CN118152548A - Medical insurance data tracing method and system based on question-answer type picture text extraction model - Google Patents
Medical insurance data tracing method and system based on question-answer type picture text extraction model Download PDFInfo
- Publication number
- CN118152548A CN118152548A CN202410586664.4A CN202410586664A CN118152548A CN 118152548 A CN118152548 A CN 118152548A CN 202410586664 A CN202410586664 A CN 202410586664A CN 118152548 A CN118152548 A CN 118152548A
- Authority
- CN
- China
- Prior art keywords
- medical insurance
- question
- medical
- tracing
- extraction model
- 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.)
- Pending
Links
- 238000000605 extraction Methods 0.000 title claims abstract description 49
- 238000000034 method Methods 0.000 title claims abstract description 41
- 238000011084 recovery Methods 0.000 claims abstract description 24
- 239000000284 extract Substances 0.000 claims description 13
- 230000008569 process Effects 0.000 claims description 9
- 230000004927 fusion Effects 0.000 claims description 7
- 230000007787 long-term memory Effects 0.000 claims description 6
- 230000006403 short-term memory Effects 0.000 claims description 6
- 238000004422 calculation algorithm Methods 0.000 claims description 4
- 238000005516 engineering process Methods 0.000 claims description 3
- 230000006870 function Effects 0.000 claims description 3
- 239000000463 material Substances 0.000 claims description 3
- 238000012545 processing Methods 0.000 abstract description 5
- 238000012549 training Methods 0.000 description 6
- 238000004364 calculation method Methods 0.000 description 3
- 238000013527 convolutional neural network Methods 0.000 description 3
- 238000007726 management method Methods 0.000 description 3
- 230000007246 mechanism Effects 0.000 description 2
- 238000013468 resource allocation Methods 0.000 description 2
- 206010039203 Road traffic accident Diseases 0.000 description 1
- 238000007792 addition Methods 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 238000007635 classification algorithm Methods 0.000 description 1
- 238000012790 confirmation Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000015654 memory Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000005728 strengthening Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 230000001960 triggered effect Effects 0.000 description 1
Landscapes
- Medical Treatment And Welfare Office Work (AREA)
Abstract
The invention provides a medical insurance data tracing method and a system based on a questioning and answering type picture text extraction model, which are used for triggering tracing events by utilizing a medical insurance fund tracing clue, extracting element information in recovery clues by the questioning and answering type picture text extraction model, automatically calculating the total payment amount of the medical insurance fund according to clues, obtaining recovery amount according to medical fee reimbursement judging amount and patient self-payment amount, efficiently helping medical insurance staff to develop the tracing work, and obtaining recovery amount based on tracing, having detailed records and related evidence, ensuring validity and rationality of the tracing, greatly reducing manual participation, enabling a user to trigger the tracing events and obtaining the required tracing amount only by carrying out simple operation according to requirements, and rapidly finishing the medical insurance reimbursement of the complaint without carrying out complicated file sending, file processing, information finishing and other works, thereby improving the efficiency and accuracy of the medical insurance reimbursement.
Description
Technical Field
The invention belongs to the technical field of computers, and particularly relates to a medical insurance data tracing method and system based on a question-answer type picture text extraction model.
Background
Medical insurance compensation is an important link in a medical insurance system, and not only relates to the sustainability of medical insurance funds and the rights and interests of paramedics, but also is the embodiment of social law treatment and moral regulations. Through an effective compensation mechanism, the healthy operation of the whole medical care system can be promoted.
In the actual management of the medical insurance fund, medical insurance staff is required to sort cases possibly needing to be paid and calculate the paid amount, wherein the medical cases related to litigation are generally processed by court trial of the medical cases related to payment of the medical insurance fund, and a responsible party and the paid amount are determined; medical insurance staff acquires judgment information from a court, and knows which cases need to be paid and the amount of paid. In the process, the manual participation degree is too high, such as file sending, file processing, information arrangement and the like, which is time-consuming and easy to make mistakes, and the condition of uneven resource allocation easily occurs due to the inefficiency of manual operation, so that the management and the use of the medical insurance foundation are affected.
