A bill processing method, device, equipment, medium and product

Abstract

The application discloses a kind of bill processing method, device, equipment, medium and product, by the bill information collected to desensitization processing, obtain desensitization bill information, send desensitization bill information to generative pre-training language model, to make desensitization bill information is stored in enterprise special bill model;Send the vote requirement instruction sent by enterprise end to generative pre-training language model, to make corresponding target bill information is filtered out in the enterprise special bill model based on vote requirement instruction;Decryption processing is carried out to target bill information sent by generative pre-training language model, obtain decryption target bill information, and send decryption target bill information to enterprise end, to make enterprise end based on decryption target bill information carries out payment settlement;Compared with prior art scheme, the technical scheme of the application can improve the accuracy and efficiency of bill information acquisition.

Description

Technical Field

[0001] This invention relates to the technical field of financial data processing, and in particular to a method, apparatus, equipment, medium, and product for processing invoices. Background Technology

[0002] When companies use inventory bills for payment, they often face a series of challenges. For example, choosing the right bill for payment requires comprehensive consideration of multiple factors, such as the acceptance period, face value, and acceptance risk. An inappropriate choice may lead to liquidity risk.

[0003] When processing bills, traditional enterprises typically incorporate information on bank acceptance bills, commercial acceptance bills, and supply chain bills into their ERP systems to facilitate subsequent billing within the ERP system. This requires the ERP system to develop multiple interfaces to connect to external billing systems to obtain bill information, resulting in complex operations and low efficiency in obtaining bill information.

[0004] While artificial intelligence (AI) technology has a strong driving force for the development of invoice processing, traditional solutions rely on Robotic Process Automation (RPA). RPA can only handle the input and storage of simple, repetitive invoice information with clearly defined steps and fixed rules. The input process requires structured data processing, which is extremely complex. ERP systems need to develop related invoice modules to store the data, but this also cannot solve the problem of storing large amounts of data and maintaining system specialization. Traditional solutions also interface with various external software through Enterprise Resource Planning (ERP) systems, but this is difficult to standardize. Therefore, how to solve the pain points of AI technology in the field of invoice processing to improve the accuracy of enterprises in selecting inventory invoices is a pressing technical problem that needs to be addressed. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide a method, apparatus, equipment, medium and product for processing invoices, so as to improve the accuracy and efficiency of invoice information acquisition.

[0006] To address the aforementioned technical problems, this invention provides a bill processing method, comprising:

[0007] Collect invoice information from enterprises, de-identify the invoice information to obtain de-identified invoice information, and send the de-identified invoice information to a generative pre-trained language model so that the generative pre-trained language model stores the de-identified invoice information in a preset enterprise-specific invoice model;

[0008] The system receives a ballot request instruction sent by the enterprise and sends the ballot request instruction to the generative pre-trained language model, so that the generative pre-trained language model can filter out the corresponding target bill information in the enterprise-specific bill model based on the ballot request instruction.

[0009] The system receives the target invoice information sent by the generative pre-trained language model, decrypts the target invoice information to obtain decrypted target invoice information, and sends the decrypted target invoice information to the enterprise so that the enterprise can make payment settlement based on the decrypted target invoice information.

[0010] In one possible implementation, the collection of invoice information from the enterprise includes:

[0011] Obtain the invoice image from the enterprise, extract the text content from the invoice image, and obtain structured invoice data;

[0012] The structured invoice data is input into a preset invoice data template to obtain the invoice information corresponding to the invoice image.

[0013] In one possible implementation, the invoice information is de-identified to obtain de-identified invoice information, specifically including:

[0014] The sensitive data in the invoice information is obtained, and the sensitive data is encrypted to obtain encrypted sensitive data.

[0015] Set a corresponding sensitive key name for the sensitive data, and use the encrypted sensitive data as a sensitive key value. Concatenate the sensitive key name and the sensitive key value in the form of a key-value pair to obtain a sensitive data key-value pair.

[0016] Obtain non-sensitive data from the invoice information, set corresponding non-sensitive key names for the non-sensitive data, and use the non-sensitive data as non-sensitive key values. Concatenate the non-sensitive key names and the non-sensitive key values ​​in the form of key-value pairs to obtain non-sensitive data key-value pairs.

[0017] The sensitive data key-value pairs and the non-sensitive data key-value pairs are concatenated to obtain de-identified invoice information.

[0018] In one possible implementation, before sending the vote request instruction to the generative pre-trained language model, the method further includes:

[0019] The ballot request instruction is de-identified to obtain the de-identified ballot filtering conditions corresponding to the ballot request instruction;

[0020] Specifically, the process of de-identifying the ballot request instruction to obtain the de-identified ballot filtering conditions corresponding to the ballot request instruction includes:

[0021] The ballot request instructions include the ballot request content and ballot filtering conditions;

[0022] Obtain sensitive data of the ballots from the ballot filtering conditions, and encrypt the sensitive data of the ballots to obtain encrypted sensitive data of the ballots;

[0023] Set a corresponding sensitive key name for the sensitive ballot data, and use the encrypted sensitive ballot data as a sensitive key value. Assemble the sensitive key name and the sensitive key value in the form of a key-value pair to obtain a sensitive ballot data key-value pair.

[0024] Obtain non-sensitive data of the ballots from the ballot filtering conditions, set a corresponding non-sensitive key name for the non-sensitive data of the ballots, and use the non-sensitive data of the ballots as non-sensitive key values. Concatenate the non-sensitive key name and the non-sensitive key value of the ballots in the form of key-value pairs to obtain non-sensitive data key-value pairs of the ballots.

[0025] The sensitive data key-value pairs and the non-sensitive data key-value pairs of the ballots are concatenated to obtain the desensitized ballot filtering conditions.

[0026] In one possible implementation, the generative pre-trained language model filters out the corresponding target bill information from the enterprise-specific bill model based on the ballot requirement instruction, specifically including:

[0027] The ballot request instruction includes the ballot request content and the de-identified ballot filtering conditions;

[0028] The generative pre-trained language model selects all first de-identified invoice information corresponding to the ballot requirement content from the enterprise-specific invoice model based on the ballot requirement content in the ballot requirement instruction, and filters all the first de-identified invoice information based on the de-identified ballot filtering conditions to obtain the target invoice information.

[0029] In one possible implementation, the generative pre-trained language model selects all first de-identified invoice information corresponding to the ballot requirement content from the enterprise-specific invoice model based on the ballot requirement content in the ballot requirement instruction, specifically including:

[0030] The generative pre-trained language model extracts keywords from the content of the ballot requirements to obtain ballot requirement keywords, and then retrieves all de-identified invoice information in the enterprise special invoice model based on the ballot requirement keywords.

[0031] When the first desensitized invoice information matching the voting requirement keywords is retrieved, the first desensitized invoice information is saved, and the retrieval of other desensitized invoice information in the enterprise-specific invoice model continues based on the voting requirement keywords until all desensitized invoice information in the enterprise-specific invoice model has been retrieved, thus obtaining all the first desensitized invoice information.

