponding determination result as the screen for the assessor;
In the machine learning step, when the correction reception display on the assessor screen receives the individual correction, the content of the correction is reflected in the learning model.
Pet insurance payment assessment method. A pet insurance payment assessment program for assessing payment of pet insurance,
to the computer,
a medical bill information obtaining step of obtaining medical treatment item names and amounts described in the medical bill from a medical bill issued by a medical institution that treats pets;
A machine that associates the data of information on the subject of assessment, including the names of medical care items described in the medical bills, with the results of the payment assessment of whether or not the subject of the assessment is subject to payment for each name of the medical care item. a machine learning process for learning;
The medical care item name acquired from the medical care details to be newly assessed by the medical care details information acquisition step is input to the learning model obtained in the machine learning step, and the determination result of whether or not the medical care item name is subject to payment. is output to the screen for the assessor, and
In the assessment processing step, the medical care item name and amount acquired from the medical care details to be newly assessed by the medical care details information acquisition step, the determination result shown in association with each medical care item name, and each medical care item name outputting a screen including a correction acceptance display for accepting an individual correction of the corresponding determination result as the screen for the assessor;
In the machine learning step, when the correction reception display on the assessor screen receives the individual correction, the content of the correction is reflected in the learning model.
Pet insurance payment assessment program.

Images