In addition, in the above manner, the basis of recovery amount is the judgment amount when the medical insurance staff expands the medical insurance fund to make a compensation work, but on one hand, the medical fee compensation judgment amount is not paid by the medical insurance fund, so that the medical insurance staff needs to perform manual calculation according to other evidences to determine the compensation part; on the other hand, the mode has no medical insurance data tracing mechanism, is easy to generate compensation disputes caused by incomplete or wrong information, and is difficult to ensure the accuracy of recovery amount.
Disclosure of Invention
The invention aims to provide a medical insurance data tracing method and a medical insurance data tracing system based on a question-answer picture text extraction model for strengthening the court linkage, improving the medical insurance compensation efficiency and accuracy and avoiding the medical insurance fund asset loss.
In order to achieve the above purpose, the present invention adopts the following technical scheme:
a medical insurance data tracing method based on a question-answer type picture text extraction model comprises the following steps:
s1, receiving a cue of a medical insurance fund recovery, and triggering a retrospective event;
the compensation clues comprise medical invoice files and medical fee compensation judgment amounts;
s2, sequentially extracting medical invoice files in recovery clues;
S3, sequentially extracting element information of each invoice by using a question-answer picture text extraction model;
S4, judging whether the corresponding invoice relates to medical insurance fund payment according to the element information, if not, executing the step S7, otherwise, executing the step S5;
S5, matching relevant records in the medical insurance system according to the element information and/or the compensation clues;
S6, if the corresponding records are matched, the medical insurance payment amount of the corresponding invoice is increased to the total amount paid by the medical insurance fund, and the self-payment amount of the corresponding invoice is increased to the total amount of the self-payment;
If not, executing a step S7;
S7, continuing to process the next medical invoice file until all medical invoice files of the compensation clues are processed, and executing the step S8;
S8, calculating the following by taking the total self-fee amount as A, the medical fee reimbursement judging amount as B and the total payment amount of the medical insurance fund as C:
D=B*（C/(A+C)）
And outputting the calculated result D to a user as recovery required amount.
According to the scheme, the trace events are triggered by using the medical insurance fund compensation clues, element information in the recovery clues is extracted through the question-answer type picture text extraction model, the medical insurance fund payment sum is automatically calculated according to the clues, recovery sum is obtained according to the medical fee compensation judgment sum and the patient self-fee sum, the medical insurance staff is effectively helped to develop the compensation work, the recovery sum is obtained based on trace, detailed records and relevant evidence are provided, and the validity and rationality of the compensation can be ensured.
In the medical insurance data tracing method based on the question-answer type picture text extraction model, the compensation clues further comprise any one or more of names, identification card numbers, case numbers, whether the medical insurance fund check functions and evidence materials are fulfilled. In the medical insurance data tracing method based on the question-answer type picture text extraction model, the method further comprises the following steps:
After the medical invoice file is processed, comparing the total payment amount of the medical insurance fund with the medical fee reimbursement judgment amount, and outputting the lower amount to the user as the amount required recovery.
In the method for tracing medical insurance data based on the question-answer picture text extraction model, in step S3, the extracted element information includes any one or more of bill number, date of invoicing, medical insurance number, payment amount of medical insurance overall fund, self-payment amount and unified social credit code of payor.
In the method for tracing the medical insurance data based on the question-answer type picture text extraction model, in step S4, whether the corresponding invoice relates to payment of the medical insurance funds is judged according to the payment amount of the medical insurance overall fund.
In the method for tracing medical insurance data based on the question-answer picture text extraction model, in step S5, the relevant records in the medical insurance system are matched according to the element information including any one or more of bill number, billing date, medical insurance number, medical insurance overall fund payment amount and uniform social credit code of the sender and the identity card number in the trace clue.
In the medical insurance data tracing method based on the question-answer type picture text extraction model, in step S3, the question-answer type picture text extraction model comprises a question feature extraction module, a focusing module and an reasoning module, and element information of a medical invoice file is extracted in the following manner:
The problem feature extraction module is used for embedding and extracting problem features of problem generation problem words for extracting element information;
The focusing module extracts image features of the medical invoice file and divides the image features into areas, and calculates the focusing score of each area on each problem according to the problem features and the image features;
The reasoning module extracts word embedding of the OCR token for the images of the plurality of areas, and performs feature fusion on the word embedding of the OCR token, the problem word embedding and the image features to obtain reasoning scores of element information extracted from the plurality of areas and corresponding to the problem;
And weighting the scores of the focus score and the reasoning score for final prediction to obtain the element information of the medical invoice file.