[0032] In one possible implementation, the target invoice information is decrypted to obtain decrypted target invoice information, specifically including:

[0033] The target invoice information includes target sensitive data key-value pairs and target non-sensitive data key-value pairs;

[0034] Obtain the target sensitive key name and target sensitive key value from the target sensitive data key value pair, decrypt the target sensitive key value to obtain the decrypted target sensitive key value, and assemble the target sensitive key name and the decrypted target sensitive key value in the form of a key value pair to obtain the target decrypted sensitive data key value pair;

[0035] The target sensitive data key-value pairs and the target non-sensitive data key-value pairs are concatenated to obtain the decrypted target ticket information.

[0036] The present invention also provides a bill processing device, comprising: a bill information storage module, a bill information filtering module, and a bill information decryption module;

[0037] The invoice information storage module is used to collect invoice information from the enterprise, perform desensitization processing on the invoice information to obtain desensitized invoice information, and send the desensitized invoice information to the generative pre-trained language model so that the generative pre-trained language model stores the desensitized invoice information in a preset enterprise-specific invoice model.

[0038] The invoice information filtering module is used to receive the ballot request instruction sent by the enterprise and send the ballot request instruction to the generative pre-trained language model, so that the generative pre-trained language model can filter out the corresponding target invoice information in the enterprise-specific invoice model based on the ballot request instruction.

[0039] The invoice information decryption module is used to receive the target invoice information sent by the generative pre-trained language model, decrypt the target invoice information to obtain decrypted target invoice information, and send the decrypted target invoice information to the enterprise so that the enterprise can make payment settlement based on the decrypted target invoice information.

[0040] In one possible implementation, the invoice information storage module is used to collect invoice information from the enterprise, specifically including:

[0041] Obtain the invoice image from the enterprise, extract the text content from the invoice image, and obtain structured invoice data;

[0042] The structured invoice data is input into a preset invoice data template to obtain the invoice information corresponding to the invoice image.

[0043] In one possible implementation, the invoice information storage module is used to perform desensitization processing on the invoice information to obtain desensitized invoice information, specifically including:

[0044] The sensitive data in the invoice information is obtained, and the sensitive data is encrypted to obtain encrypted sensitive data.

[0045] Set a corresponding sensitive key name for the sensitive data, and use the encrypted sensitive data as a sensitive key value. Concatenate the sensitive key name and the sensitive key value in the form of a key-value pair to obtain a sensitive data key-value pair.

[0046] Obtain non-sensitive data from the invoice information, set corresponding non-sensitive key names for the non-sensitive data, and use the non-sensitive data as non-sensitive key values. Concatenate the non-sensitive key names and the non-sensitive key values ​​in the form of key-value pairs to obtain non-sensitive data key-value pairs.

[0047] The sensitive data key-value pairs and the non-sensitive data key-value pairs are concatenated to obtain de-identified invoice information.

[0048] In one possible implementation, the ticket information filtering module, before sending the ballot request instruction to the generative pre-trained language model, is further configured to:

[0049] The ballot request instruction is de-identified to obtain the de-identified ballot filtering conditions corresponding to the ballot request instruction;

[0050] The ballot information filtering module is used to de-identify the ballot request instruction to obtain the de-identified ballot filtering conditions corresponding to the ballot request instruction, specifically including:

[0051] The ballot request instructions include the ballot request content and ballot filtering conditions;

[0052] Obtain sensitive data of the ballots from the ballot filtering conditions, and encrypt the sensitive data of the ballots to obtain encrypted sensitive data of the ballots;

[0053] Set a corresponding sensitive key name for the sensitive ballot data, and use the encrypted sensitive ballot data as a sensitive key value. Assemble the sensitive key name and the sensitive key value in the form of a key-value pair to obtain a sensitive ballot data key-value pair.

[0054] Obtain non-sensitive data of the ballots from the ballot filtering conditions, set a corresponding non-sensitive key name for the non-sensitive data of the ballots, and use the non-sensitive data of the ballots as non-sensitive key values. Concatenate the non-sensitive key name and the non-sensitive key value of the ballots in the form of key-value pairs to obtain non-sensitive data key-value pairs of the ballots.

[0055] The sensitive data key-value pairs and the non-sensitive data key-value pairs of the ballots are concatenated to obtain the desensitized ballot filtering conditions.

[0056] In one possible implementation, the invoice information filtering module is used by the generative pre-trained language model to filter out corresponding target invoice information from the enterprise-specific invoice model based on the ballot requirement instruction, specifically including:

[0057] The ballot request instruction includes the ballot request content and the de-identified ballot filtering conditions;

[0058] The generative pre-trained language model selects all first de-identified invoice information corresponding to the ballot requirement content from the enterprise-specific invoice model based on the ballot requirement content in the ballot requirement instruction, and filters all the first de-identified invoice information based on the de-identified ballot filtering conditions to obtain the target invoice information.

[0059] In one possible implementation, the invoice information filtering module is used by the generative pre-trained language model to select all first de-identified invoice information corresponding to the ballot requirement content from the enterprise-specific invoice model based on the ballot requirement content in the ballot requirement instruction, specifically including:

[0060] The generative pre-trained language model extracts keywords from the content of the ballot requirements to obtain ballot requirement keywords, and then retrieves all de-identified invoice information in the enterprise special invoice model based on the ballot requirement keywords.

[0061] When the first desensitized invoice information matching the voting requirement keywords is retrieved, the first desensitized invoice information is saved, and the retrieval of other desensitized invoice information in the enterprise-specific invoice model continues based on the voting requirement keywords until all desensitized invoice information in the enterprise-specific invoice model has been retrieved, thus obtaining all the first desensitized invoice information.

[0062] In one possible implementation, the bill information decryption module is used to decrypt the target bill information to obtain decrypted target bill information, specifically including:

[0063] The target invoice information includes target sensitive data key-value pairs and target non-sensitive data key-value pairs;

[0064] Obtain the target sensitive key name and target sensitive key value from the target sensitive data key value pair, decrypt the target sensitive key value to obtain the decrypted target sensitive key value, and assemble the target sensitive key name and the decrypted target sensitive key value in the form of a key value pair to obtain the target decrypted sensitive data key value pair;

[0065] The target sensitive data key-value pairs and the target non-sensitive data key-value pairs are concatenated to obtain the decrypted target ticket information.

[0066] The present invention also provides a terminal device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor executes the computer program to implement the ticket management method as described in any of the preceding claims.

[0067] The present invention also provides a computer-readable storage medium comprising a stored computer program, wherein, when the computer program is executed, it controls the device on which the computer-readable storage medium is located to perform the ticket processing method as described in any of the preceding claims.

[0068] The present invention also provides a computer program product, which, when run on a computer device, enables the computer device to implement the ticket processing method as described in any of the above claims.