In the medical insurance data tracing method based on the question-answering type picture text extraction model, the problem feature extraction module comprises a word2vec model and a long-term and short-term memory network;
The word2vec model embeds problem generating problem words into a long-term and short-term memory network;
The long-term and short-term memory network extracts problem features based on the word embedding;
the questions include the question of obtaining the element information of bill number, uniform social credit code of the sender, medical insurance number, general fund payment amount of medical insurance, date of billing, and self-payment amount, such as "what is the bill number in the figure? What are the uniform social credit codes of the payees in the figure? What are the medical insurance numbers in the "and" figure? What are the "and" how do the medical insurance orchestration funds paid in the figure? "," what day of the drawing is the date of billing? "what is the amount paid for the self-fee in the figure? ".
In the medical insurance data tracing method based on the question-answering type picture text extraction model, the focusing module identifies each region in the medical ticketing file image based on the fast R-CNN algorithm, calculates the focusing score of each region for each question by combining the question features and the image features of each region, and divides the boundary region into a question-answering region and a non-question-answering region according to the focusing score;
the reasoning module extracts OCR token for a plurality of area images based on OCR technology, extracts OCRtoken word embedding by utilizing FastText model, and finally performs feature fusion on word embedding, problem word embedding and image features of the OCR token to obtain the reasoning score of the element information extracted from a plurality of areas and corresponding to the problem.
A medical insurance data traceability system based on a question-answer picture text extraction model is used for executing the method.
The invention has the advantages that:
1) The manual participation degree is greatly reduced, a user can trigger a trace-back event and obtain the amount to be paid by performing simple operation as required, and the amount to be paid can be rapidly paid for the medical insurance payment cases of the complaints without performing complicated file sending, file processing, information arrangement and other works, so that the payment efficiency and accuracy of the medical insurance foundation are improved;
2) Providing accurate data support for medical insurance compensation staff, reasonably and effectively providing results, and helping the staff to develop compensation more efficiently;
3) The method has the advantages that the compensation flow of the medical insurance fund is optimized through the automatic and intelligent flow, the operation efficiency of the whole system is improved, a large number of complaint medical insurance fund payment cases can be efficiently processed, the same processing standard and flow are ensured to be applied to different cases, the problems that management and use of the medical insurance fund are affected due to uneven resource allocation and the like are avoided, and meanwhile, the consistency and fairness of compensation work can be improved;
4) A series of information related to the case is used as a compensation clue to develop a traceable event, and accurate recovery amount is automatically calculated according to detailed payment records and related evidence, so that the validity and rationality of compensation can be determined;
5) Extracting a question-answer text of the invoice, calculating the payment amount of the medical insurance fund related to the related case based on the extracted element information, and helping medical insurance fund staff to realize the arrangement of the amount required recovery under the reliable tracing of the medical insurance fund;
6) Aiming at the characteristics of the invoice, a targeted question-answer text extraction method is provided, different areas are focused and divided firstly, the focusing score of each area on each question is obtained according to the characteristics of the question, the question area is screened out, then an reasoning module is used for obtaining the reasoning score by combining word embedding, question word embedding and image characteristics of OCR token, and finally the prediction is obtained based on weighting of the focusing score and the reasoning score, so that the recognition accuracy and the calculation efficiency can be effectively improved, and the accuracy of the answer is improved.
Drawings
FIG. 1 is a flow chart of a method for tracing medical insurance data based on a question-answer picture text extraction model;
fig. 2 is a block diagram of a questioning-and-answering type picture text extraction network of a medical insurance data tracing method and a system thereof based on a questioning-and-answering type picture text extraction model.
Detailed Description
The invention will be described in further detail with reference to the drawings and the detailed description.
The scheme discloses a medical insurance data method and a system based on a question-answer type picture text extraction model, which are connected with a court system and receive medical insurance foundation compensation clues from the court system. When a case relates to the compensation amount of medical expense, the court system automatically or manually pushes the compensation clue structured data of the case to the system, the case is added to the medical insurance fund compensation clue list by the system, a medical insurance worker clicks a receiving button aiming at the data in the medical insurance fund compensation clue list, after clicking, the medical insurance worker checks details of the clue through a popup window, and after clicking confirmation, a background service interface is called to start a traceability event. After the execution of the trace-back event is finished, the system returns a result to be processed to the medical insurance staff and reminds the medical insurance staff to process the result.
Structured data refers to data that organizes case-related information in a standardized, easily handled format that is organized into records or tables in a database to facilitate subsequent trace-back events.
Specifically, as shown in fig. 1, the method for tracing medical insurance data based on the question-answer type picture text extraction model provided in this embodiment is as follows:
S1, receiving a medical insurance fund compensation clue, including a name, an identity card number, a case number, a medical fee compensation judgment amount, whether the medical insurance fund compensation clue is fulfilled, a medical insurance fund check function, an evidence material (such as a medical fee invoice, and the like, which is generally in a picture format, and is converted into a picture format if not in the picture format), and triggering a traceability event;
S2, sequentially extracting medical invoice pictures in recovery clues;
S3, extracting element information of the invoice from all medical invoice pictures sequentially by using a question-answer picture text extraction model;
S4, judging whether the corresponding invoice relates to medical insurance fund payment according to the element information, if not, executing the step S7, otherwise, executing the step S5;
S5, matching relevant records in the medical insurance system according to the element information and/or the compensation clues;
s6, if the corresponding records are matched, the medical insurance payment amount of the corresponding invoice is increased to the medical insurance fund payment amount of the current time, and the self-payment amount of the corresponding invoice is increased to the total self-payment amount of the current time;
If not, executing a step S7;
S7, continuing to process the next medical invoice file until all medical invoice files of the compensation clues are processed, and executing the step S8;
S8, after the picture processing is finished, the total self-fee amount is A, the medical fee reimbursement judgment amount is B, and the total medical insurance fund payment amount is C, so that the following calculation is performed:
D=B*（C/(A+C)）
And outputting the calculated result D to a user as recovery required amount.
The total payment amount of the current medical insurance fund, the total self-fee amount and the medical fee reimbursement judgment amount can be simultaneously output to the user for reference of medical insurance staff.
The process can sequentially process one Zhang Yiliao invoice picture, namely, sequentially execute the steps S2-S7 on each medical invoice picture, or execute the steps S2 and S3 on all medical invoice files first, and then execute the steps S4-S7 on one Zhang Fapiao; or step S2 is performed first, and then steps S3-S7 are performed on a sheet Zhang Fapiao. The drawings of the present embodiment are given as examples of the first mode.
Specifically, the element information extracted in step S3 includes a bill number, an invoicing date, a medical insurance number, a payment amount of a medical insurance overall fund, a self-payment amount, and the like;
in step S4, whether the corresponding invoice relates to the payment of the medical insurance fund is judged according to the payment amount of the medical insurance fund, if the payment amount is 0, the payment of the medical insurance fund is not related, and if the payment amount is more than 0, the payment of the medical insurance fund is related.
In step S5, the relevant records in the medical insurance system are matched according to the element information and the identification card number in the compensation clue. The system can be an independent system connected with the medical insurance system with the medical insurance records, or can be the medical insurance system directly.
Further, as shown in fig. 2, in step S3, the question-answering type picture text extraction model includes a question feature extraction module, a focusing module and an inference module, and extracts element information of the medical invoice file by:
1) Question feature extraction module, for "what is the ticket number in the graph? What are the uniform social credit codes of the payees in the figure? What are the medical insurance numbers in the "and" figure? What are the "and" how do the medical insurance orchestration funds paid in the figure? "," what day of the drawing is the date of billing? "what is the amount paid for the self-fee in the figure? The problems are embedded by extracting words through a word2vec model and fed into a long-short-term memory network LSTM layer to extract problem characteristics, wherein the characteristics comprise context information of the problems, and the problems are convenient for a follow-up focusing module to use.
2) The focusing module extracts image features of the medical ticket issuing picture based on the fast R-CNN algorithm, performs region division based on the image features, calculates the focusing score of each region on each problem by combining the image features and the problem features of each region, and divides the boundary region into a question-answering region and a non-question-answering region according to the focusing score. Each region has a focus score for each question, with higher focus scores indicating a higher correlation of the region with the corresponding question.