[0069] The present invention provides a method, apparatus, device, medium, and product for processing invoices, which, compared with the prior art, has the following beneficial effects:

[0070] By collecting invoice information from enterprises, de-identifying the invoice information to obtain de-identified invoice information, and sending the de-identified invoice information to a generative pre-trained language model, the generative pre-trained language model stores the de-identified invoice information in a preset enterprise-specific invoice model; receiving a ballot request instruction from the enterprise and sending the ballot request instruction to the generative pre-trained language model, the generative pre-trained language model filters out the corresponding target invoice information in the enterprise-specific invoice model based on the ballot request instruction; receiving the target invoice information sent by the generative pre-trained language model, decrypting the target invoice information to obtain decrypted target invoice information, and sending the decrypted target invoice information to the enterprise, the enterprise performs payment settlement based on the decrypted target invoice information; compared with the prior art, the technical solution of the present invention improves the efficiency of enterprise-specific invoice information processing. Desensitization prevents the leakage and misuse of sensitive data, protecting corporate privacy and data security. Sending desensitized invoice information to a generative pre-trained language model for storage in a dedicated corporate invoice model allows for quick and convenient management of large amounts of invoice information, reducing storage pressure on the ERP system compared to traditional methods. Furthermore, using the generative pre-trained language model to filter target invoice information from the dedicated corporate invoice model based on voting requirements improves accuracy and efficiency. Compared to manual filtering, it quickly identifies invoices that meet voting requirements, reducing the possibility of errors and omissions. Finally, after decrypting the target invoice information, it is sent to the corporate end, enabling payment and settlement operations, ensuring the accuracy and reliability of the payment process and preventing payment disputes caused by unclear or incorrect invoice information. Attached Figure Description

[0071] Figure 1 This is a flowchart illustrating an embodiment of a bill processing method provided by the present invention;

[0072] Figure 2 This is a schematic diagram of the structure of one embodiment of the ticket processing device provided by the present invention;

[0073] Figure 3 This is a schematic diagram of another invoice processing flow according to an embodiment of the present invention. Detailed Implementation

[0074] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0075] Example 1, see Figure 1 , Figure 1 This is a flowchart illustrating an embodiment of a ticket processing method provided by the present invention, as shown below. Figure 1 As shown, the method includes steps 101-103, as detailed below:

[0076] Step 101: Collect invoice information from the enterprise, perform desensitization processing on the invoice information to obtain desensitized invoice information, and send the desensitized invoice information to the generative pre-trained language model so that the generative pre-trained language model stores the desensitized invoice information in a preset enterprise-specific invoice model.

[0077] In one embodiment, an image of a receipt from the enterprise is acquired, and text content is extracted from the receipt image to obtain structured receipt data. The structured receipt data is then input into a preset receipt data template to obtain the receipt information corresponding to the receipt image.

[0078] Specifically, multiple bank bill systems are integrated into the enterprise ERP management software. Through RPA software, simulating human operation, a large number of bill images are collected from the multiple bank bill systems based on the enterprise ERP management software interface. The bill images include paper bill images and electronic bill images.

[0079] Specifically, the text content of the invoice image is extracted based on OCR optical character recognition technology to obtain structured invoice data. This includes: image processing of the invoice image to obtain a preprocessed invoice image, wherein the image processing includes image scaling, image binarization, and image denoising; target detection is performed on the preprocessed invoice image based on a target detection algorithm to obtain text region bounding boxes in the preprocessed invoice image; text recognition is performed on the text within the text region bounding boxes based on an OCR engine to obtain text recognition results; and text is extracted from the text recognition results using regular expressions. The key fields include the bill type, face value, issue date, payment date, drawer, payee, acceptor, bill number, and endorsement. Combined with the position information of the text region rectangle and the bill layout features, other detailed data from the text recognition results are extracted. The bill type, face value, issue date, payment date, drawer, payee, acceptor, bill number, endorsement, and other detailed data are input into the corresponding positions of a preset bill data template to obtain the bill information corresponding to the bill image.

[0080] Preferably, the type of negotiable instrument is a commercial draft, bank draft, or government bill, etc.; the endorsement refers to each endorsement record during the circulation of the negotiable instrument; and the other detailed data includes password, purpose of payment, margin, credit line, etc.

[0081] Preferably, in addition to extracting text content from the invoice image based on OCR optical character recognition technology to obtain structured invoice data, text content can also be extracted from the invoice image based on data scraping to obtain structured invoice data.

[0082] In one embodiment, a scheduled task is set up through RPA software or through enterprise ERP management software to start the RPA software data collection process. When the RPA software detects that there are invoice images on the enterprise side, it performs the invoice image collection operation.

[0083] In one embodiment, the process of desensitizing the invoice information to obtain desensitized invoice information specifically includes: acquiring sensitive data from the invoice information; encrypting the sensitive data to obtain encrypted sensitive data; setting corresponding sensitive key names for the sensitive data and using the encrypted sensitive data as sensitive key values; assembling the sensitive key names and sensitive key values ​​as key-value pairs to obtain sensitive data key-value pairs; acquiring non-sensitive data from the invoice information; setting corresponding non-sensitive key names for the non-sensitive data and using the non-sensitive data as non-sensitive key values; assembling the non-sensitive key names and non-sensitive key values ​​as key-value pairs to obtain non-sensitive data key-value pairs; and concatenating the sensitive data key-value pairs and non-sensitive data key-value pairs to obtain desensitized invoice information.

[0084] Specifically, the process involves obtaining sensitive data from the bill information, including the face value and payee; and obtaining non-sensitive data from the bill information, including the bill type, issue date, payment date, drawer, acceptor, bill number, endorsement details, and other details.

[0085] Specifically, the sensitive data is encrypted. When the encrypted sensitive data is obtained, the AES symmetric encryption algorithm is used to encrypt the face value of the ticket and the payee in the sensitive data respectively to obtain the encrypted face value of the ticket and the encrypted payee. The encrypted face value of the ticket and the encrypted payee are then integrated to obtain the encrypted sensitive data.

[0086] Preferably, the face value of the ticket and the payee in the sensitive data can be encrypted separately based on the DES symmetric encryption algorithm to obtain encrypted face value and encrypted payee. The encrypted face value and the encrypted payee can then be integrated to obtain encrypted sensitive data.

[0087] Specifically, corresponding sensitive key names are set for the face value and the payee, respectively, to obtain the face value sensitive key name and the payee sensitive key name. The encrypted face value is used as the encrypted face value sensitive key value, and the encrypted payee is used as the encrypted payee sensitive key value. The face value sensitive key name and the encrypted face value sensitive key value are assembled in the form of key-value pairs to obtain the encrypted face value sensitive data key-value pair. The payee sensitive key name and the encrypted payee sensitive key value are assembled in the form of key-value pairs to obtain the encrypted payee sensitive data key-value pair. The encrypted face value sensitive data key-value pair and the encrypted payee sensitive data key-value pair are integrated to obtain the sensitive data key-value pair.

[0088] Specifically, non-sensitive key names are set for the bill type, issue date, payment date, drawer, acceptor, bill number, endorsement, and other detailed data, resulting in non-sensitive key names for bill type, issue date, payment date, drawer, acceptor, bill number, endorsement, and other detailed data. Then, the bill type, issue date, payment date, drawer, acceptor, bill number, endorsement, and other detailed data are used as non-sensitive key values, resulting in non-sensitive key values ​​for bill type, issue date, payment date, drawer, acceptor, bill number, endorsement, and other detailed data.