3) And the reasoning module is used for realizing OCR token extraction of the images of each region based on an OCR technology, extracting word embedding of the OCR token through a FastText model, completing OCR embedding, question word embedding and feature fusion of image features, classifying the detected text as an answer library, and calculating to obtain the reasoning score of the element information extracted from each region and corresponding to the question. Feature fusion and classification can use specific neural networks, such as feature fusion algorithms and classification algorithms based on deep learning theory, such as full-connection layers, convolution layers and the like, and form an inference module together with OCR models, fastText models and the like.
Here, the word embedding of the OCR token may be extracted for all the area images divided by the focusing module, or an area with a focus score higher than a set value for at least one question may be divided into question-answering areas, and then the word embedding of the OCR token may be extracted for only the question-answering area images. This set value is determined by a person skilled in the art according to the actual circumstances.
4) And extracting results, namely adding the weighted scores of the focus scores and the reasoning scores for final prediction.
The final prediction result is based on the weighted score, which is helpful to improve the accuracy of the answer. For a question, if the focus score of a region is high, but the inference score is not high, it may be because the region is related to the question, but does not necessarily contain the correct answer; conversely, if the inference score is high but the focus score is not high, it may be that the answer is semantically reasonable but not an accurate answer to the question. The weighted score allows the system to find a balance between these two aspects, thereby improving overall prediction quality.
The above mentioned models, such as word2vec model, LSTM, fast R-CNN, OCR model, fastText model, etc. are all trained models, and the innovation of the present solution is not to train the model, and the training method of the model should be clear to those skilled in the art, so the training process will not be repeated here. The models can be used directly with the pre-training network or after fine tuning the pre-training network with a specific data set, or from scratch with a specific data set, as desired.
The modules can train simultaneously or separately, and the embodiment uses a pre-training network for the word2vec model and the LSTM. And carrying out joint training on the focusing module and the reasoning module.
In order to make the reader better understand the present disclosure, the following further describes the present disclosure by taking an application scenario as an example:
A medical insurance participant is subjected to medical fee reimbursement due to traffic accidents, the medical fee spends 12 ten thousand yuan, and the court judges that the reimbursement amount is 10 ten thousand yuan.
The court system pushes recovery cues for the case to the system in the form of structured data. These cues include: name, identification number, case number, medical fee reimbursement decision amount, whether or not it has been fulfilled, medical insurance fund check box, medical invoice, etc., wherein there are 38 medical invoices.
The system receives the compensation clues and starts a tracing event;
The system sequentially extracts the medical invoice pictures in recovery clues, and extracts element information, such as bill numbers, invoicing dates, medical insurance numbers, overall fund payment amount, self-fee amount and the like, of all the medical invoice pictures by using a question-answer picture text extraction model.
And judging whether the invoice relates to the payment of the medical insurance foundation according to the extracted element information. If the overall foundation payment is 0, then no foundation payment is involved; if the payment amount is greater than 0, a medical insurance fund payment is involved. The result is that all 38 medical invoices involve a payment of the medical insurance foundation.
The system matches the relevant records in the medical insurance system according to the element information and the identification card number in the compensation clues.
The system adds the medical insurance payment amount of the corresponding invoice to the medical insurance fund payment total amount, adds the self-payment amount to the self-payment total amount, and after all 38 invoices are added, the self-payment amount of 3 ten thousand yuan and the medical insurance fund payment total amount of 9 ten thousand yuan are calculated.
According to formula d=b (C/(a+c))=10 (9/(3+9))=7.5 ten thousand.
The system outputs the recovery U of the calculated amount to the medical insurance staff, and provides the total payment amount, the total self-payment amount and the medical fee reimbursement judgment amount of the medical insurance fund for reference.
And the medical insurance staff performs subsequent compensation work according to the compensation amount provided by the system.
In the implementation process of the scheme, the system automatically calculates the to-be-paid medical insurance funds based on the trigger of the cue of the medical insurance funds recovery, and medical insurance staff only need to receive recovery amount calculated by the system, so that a large number of complaint medical insurance funds payment cases can be efficiently processed, timeliness, effectiveness and accuracy of the medical insurance funds to be paid are improved, and the asset loss of public medical insurance funds is avoided.
The specific embodiments described herein are offered by way of example only to illustrate the spirit of the invention. Those skilled in the art may make various modifications or additions to the described embodiments or substitutions thereof without departing from the spirit of the invention or exceeding the scope of the invention as defined in the accompanying claims.
Although terms such as medical insurance fund tracking clues, medical invoice files, medical fee reimbursement decision amounts, element information, medical insurance fund payment amounts, and the like are used more herein, the possibility of using other terms is not excluded. These terms are used merely for convenience in describing and explaining the nature of the invention; they are to be interpreted as any additional limitation that is not inconsistent with the spirit of the present invention.
Claims (10)
1. A medical insurance data tracing method based on a question-answer type picture text extraction model is characterized by comprising the following steps of:
s1, receiving a cue of a medical insurance fund recovery, and triggering a retrospective event;
the compensation clues comprise medical invoice files and medical fee compensation judgment amounts;
s2, sequentially extracting medical invoice files in recovery clues;
S3, sequentially extracting element information of each invoice by using a question-answer picture text extraction model;
S4, judging whether the corresponding invoice relates to medical insurance fund payment according to the element information, if not, executing the step S7, otherwise, executing the step S5;
S5, matching relevant records in the medical insurance system according to the element information and/or the compensation clues;
S6, if the corresponding records are matched, the medical insurance payment amount of the corresponding invoice is increased to the total amount paid by the medical insurance fund, and the self-payment amount of the corresponding invoice is increased to the total amount of the self-payment;
If not, executing a step S7;
S7, continuing to process the next medical invoice file until all medical invoice files of the compensation clues are processed, and executing the step S8;
S8, calculating the following by taking the total self-fee amount as A, the medical fee reimbursement judging amount as B and the total payment amount of the medical insurance fund as C:
D=B*（C/(A+C)）
And outputting the calculated result D to a user as recovery required amount.
2. The method for tracing medical insurance data based on question-answer picture text extraction model according to claim 1, wherein the compensation clues further comprise any one or more of name, identification card number, case number, whether or not they have been fulfilled, medical insurance fund check function and evidence material.
3. The method for tracing medical insurance data based on question-answer picture text extraction model according to claim 2, wherein in step S3, the extracted element information includes any one or more of bill number, date of invoicing, medical insurance number, payment amount of medical insurance overall fund, self-payment amount, and unified social credit code of payor.
4. The method for tracing medical insurance data based on question-answering type picture text extraction model according to claim 3, wherein in step S4, whether corresponding invoices relate to payment of medical insurance funds is judged according to the payment amount of medical insurance overall funds.
5. The method for tracing medical insurance data based on question-answer picture text extraction model according to claim 4, wherein in step S5, the element information including any one or more of bill number, billing date, medical insurance number, payment amount of medical insurance overall fund, uniform social credit code of the sender and the identification card number in the compensation clue are matched with the related records in the medical insurance system.
6. The method for tracing medical insurance data based on question-answering type picture text extraction model according to claim 1, wherein in step S3, the question-answering type picture text extraction model includes a question feature extraction module, a focusing module and an reasoning module, and element information of a medical invoice file is extracted by:
The problem feature extraction module is used for embedding and extracting problem features of problem generation problem words for extracting element information;
The focusing module extracts image features of the medical invoice file and divides the image features into areas, and calculates the focusing score of each area on each problem according to the problem features and the image features;
The reasoning module extracts word embedding of the OCR token for the images of the plurality of areas, and performs feature fusion on the word embedding of the OCR token, the problem word embedding and the image features to obtain reasoning scores of element information extracted from the plurality of areas and corresponding to the problem;
And weighting the scores of the focus score and the reasoning score for final prediction to obtain the element information of the medical invoice file.
7. The method for tracing medical insurance data based on question-answering type picture text extraction model according to claim 6, wherein the question feature extraction module comprises a word2vec model and a long-term and short-term memory network;
the word2vec model embeds the problem generating word and inputs the word into a long-term and short-term memory network;
The long-term and short-term memory network is used for embedding and extracting the problem features based on the problem words.