[0089] Specifically, the non-sensitive key name and value of the bill type are assembled into key-value pairs to obtain non-sensitive data key-value pairs for the bill type; the non-sensitive key name and value of the issuance date are assembled into key-value pairs to obtain non-sensitive data key-value pairs for the issuance date; the non-sensitive key name and value of the payment date are assembled into key-value pairs to obtain non-sensitive data key-value pairs for the payment date; and the non-sensitive key name and value of the drawer are assembled into key-value pairs to obtain non-sensitive data key-value pairs for the drawer. Yes; the acceptor's non-sensitive key name and the acceptor's non-sensitive key value are assembled into key-value pairs to obtain acceptor's non-sensitive data key-value pairs; the ticket number's non-sensitive key name and the ticket number's non-sensitive key value are assembled into key-value pairs to obtain ticket number's non-sensitive data key-value pairs; the circulation endorsement's non-sensitive key name and the circulation endorsement's non-sensitive key value are assembled into key-value pairs to obtain circulation endorsement's non-sensitive data key-value pairs; the other detail data's non-sensitive key name and the other detail data's non-sensitive key value are assembled into key-value pairs to obtain other detail data's non-sensitive data key-value pairs.

[0090] Specifically, the non-sensitive data key-value pairs of the bill type, the non-sensitive data key-value pairs of the issuance date, the non-sensitive data key-value pairs of the payment date, the non-sensitive data key-value pairs of the drawer, the non-sensitive data key-value pairs of the acceptor, the non-sensitive data key-value pairs of the bill number, the non-sensitive data key-value pairs of the circulation endorsement, and the non-sensitive data key-value pairs of other detailed data are integrated to obtain non-sensitive data key-value pairs.

[0091] Specifically, the encrypted face value sensitive data key-value pairs, the encrypted payee sensitive data key-value pairs, the non-sensitive data key-value pairs of the bill type, the non-sensitive data key-value pairs of the issue date, the non-sensitive data key-value pairs of the payment date, the non-sensitive data key-value pairs of the issuer, the non-sensitive data key-value pairs of the acceptor, the non-sensitive data key-value pairs of the bill number, the non-sensitive data key-value pairs of the circulation endorsement, and the non-sensitive data key-value pairs of other detailed data are concatenated to obtain de-identified bill information.

[0092] In one embodiment, the generative pre-trained language model can be a generative pre-trained transform model (GPT), such as the chatgpt language model. A corporate-specific invoice model is constructed in the generative pre-trained language model to store the company's invoice information. The de-identified invoice information is sent to the generative pre-trained language model through RPA software, so that the generative pre-trained language model stores the de-identified invoice information in the corporate-specific invoice model.

[0093] Specifically, the enterprise-specific bill model built in the generative pre-trained language model is formed by combining a large amount of model pre-training and model fine-tuning with data from the enterprise's own vertical field, thereby forming a large model specific to a certain enterprise; preferably, the data from the enterprise's own vertical field includes credit assessment data of bill acceptors.

[0094] In one embodiment, by storing invoice information in a dedicated enterprise invoice model, unlike existing technologies where enterprise invoice processing requires storing invoice information in an ERP system, which necessitates the development of interfaces for the ERP system to connect with bank invoice systems, this approach further frees up ERP system space. Furthermore, by simulating, enhancing, and expanding the interaction between financial personnel and the computer system using RPA software, this approach assists financial personnel in completing the large-volume, repetitive, and easily standardized invoice information collection process. This optimizes financial processes, reduces financial operating costs, and improves the efficiency and quality of invoice information processing. Subsequently, the collected invoice information is input and stored in the "dedicated enterprise invoice model" using RPA software, and sensitive information on the invoices is encrypted during transmission. The "dedicated enterprise invoice model" acts as a super brain.

[0095] Step 102: Receive the ballot request instruction sent by the enterprise and send the ballot request instruction to the generative pre-trained language model so that the generative pre-trained language model can filter out the corresponding target ticket information in the enterprise-specific ticket model based on the ballot request instruction.

[0096] In one embodiment, when the enterprise ERP management software in the enterprise terminal performs payment settlement, it needs to select a bill. At this time, based on the bill to be selected, the voting requirements are set and a voting requirement instruction is generated, and the generated voting requirement instruction is sent to the RPA software.

[0097] In one embodiment, before sending the ballot request instruction to the generative pre-trained language model, the ballot request instruction needs to be de-identified to obtain the de-identified ballot filtering conditions corresponding to the ballot request instruction.

[0098] In one embodiment, the ballot request instruction includes ballot request content and ballot filtering conditions; preferably, the ballot request can be multi-dimensional, including filtering requirements based on the information recorded on the ballot itself, as well as filtering requirements based on external conditions; for example, when a company pays a supplier using a bill, the supplier has requirements for the bill, such as a maturity date within one month, the bill not being accepted by a specific bank, and the endorser not being a specific company, etc. These are all filtering requirements.

[0099] In one embodiment, sensitive data of the ballots in the ballot filtering conditions is obtained, and the sensitive data of the ballots is encrypted to obtain encrypted sensitive data of the ballots.

[0100] Specifically, the sensitive data in the ballot filtering conditions includes one or more of the following: the face value of the ballot, the ballot recipient, and the ballot supplier.

[0101] Specifically, when the ballot sensitive data simultaneously includes the ballot face value, the ballot recipient, and the ballot supplier, the ballot sensitive data is encrypted to obtain encrypted ballot sensitive data. Then, the AES symmetric encryption algorithm is used to encrypt the ballot face value, the ballot recipient, and the ballot supplier in the ballot sensitive data to obtain encrypted ballot face value, encrypted ballot recipient, and encrypted ballot supplier. Finally, the encrypted ballot face value and the encrypted recipient are integrated to obtain the encrypted ballot sensitive data.

[0102] Preferably, the face value of the ballot, the ballot recipient, and the ballot supplier in the ballot sensitive data can also be encrypted using the DES symmetric encryption algorithm to obtain encrypted ballot face value, encrypted ballot recipient, and encrypted ballot supplier. The encrypted face value and the encrypted recipient are then integrated to obtain encrypted ballot sensitive data.

[0103] In one embodiment, a corresponding sensitive key name is set for the sensitive ballot data, and the encrypted sensitive ballot data is used as a sensitive key value. The sensitive key name and the sensitive key value are assembled in the form of key-value pairs to obtain sensitive ballot data key-value pairs.