8. The method for tracing medical insurance data based on a question-answering type picture text extraction model according to claim 7, wherein the focusing module identifies each region in the medical ticketing file image based on a fast R-CNN algorithm, calculates a focusing score of each region for each question by combining the question features and the image features of each region, and divides the boundary region into a question-answering region and a non-question-answering region according to the focusing score.
9. The method for tracing medical insurance data based on question-answering type picture text extraction model according to claim 8, wherein the reasoning module extracts OCR token for a plurality of area images based on OCR technology, extracts word embedding of the OCR token by utilizing FastText model, and finally performs feature fusion on word embedding, question word embedding and image features of the OCR token to obtain reasoning scores of element information extracted from a plurality of areas and corresponding to the questions.
10. A medical insurance data traceability system based on a question-answer picture text extraction model, which is used for executing the method of any one of claims 1-9.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410586664.4A CN118152548A (en) | 2024-05-13 | 2024-05-13 | Medical insurance data tracing method and system based on question-answer type picture text extraction model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410586664.4A CN118152548A (en) | 2024-05-13 | 2024-05-13 | Medical insurance data tracing method and system based on question-answer type picture text extraction model |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN118152548A true CN118152548A (en) | 2024-06-07 |
Family
ID=91290590
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202410586664.4A Pending CN118152548A (en) | 2024-05-13 | 2024-05-13 | Medical insurance data tracing method and system based on question-answer type picture text extraction model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN118152548A (en) |
Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107798514A (en) * | 2017-11-22 | 2018-03-13 | 阿里巴巴集团控股有限公司 | The method and apparatus that Claims Resolution is realized based on credit |
| WO2019090983A1 (en) * | 2017-11-10 | 2019-05-16 | 平安科技（深圳）有限公司 | Claim settlement method, apparatus and system based on medical insurance |
| CN112926577A (en) * | 2021-02-20 | 2021-06-08 | 杭州海量信息技术有限公司 | Medical bill image structuring method and device and computer readable medium |
| CN113360621A (en) * | 2021-06-22 | 2021-09-07 | 辽宁工程技术大学 | Scene text visual question-answering method based on modal inference graph neural network |
| KR20210130362A (en) * | 2020-04-22 | 2021-11-01 | (주)한국보험보장연구소 | Method and server for providing question and answer service related to insurance based on insurance policy analysis |
| CN113705733A (en) * | 2021-09-29 | 2021-11-26 | 平安医疗健康管理股份有限公司 | Medical bill image processing method and device, electronic device and storage medium |
| CN113902573A (en) * | 2021-09-24 | 2022-01-07 | 北京健康之家科技有限公司 | Method and device for processing claim settlement data, storage medium and terminal |
| CN116228265A (en) * | 2023-03-24 | 2023-06-06 | 北京中诺链捷数字科技有限公司 | Invoice risk identification method, device and equipment |
| CN116433391A (en) * | 2023-04-19 | 2023-07-14 | 平安科技（深圳）有限公司 | Method and device for processing claim information, electronic equipment and storage medium |
| CN116563033A (en) * | 2023-04-17 | 2023-08-08 | 因朔桔(上海)智能科技有限公司 | Data-driven automatic claim settlement system |
| CN116912853A (en) * | 2023-07-26 | 2023-10-20 | 云知声智能科技股份有限公司 | OCR (optical character recognition) method and system based on medical expense list image |
-
2024
- 2024-05-13 CN CN202410586664.4A patent/CN118152548A/en active Pending
Patent Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2019090983A1 (en) * | 2017-11-10 | 2019-05-16 | 平安科技（深圳）有限公司 | Claim settlement method, apparatus and system based on medical insurance |
| CN107798514A (en) * | 2017-11-22 | 2018-03-13 | 阿里巴巴集团控股有限公司 | The method and apparatus that Claims Resolution is realized based on credit |
| KR20210130362A (en) * | 2020-04-22 | 2021-11-01 | (주)한국보험보장연구소 | Method and server for providing question and answer service related to insurance based on insurance policy analysis |