[0104] Specifically, corresponding sensitive key names are set for the face value of the ballot, the payee of the ballot, and the ballot supplier, resulting in sensitive key names for the face value of the ballot, the payee of the ballot, and the supplier of the ballot. The face value of the encrypted ballot is used as the sensitive key value for the encrypted ballot, the payee of the encrypted ballot is used as the sensitive key value for the encrypted ballot, and the supplier of the encrypted ballot is used as the sensitive key value for the encrypted ballot. The sensitive key names and sensitive key values ​​for the face value of the ballot are then assembled as key-value pairs to obtain the encrypted ballot. The sensitive data key-value pair for the face value of the ballot is obtained by assembling the sensitive key name of the ballot recipient and the sensitive key value of the encrypted ballot recipient as key-value pairs, and by assembling the sensitive key name of the ballot supplier and the sensitive key value of the encrypted ballot supplier as key-value pairs, and by integrating the sensitive data key-value pairs for the face value of the encrypted ballot, the sensitive data key-value pairs for the encrypted ballot recipient, and the sensitive data key-value pairs for the encrypted ballot supplier, to obtain the sensitive data key-value pair for the ballot.

[0105] In one embodiment, non-sensitive data of the ballots in the ballot filtering conditions is obtained, a corresponding non-sensitive key name is set for the non-sensitive data of the ballots, and the non-sensitive data of the ballots is used as a non-sensitive key value of the ballots. The non-sensitive key name of the ballots and the non-sensitive key value of the ballots are assembled in the form of key-value pairs to obtain non-sensitive data key-value pairs of the ballots.

[0106] Specifically, the non-sensitive data of the ballot includes one or more of the following: ballot type, date of issue, date of payment, issuer, acceptor, ballot number, endorsement, and other details.

[0107] Specifically, when the non-sensitive data of the ballot simultaneously contains ballot type, ballot issuance date, ballot payment date, ballot issuer, ballot acceptor, ballot number, ballot circulation endorsement, and other ballot details, corresponding non-sensitive key names are set for each of these data, resulting in non-sensitive key names for ballot type, ballot issuance date, ballot payment date, ballot issuer, ballot acceptor, ballot number, ballot circulation endorsement, and other ballot details. The system uses the following non-sensitive key names: name, ballot number, ballot circulation endorsement, and other ballot details. It then uses the following non-sensitive key values ​​for the ballot type, ballot issuance date, ballot payment date, ballot issuer, ballot acceptor, ballot number, ballot circulation endorsement, and other ballot details: ballot type, ballot issuance date, ballot payment date, ballot issuer, ballot acceptor, ballot number, ballot circulation endorsement, and other ballot details.

[0108] Specifically, the non-sensitive key name and value of the ballot type are assembled into key-value pairs to obtain non-sensitive data key-value pairs for the ballot type; the non-sensitive key name and value of the ballot issuance date are assembled into key-value pairs to obtain non-sensitive data key-value pairs for the ballot issuance date; the non-sensitive key name and value of the ballot payment date are assembled into key-value pairs to obtain non-sensitive data key-value pairs for the ballot payment date; and the non-sensitive key name and value of the ballot issuer are assembled into key-value pairs to obtain non-sensitive data key-value pairs for the ballot issuer. Yes; the non-sensitive key name and value of the ballot acceptor are assembled into key-value pairs to obtain non-sensitive data key-value pairs for the ballot acceptor; the non-sensitive key name and value of the ballot number are assembled into key-value pairs to obtain non-sensitive data key-value pairs for the ballot number; the non-sensitive key name and value of the ballot circulation endorsement are assembled into key-value pairs to obtain non-sensitive data key-value pairs for the ballot circulation endorsement; the non-sensitive key name and value of other ballot detail data are assembled into key-value pairs to obtain non-sensitive data key-value pairs for other ballot detail data.

[0109] Specifically, the non-sensitive data key-value pairs of the ballot type, the ballot issuance date, the ballot payment date, the ballot issuer, the ballot acceptor, the ballot number, the ballot circulation endorsement, and other ballot details are integrated to obtain the ballot non-sensitive data key-value pairs.

[0110] In one embodiment, the sensitive data key-value pairs and the non-sensitive data key-value pairs of the ballots are concatenated to obtain the desensitized ballot filtering conditions.

[0111] Specifically, the sensitive data key-value pairs of the encrypted ballot face value, the sensitive data key-value pairs of the encrypted ballot payee, the sensitive data key-value pairs of the encrypted ballot supplier, the non-sensitive data key-value pairs of the ballot type, the non-sensitive data key-value pairs of the ballot issuance date, the non-sensitive data key-value pairs of the ballot payment date, the non-sensitive data key-value pairs of the ballot issuer, the non-sensitive data key-value pairs of the ballot acceptor, the non-sensitive data key-value pairs of the ballot number, the non-sensitive data key-value pairs of the ballot circulation endorsement, and the non-sensitive data key-value pairs of other ballot details are concatenated to obtain the de-identified ballot filtering conditions.

[0112] In one embodiment, the ballot request instruction is sent to a generative pre-trained language model via RPA software, so that the generative pre-trained language model can use its powerful retrieval function to filter out the target bill information corresponding to the ballot request instruction from the enterprise-specific bill model.

[0113] In one embodiment, the generative pre-trained language model selects all first de-identified invoice information corresponding to the ballot requirement content from the enterprise-specific invoice model based on the ballot requirement content in the ballot requirement instruction, and filters all the first de-identified invoice information based on the de-identified ballot filtering conditions to obtain the target invoice information.

[0114] Specifically, when the generative pre-trained language model selects all first de-identified invoice information corresponding to the ballot requirement content from the enterprise-specific invoice model based on the ballot requirement content in the ballot requirement instruction, the generative pre-trained language model extracts keywords from the ballot requirement content to obtain ballot requirement keywords. Based on the ballot requirement keywords, it sequentially retrieves all de-identified invoice information in the enterprise-specific invoice model. When it retrieves first de-identified invoice information that matches the ballot requirement keywords, it saves the first de-identified invoice information and continues to retrieve other de-identified invoice information in the enterprise-specific invoice model based on the ballot requirement keywords until all de-identified invoice information in the enterprise-specific invoice model has been retrieved, thus obtaining all first de-identified invoice information.

[0115] Specifically, the text preprocessing of the ballot requirements content yields preprocessed ballot requirements content, including the removal of special characters and punctuation marks; the preprocessed ballot requirements content is segmented into multiple words using a Chinese word segmentation tool to obtain segmentation results; stop word removal is performed on the segmentation results to obtain a first segmentation result; and keyword extraction is performed on the first segmentation result based on a keyword extraction algorithm to obtain ballot requirements keywords.

[0116] Preferably, the keywords required for the ballot include one or more of the following: ballot type, ballot issuance date, ballot payment date, ballot issuer, ballot acceptor, ballot number, ballot circulation endorsement, and other ballot details.

[0117] In one embodiment, when all the first de-identified ticket information is filtered based on the de-identified ballot filtering conditions to obtain the target ticket information, it is determined whether each first de-identified ticket information meets the de-identified ballot filtering conditions. If the first de-identified ticket information meets the de-identified ballot filtering conditions, the first de-identified ticket information is retained; otherwise, the first de-identified ticket information is discarded, and the retained first de-identified ticket information is used as the target ticket information.

[0118] In one embodiment, after the generative pre-trained language model outputs the target ticket information, it also sends the target ticket information to the RPA software.

[0119] In one embodiment, a generative pre-trained language model is used to filter a large amount of bill information stored in a corporate-specific bill model based on the ballot filtering conditions. This avoids the bias caused by manual selection and improves the accuracy and efficiency of bill information selection.

[0120] Step 103: Receive the target invoice information sent by the generative pre-trained language model, decrypt the target invoice information to obtain decrypted target invoice information, and send the decrypted target invoice information to the enterprise so that the enterprise can make payment settlement based on the decrypted target invoice information.

[0121] In one embodiment, after receiving the target ticket information sent by the generative pre-trained language model, the RPA software further decrypts the target ticket information to obtain decrypted target ticket information.

[0122] In one embodiment, the target ticket information includes target sensitive data key-value pairs and target non-sensitive data key-value pairs.

[0123] Specifically, the target sensitive data key-value pairs include target encrypted face value sensitive data key-value pairs and target encrypted payee sensitive data key-value pairs; the non-sensitive data key-value pairs include target bill type non-sensitive data key-value pairs, target issue date non-sensitive data key-value pairs, target payment date non-sensitive data key-value pairs, target drawer non-sensitive data key-value pairs, target acceptor non-sensitive data key-value pairs, target bill number non-sensitive data key-value pairs, target circulation endorsement non-sensitive data key-value pairs, and target other detailed data non-sensitive data key-value pairs.

[0124] In one embodiment, the target sensitive key name and target sensitive key value in the target sensitive data key-value pair are obtained, the target sensitive key value is decrypted to obtain the decrypted target sensitive key value, and the target sensitive key name and the decrypted target sensitive key value are assembled in the form of key-value pairs to obtain the target decrypted sensitive data key-value pair.

[0125] Specifically, the target sensitive key name and target sensitive key value are obtained from the target encrypted ticket amount sensitive data key-value pair and the target encrypted payee sensitive data key-value pair, respectively, to obtain the target encrypted ticket amount sensitive data key name, target encrypted ticket amount sensitive data key value, target encrypted payee sensitive data key name, and target encrypted payee sensitive data key value; the target encrypted ticket amount sensitive data key value and the target encrypted payee sensitive data key value are decrypted respectively to obtain the target decrypted ticket amount sensitive data key value and the target decrypted payee sensitive data key value.

[0126] Specifically, during the decryption process, the user's encryption key is obtained, and the sensitive data key values ​​of the target encrypted ticket amount and the target encrypted payee are decrypted based on the decryption algorithm and the key.

[0127] Specifically, the target sensitive key name and the decryption target sensitive key value are assembled in the form of key-value pairs to obtain the target decryption sensitive data key-value pairs.

[0128] The target encrypted face value sensitive data key name and the target decrypted face value sensitive data key value are combined to obtain a target decrypted face value sensitive data key-value pair; the target encrypted payee sensitive data key name and the target decrypted payee sensitive data key value are combined to obtain a target decrypted payee sensitive data key-value pair; the target decrypted face value sensitive data key-value pair and the target decrypted payee sensitive data key-value pair are integrated to obtain a target decrypted sensitive data key-value pair.

[0129] In one embodiment, the target decryption sensitive data key-value pairs and the target non-sensitive data key-value pairs are concatenated to obtain the decrypted target ticket information.

[0130] Specifically, the target decrypted bill amount sensitive data key-value pair, the target decrypted payee sensitive data key-value pair, the target bill type non-sensitive data key-value pair, the target issue date non-sensitive data key-value pair, the target payment date non-sensitive data key-value pair, the target drawer non-sensitive data key-value pair, the target acceptor non-sensitive data key-value pair, the target bill number non-sensitive data key-value pair, the target circulation endorsement non-sensitive data key-value pair, and the target other detailed data non-sensitive data key-value pair are concatenated to obtain the decrypted target bill information.

[0131] In one embodiment, the decrypted target invoice information is sent to the enterprise terminal using RPA software, so that the enterprise terminal can perform payment settlement based on the decrypted target invoice information; for example... Figure 3 As shown, Figure 3 This is another schematic diagram of a bill processing flow provided in this embodiment.

[0132] In summary, the bill processing method provided in this embodiment collects bill information on the enterprise's ERP management software using RPA software, avoiding the need for the ERP system to develop interfaces to connect with multiple bank bill systems. The bill information is stored in a dedicated enterprise bill model using a generative pre-trained language model, reducing the storage requirements of the ERP system. When the enterprise performs payment settlement, it sends bill request instructions to the generative pre-trained language model via RPA software, enabling the model to directly output the target bill information. This information is then fed back to the enterprise via RPA software for payment settlement. This solution combines the expertise of RPA software, the generative pre-trained language model, and the enterprise ERP management software, simplifying complex processes and automating business processes. It improves the accuracy and efficiency of bill information acquisition and uses RPA software and the generative pre-trained language model as a "glue" between bill information and payment settlement, effectively helping to solve the problems of "information silos" and "separation of business and finance" within enterprises. This helps enterprises achieve a "flexible integration" of the bill value chain and the construction of a "large accounting" system.

[0133] Example 2, see Figure 2 , Figure 2 This is a schematic diagram of one embodiment of a ticket processing device provided by the present invention, as shown below. Figure 2 As shown, the device includes a bill information storage module 201, a bill information filtering module 202, and a bill information decryption module 203, as detailed below:

[0134] The invoice information storage module 201 is used to collect invoice information from the enterprise, perform desensitization processing on the invoice information to obtain desensitized invoice information, and send the desensitized invoice information to the generative pre-trained language model so that the generative pre-trained language model stores the desensitized invoice information in a preset enterprise-specific invoice model.

[0135] The invoice information filtering module 202 is used to receive the ballot request instruction sent by the enterprise and send the ballot request instruction to the generative pre-trained language model, so that the generative pre-trained language model can filter out the corresponding target invoice information in the enterprise-specific invoice model based on the ballot request instruction.

[0136] The invoice information decryption module 203 is used to receive the target invoice information sent by the generative pre-trained language model, decrypt the target invoice information to obtain decrypted target invoice information, and send the decrypted target invoice information to the enterprise so that the enterprise can make payment settlement based on the decrypted target invoice information.

[0137] In one embodiment, the invoice information storage module 201 is used to collect invoice information from the enterprise, specifically including: acquiring an invoice image from the enterprise, extracting text content from the invoice image to obtain structured invoice data; and inputting the structured invoice data into a preset invoice data template to obtain invoice information corresponding to the invoice image.

[0138] In one embodiment, the invoice information storage module 201 is used to perform desensitization processing on the invoice information to obtain desensitized invoice information. Specifically, this includes: acquiring sensitive data from the invoice information; encrypting the sensitive data to obtain encrypted sensitive data; setting corresponding sensitive key names for the sensitive data and using the encrypted sensitive data as sensitive key values; assembling the sensitive key names and sensitive key values ​​as key-value pairs to obtain sensitive data key-value pairs; acquiring non-sensitive data from the invoice information; setting corresponding non-sensitive key names for the non-sensitive data and using the non-sensitive data as non-sensitive key values; assembling the non-sensitive key names and non-sensitive key values ​​as key-value pairs to obtain non-sensitive data key-value pairs; and concatenating the sensitive data key-value pairs and non-sensitive data key-value pairs to obtain desensitized invoice information.

[0139] In one embodiment, the ballot information filtering module 202 is used to perform desensitization processing on the ballot request instruction before sending the ballot request instruction to the generative pre-trained language model, so as to obtain the desensitized ballot filtering conditions corresponding to the ballot request instruction.

[0140] In one embodiment, the ballot information filtering module 202 is used to desensitize the ballot requirement instruction to obtain desensitized ballot filtering conditions corresponding to the ballot requirement instruction. Specifically, the process includes: the ballot requirement instruction includes ballot requirement content and ballot filtering conditions; obtaining sensitive ballot data from the ballot filtering conditions, encrypting the sensitive ballot data to obtain encrypted sensitive ballot data; setting a corresponding sensitive ballot key name for the sensitive ballot data, and using the encrypted sensitive ballot data as a sensitive ballot key value, assembling the sensitive ballot key name and the sensitive ballot key value as key-value pairs to obtain sensitive ballot data key-value pairs; obtaining non-sensitive ballot data from the ballot filtering conditions, setting a corresponding non-sensitive ballot key name for the non-sensitive ballot data, and using the non-sensitive ballot data as a non-sensitive ballot key value, assembling the non-sensitive ballot key name and the non-sensitive ballot key value as key-value pairs to obtain non-sensitive ballot data key-value pairs; and concatenating the sensitive ballot data key-value pairs and the non-sensitive ballot data key-value pairs to obtain desensitized ballot filtering conditions.

[0141] In one embodiment, the invoice information filtering module 202 is used by the generative pre-trained language model to filter out corresponding target invoice information from the enterprise-specific invoice model based on the ballot requirement instruction. Specifically, the ballot requirement instruction includes the ballot requirement content and the desensitized ballot filtering conditions. Based on the ballot requirement content in the ballot requirement instruction, the generative pre-trained language model selects all first desensitized invoice information corresponding to the ballot requirement content from the enterprise-specific invoice model, and filters all the first desensitized invoice information based on the desensitized ballot filtering conditions to obtain the target invoice information.

[0142] In one embodiment, the invoice information filtering module 202 is used for the generative pre-trained language model to select all first de-identified invoice information corresponding to the ballot requirement content in the ballot requirement instruction from the enterprise-specific invoice model. Specifically, the generative pre-trained language model extracts keywords from the ballot requirement content to obtain ballot requirement keywords, and sequentially searches all de-identified invoice information in the enterprise-specific invoice model based on the ballot requirement keywords. When first de-identified invoice information matching the ballot requirement keywords is found, the first de-identified invoice information is saved, and the search for other de-identified invoice information in the enterprise-specific invoice model continues based on the ballot requirement keywords until all de-identified invoice information in the enterprise-specific invoice model has been searched, thus obtaining all first de-identified invoice information.

[0143] In one embodiment, the bill information decryption module 203 is used to decrypt the target bill information to obtain decrypted target bill information. Specifically, the decryption module includes: the target bill information comprising target sensitive data key-value pairs and target non-sensitive data key-value pairs; obtaining the target sensitive key name and target sensitive key value from the target sensitive data key-value pairs; decrypting the target sensitive key value to obtain a decrypted target sensitive key value; assembling the target sensitive key name and the decrypted target sensitive key value as key-value pairs to obtain a target decrypted sensitive data key-value pair; and concatenating the target decrypted sensitive data key-value pair and the target non-sensitive data key-value pair to obtain the decrypted target bill information.

[0144] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working process of the device described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0145] It should be noted that the above-described embodiments of the ticket processing device are merely illustrative. The modules described as separate components may or may not be physically separate, and the components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0146] Based on the above-described embodiments of the bill processing method, another embodiment of the present invention provides a bill processing terminal device, which includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the bill processing method of any embodiment of the present invention.

[0147] For example, in this embodiment, the computer program can be divided into one or more modules, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules can be a series of computer program instruction segments capable of performing specific functions, which describe the execution process of the computer program in the ticket processing terminal device.

[0148] The document processing terminal equipment may be a desktop computer, laptop, handheld computer, or cloud server, etc. The document processing terminal equipment may include, but is not limited to, a processor and a memory.

[0149] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the ticket processing terminal equipment, connecting all parts of the equipment via various interfaces and lines.

[0150] The memory can be used to store the computer programs and / or modules. The processor implements various functions of the ticket processing terminal device by running or executing the computer programs and / or modules stored in the memory and calling data stored in the memory. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function, etc.; the data storage area may store data created based on the use of the mobile phone, etc. In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0151] Based on the above embodiments of the bill processing method, another embodiment of the present invention provides a storage medium, the storage medium including a stored computer program, wherein, when the computer program is running, the device where the storage medium is located controls the execution of the bill processing method of any embodiment of the present invention.

[0152] In this embodiment, the storage medium is a computer-readable storage medium, and the computer program includes computer program code, which can be in the form of source code, object code, executable file, or some intermediate form. The computer-readable medium can include any entity or device capable of carrying the computer program code, recording media, USB flash drive, portable hard drive, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium can be appropriately added or removed according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electrical carrier signals and telecommunication signals.

[0153] Furthermore, embodiments of the present invention also provide a computer program product, which, when run on a computer device, enables the computer device to implement the ticket processing method as described in any of the above embodiments.

[0154] In summary, the present invention provides a method, apparatus, device, medium, and product for processing invoices. This involves de-identifying the collected invoice information to obtain de-identified invoice information, sending this de-identified invoice information to a generative pre-trained language model for storage in a dedicated enterprise invoice model; sending voting request instructions from the enterprise to the generative pre-trained language model to filter out corresponding target invoice information based on these instructions; decrypting the target invoice information sent by the generative pre-trained language model to obtain decrypted target invoice information, and sending this decrypted target invoice information to the enterprise for payment settlement. Compared with existing technologies, the present invention improves the accuracy and efficiency of invoice information acquisition.

[0155] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and substitutions can be made without departing from the technical principles of the present invention, and these improvements and substitutions should also be considered within the scope of protection of the present invention.

Claims

1. A method for processing invoices, characterized in that, include: Collect invoice information from enterprises, de-identify the invoice information to obtain de-identified invoice information, and send the de-identified invoice information to a generative pre-trained language model so that the generative pre-trained language model stores the de-identified invoice information in a preset enterprise-specific invoice model; The system receives a ballot request instruction sent by the enterprise and sends the ballot request instruction to the generative pre-trained language model, so that the generative pre-trained language model can filter out the corresponding target bill information in the enterprise-specific bill model based on the ballot request instruction. Before sending the ballot request instruction to the generative pre-trained language model, the method further includes: de-sensitizing the ballot request instruction to obtain the de-sensitized ballot filtering conditions corresponding to the ballot request instruction. The process of desensitizing the ballot request instruction to obtain the corresponding desensitized ballot filtering conditions specifically includes: the ballot request instruction comprising ballot request content and ballot filtering conditions; obtaining sensitive ballot data from the ballot filtering conditions, encrypting the sensitive ballot data to obtain encrypted sensitive ballot data; setting a corresponding sensitive ballot key name for the sensitive ballot data, and using the encrypted sensitive ballot data as a sensitive ballot key value, assembling the sensitive ballot key name and the sensitive ballot key value as key-value pairs to obtain sensitive ballot data key-value pairs; obtaining non-sensitive ballot data from the ballot filtering conditions, setting a corresponding non-sensitive ballot key name for the non-sensitive ballot data, and using the non-sensitive ballot data as a non-sensitive ballot key value, assembling the non-sensitive ballot key name and the non-sensitive ballot key value as key-value pairs to obtain non-sensitive ballot data key-value pairs; and concatenating the sensitive ballot data key-value pairs and the non-sensitive ballot data key-value pairs to obtain the desensitized ballot filtering conditions. The system receives the target invoice information sent by the generative pre-trained language model, decrypts the target invoice information to obtain decrypted target invoice information, and sends the decrypted target invoice information to the enterprise so that the enterprise can make payment settlement based on the decrypted target invoice information.

2. The bill processing method as described in claim 1, characterized in that, Collect invoice information from enterprises, specifically including: Obtain the invoice image from the enterprise, extract the text content from the invoice image, and obtain structured invoice data; The structured invoice data is input into a preset invoice data template to obtain the invoice information corresponding to the invoice image.

3. The bill processing method as described in claim 1, characterized in that, The invoice information is de-identified to obtain de-identified invoice information, specifically including: The sensitive data in the invoice information is obtained, and the sensitive data is encrypted to obtain encrypted sensitive data. Set a corresponding sensitive key name for the sensitive data, and use the encrypted sensitive data as a sensitive key value. Concatenate the sensitive key name and the sensitive key value in the form of a key-value pair to obtain a sensitive data key-value pair. Obtain non-sensitive data from the invoice information, set corresponding non-sensitive key names for the non-sensitive data, and use the non-sensitive data as non-sensitive key values. Concatenate the non-sensitive key names and the non-sensitive key values ​​in the form of key-value pairs to obtain non-sensitive data key-value pairs. The sensitive data key-value pairs and the non-sensitive data key-value pairs are concatenated to obtain de-identified invoice information.

4. The bill processing method as described in claim 1, characterized in that, The generative pre-trained language model filters out the corresponding target bill information from the enterprise-specific bill model based on the ballot requirement instruction, specifically including: The ballot request instruction includes the ballot request content and the de-identified ballot filtering conditions; The generative pre-trained language model selects all first de-identified invoice information corresponding to the ballot requirement content from the enterprise-specific invoice model based on the ballot requirement content in the ballot requirement instruction, and filters all the first de-identified invoice information based on the de-identified ballot filtering conditions to obtain the target invoice information.

5. A bill processing method as described in claim 4, characterized in that, The generative pre-trained language model, based on the voting requirement content in the voting requirement instruction, selects all first de-identified invoice information corresponding to the voting requirement content from the enterprise-specific invoice model, specifically including: The generative pre-trained language model extracts keywords from the content of the ballot requirements to obtain ballot requirement keywords, and then retrieves all de-identified invoice information in the enterprise special invoice model based on the ballot requirement keywords. When the first desensitized invoice information matching the voting requirement keywords is retrieved, the first desensitized invoice information is saved, and the retrieval of other desensitized invoice information in the enterprise-specific invoice model continues based on the voting requirement keywords until all desensitized invoice information in the enterprise-specific invoice model has been retrieved, thus obtaining all the first desensitized invoice information.

6. The bill processing method as described in claim 1, characterized in that, The target invoice information is decrypted to obtain decrypted target invoice information, specifically including: The target invoice information includes target sensitive data key-value pairs and target non-sensitive data key-value pairs; Obtain the target sensitive key name and target sensitive key value from the target sensitive data key value pair, decrypt the target sensitive key value to obtain the decrypted target sensitive key value, and assemble the target sensitive key name and the decrypted target sensitive key value in the form of a key value pair to obtain the target decrypted sensitive data key value pair; The target sensitive data key-value pairs and the target non-sensitive data key-value pairs are concatenated to obtain the decrypted target ticket information.

7. A ticket processing device, characterized in that, include: The system includes a bill information storage module, a bill information filtering module, and a bill information decryption module. The invoice information storage module is used to collect invoice information from the enterprise, perform desensitization processing on the invoice information to obtain desensitized invoice information, and send the desensitized invoice information to the generative pre-trained language model so that the generative pre-trained language model stores the desensitized invoice information in a preset enterprise-specific invoice model. The invoice information filtering module is used to receive the ballot request instruction sent by the enterprise and send the ballot request instruction to the generative pre-trained language model, so that the generative pre-trained language model can filter out the corresponding target invoice information in the enterprise-specific invoice model based on the ballot request instruction. Before sending the ballot request instruction to the generative pre-trained language model, the method further includes: de-sensitizing the ballot request instruction to obtain the de-sensitized ballot filtering conditions corresponding to the ballot request instruction. The process of desensitizing the ballot request instruction to obtain the corresponding desensitized ballot filtering conditions specifically includes: the ballot request instruction comprising ballot request content and ballot filtering conditions; obtaining sensitive ballot data from the ballot filtering conditions, encrypting the sensitive ballot data to obtain encrypted sensitive ballot data; setting a corresponding sensitive ballot key name for the sensitive ballot data, and using the encrypted sensitive ballot data as a sensitive ballot key value, assembling the sensitive ballot key name and the sensitive ballot key value as key-value pairs to obtain sensitive ballot data key-value pairs; obtaining non-sensitive ballot data from the ballot filtering conditions, setting a corresponding non-sensitive ballot key name for the non-sensitive ballot data, and using the non-sensitive ballot data as a non-sensitive ballot key value, assembling the non-sensitive ballot key name and the non-sensitive ballot key value as key-value pairs to obtain non-sensitive ballot data key-value pairs; and concatenating the sensitive ballot data key-value pairs and the non-sensitive ballot data key-value pairs to obtain the desensitized ballot filtering conditions. The invoice information decryption module is used to receive the target invoice information sent by the generative pre-trained language model, decrypt the target invoice information to obtain decrypted target invoice information, and send the decrypted target invoice information to the enterprise so that the enterprise can make payment settlement based on the decrypted target invoice information.

8. A terminal device, characterized in that, It includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor executes the computer program to implement the ticket processing method as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored computer program, wherein, when the computer program is executed, it controls the device on which the computer-readable storage medium is located to perform the ticket processing method as described in any one of claims 1 to 6.

10. A computer program product, comprising a computer program or instructions, characterized in that, The computer program or instructions perform the bill processing method as described in any one of claims 1 to 6.

Images