| CN112926577A (en) * | 2021-02-20 | 2021-06-08 | 杭州海量信息技术有限公司 | Medical bill image structuring method and device and computer readable medium |
| CN113360621A (en) * | 2021-06-22 | 2021-09-07 | 辽宁工程技术大学 | Scene text visual question-answering method based on modal inference graph neural network |
| CN113902573A (en) * | 2021-09-24 | 2022-01-07 | 北京健康之家科技有限公司 | Method and device for processing claim settlement data, storage medium and terminal |
| CN113705733A (en) * | 2021-09-29 | 2021-11-26 | 平安医疗健康管理股份有限公司 | Medical bill image processing method and device, electronic device and storage medium |
| CN116228265A (en) * | 2023-03-24 | 2023-06-06 | 北京中诺链捷数字科技有限公司 | Invoice risk identification method, device and equipment |
| CN116563033A (en) * | 2023-04-17 | 2023-08-08 | 因朔桔(上海)智能科技有限公司 | Data-driven automatic claim settlement system |
| CN116433391A (en) * | 2023-04-19 | 2023-07-14 | 平安科技（深圳）有限公司 | Method and device for processing claim information, electronic equipment and storage medium |
| CN116912853A (en) * | 2023-07-26 | 2023-10-20 | 云知声智能科技股份有限公司 | OCR (optical character recognition) method and system based on medical expense list image |
Non-Patent Citations (4)
| Title |
|---|
| IOANNIS PAPAIOANNOU: "Combining Chat and Task-Based Multimodal Dialogue for More Engaging HRI: A Scalable Method Using Reinforcement Learning", HRI \'17: PROCEEDINGS OF THE COMPANION OF THE 2017 ACM/IEEE INTERNATIONAL CONFERENCE ON HUMAN-ROBOT INTERACTION, 6 March 2017 (2017-03-06) * |
| 徐恺;王振宇;王旭;秦华;龙宇轩: "基于强化学习的任务型对话策略研究综述", 计算机学报, 25 January 2024 (2024-01-25) * |
| 沈焕根;: "基本医疗保险先行支付及追偿机制探讨", 中国医疗保险, no. 08, 16 August 2011 (2011-08-16) * |
| 郑中鹏;姜虹;: "广东省医疗损害责任诉讼中基本医疗保险基金流失现状", 医学与社会, no. 01, 10 January 2017 (2017-01-10) * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US20230377032A1 (en) | System and method for processing transaction records for users | |
| US11783422B1 (en) | Implementing machine learning for life and health insurance claims handling | |
| US7917378B2 (en) | System for processing healthcare claim data | |
| CN110929039B (en) | Data processing method, device, equipment and storage medium | |
| CN109034727A (en) | Self-service electronic government affairs processing method | |
| CN108509484B (en) | Classifier construction and intelligent question and answer method, device, terminal and readable storage medium | |
| CN109584034A (en) | Legal document generation method and system | |
| US20200151386A1 (en) | System and interface for electronic communication generation and tracking | |
| CN112712429A (en) | Remittance service auditing method, remittance service auditing device, computer equipment and storage medium | |
| US20140207631A1 (en) | Systems and Method for Analyzing and Validating Invoices | |
| CN110362799A (en) | Processing method, device and computer equipment are generated based on the award arbitrated online | |
| US20220292268A1 (en) | Smart contract generation system and methods | |
| CN110517007A (en) | Request for arbitration, which is put on record, audits processing method, device and computer equipment | |
| CN110858353B (en) | Method and system for obtaining case judge result | |
| CN112347254A (en) | News text classification method and device, computer equipment and storage medium | |
| CN116071150A (en) | Data processing method, bank product popularization, wind control system, server and medium | |
| CN111898611A (en) | Method, device, equipment and medium for generating collection voucher combining RPA and AI | |
| CN111061948A (en) | User label recommendation method and device, computer equipment and storage medium | |
| CN113255929B (en) | Method and device for acquiring interpretable reasons of abnormal user | |
| CN112364136B (en) | Keyword generation method, device, equipment and storage medium | |
| CN111400529B (en) | Data processing method and device | |
| CN117114901A (en) | Method, device, equipment and medium for processing insurance data based on artificial intelligence | |
| CN112835910A (en) | Enterprise information and policy information processing method and device | |
| CN118152548A (en) | Medical insurance data tracing method and system based on question-answer type picture text extraction model | |
| CN113779998A (en) | Structured processing method, device and equipment of service information and storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination |