An asset bill cancellation management method, electronic equipment and medium

Abstract

The application provides an asset bill cancellation management method, electronic equipment and medium, and relates to the technical field of data processing. First, bill-related data is acquired, covering structured and unstructured data, and unified cancellation data sets are generated by associating the data through key fields. Then, a multi-modal cancellation rule library is defined, including basic amount, time dimension and unstructured text enhancement rules, a multi-modal cancellation decision engine is constructed, the unified cancellation data sets are input, and cancellation results are output in combination with the rule library. The multi-modal cancellation decision engine is also provided with functions such as confidence score, offset cancellation, deposit cancellation and excess payment processing to further solve the cancellation error problem. Finally, a management report is generated based on the cancellation results. The application can comprehensively and accurately process asset bill cancellation by integrating multi-source data, flexibly formulating cancellation rules and using an intelligent decision engine, thereby providing strong support for enterprise management.

Description

TECHNICAL FIELD

[0001] The present application relates to an asset bill cancellation management method, an electronic device and a medium, belonging to the technical field of data processing. BACKGROUND

[0002] In the financial management work of enterprises, asset bill cancellation is a crucial and tedious task. Accurate and efficient bill cancellation can clearly reflect the financial situation of the enterprise, providing strong support for the enterprise's fund management and decision-making. With the continuous expansion of the business scale of enterprises and the increasing complexity of transaction data, the traditional manual cancellation method not only has low efficiency, but also is prone to errors, and has been difficult to meet the needs of modern enterprise financial management. Therefore, exploring an automated and intelligent asset bill cancellation method has become a research hotspot in the field of enterprise financial management.

[0003] For example, the Chinese invention patent with publication number CN119831759A discloses a method for automatic cancellation of accounts receivable based on artificial intelligence. This patent extracts the information features of accounts receivable and orders, uses a classification model to classify accounts receivable, and then performs matching, grouping and cancellation operations. However, this patent mainly focuses on the processing and matching of accounts receivable and order information features based on artificial intelligence, relies on order information and classification models, and has relatively simple cancellation rules, making it difficult to cope with complex business scenarios. Moreover, it lacks the fusion and comprehensive decision-making of multi-modal information, making it difficult to handle ambiguous or conflicting data.

[0004] Therefore, there is an urgent need for a method that can more comprehensively, flexibly and accurately handle asset bill cancellation by combining multi-source data and diversified cancellation rules. SUMMARY

[0005] In order to solve the problems existing in the prior art, the present application provides an asset bill cancellation management method, an electronic device and a medium.

[0006] The technical solution of the present application is as follows:

[0007] On the one hand, the present application provides an asset bill cancellation management method, which comprises:

[0008] Obtaining bill-related data, including structured data and unstructured data, and correlating the bill-related data through key fields to generate a unified cancellation data set;

[0009] Defining a multi-modal cancellation rule library, which includes basic amount matching rules for providing amount matching for cancellation decisions, time dimension rules for time determination, unstructured text enhancement rules for unstructured text analysis, and account consistency rules, wherein:

[0010] The basic amount matching rule is used to determine the cancellation state according to the matching relationship between the transaction amount and the bill amount, including marking as accurate cancellation when the transaction amount is equal to the bill amount, marking as error allowed cancellation when the absolute difference between the transaction amount and the bill amount is less than or equal to the product of the preset error rate and the bill amount, marking as multi-transaction cancellation of a single bill when the sum of the amounts of multiple transactions is equal to the amount of a single bill, and marking as single-transaction cancellation of multiple bills when the amount of a single transaction is equal to the sum of the amounts of multiple bills;

[0011] The time dimension rule is used to determine the cancellation time limit state according to the transaction date, the bill date and the contract agreement, including marking as on-time cancellation when the transaction date is within the bill date plus the contract agreed account period, marking as early cancellation when the transaction date is earlier than the bill date, and marking as overdue cancellation when the transaction date is later than the bill date plus the contract agreed account period;

[0012] The unstructured text enhancement rule is used to parse the contract terms and the bill note text to enhance the cancellation decision, including term-driven cancellation and bill note analysis, the term-driven cancellation is used to identify the discount, penalty or installment payment terms and calculate the amount and state to be cancelled accordingly, and the bill note analysis is used to identify the cancellation object, special purpose or dispute information in the note to assist the cancellation confirmation;

[0013] The account consistency rule is used to match the payment bank account and the collection bank account, and if the matching result is inconsistent, it is marked as a serious exception;

[0014] A multi-modal cancellation decision engine is constructed based on rule priority scheduling and multi-modal information fusion decision, the multi-modal cancellation decision engine takes a unified cancellation data set as input, applies the rules in the multi-modal cancellation rule library for cancellation decision calculation and state marking, and outputs the cancellation result, wherein the rule priority scheduling is to apply the rules in the order of account consistency rule>basic amount matching rule>time dimension rule>unstructured text enhancement rule;

[0015] Based on the cancellation result, a corresponding cancellation management report is generated to obtain bill-related data, including structured data and unstructured data, the bill-related data is associated through key fields to generate a unified cancellation data set;

[0016] A multi-modal cancellation rule library is defined, which includes rules for providing amount matching, time determination and unstructured text analysis for cancellation decision; a multi-modal cancellation decision engine is constructed based on rule priority scheduling and multi-modal information fusion decision, the multi-modal cancellation decision engine takes a unified cancellation data set as input, applies the rules in the multi-modal cancellation rule library for cancellation decision calculation and state marking, and outputs the cancellation result;

[0017] Based on the result of the cancellation, a corresponding cancellation management report is generated.

[0018] Preferably, the bill-related data is specifically obtained as follows:

[0019] Structured data of the lease contract is extracted and stored as a lease contract table; structured data of the bill is obtained and stored as a bill table; structured data of the bank statement is obtained and stored as a bank fund income and expenditure table;

[0020] The PDF text of the lease contract and the bill image are taken as unstructured data, and the unstructured data is subjected to character recognition; the recognized text data is subjected to natural language processing, and unstructured key information, including contract terms and bill notes, is extracted; the contract terms are stored in the lease contract table corresponding to the lease contract record, and the bill notes are stored in the bill table corresponding to the bill record.

[0021] Preferably, the bill-related data is associated by means of key fields, specifically by means of joint keys (contract ID, customer ID, payment bank account, collection bank account, and bill date) to establish cross-modal association, wherein:

[0022] The structured data of the lease contract includes contract ID, customer ID, contract specified amount, contract start date, contract end date, payment bank account, and collection bank account; the structured data of the bill includes bill ID, contract ID, customer ID, bill amount, bill date, payment bank account, and collection bank account; and the structured data of the bank statement includes transaction ID, transaction amount, transaction date, payment bank account, and collection bank account.

[0023] Based on the lease contract table, each row is traversed, for each lease contract record, the contract ID, customer ID, payment bank account, and collection bank account are searched in the bill table, and if the bill date is also within the range of the contract specified start date and end date, it is considered that the lease contract record and the bill record are matched successfully, the matched lease contract record and bill record are integrated together, and lease contract-bill association data is obtained.

[0024] In the bank fund income and expenditure table, find the transaction record in which the payment bank account and the collection bank account in the lease contract-bill association data are consistent, find the bank transaction record in which the payment bank account and the collection bank account in the lease contract-bill association data are consistent in the bank fund income and expenditure table, and the transaction date is within the contract start date and end date range, it is considered that the lease contract-bill association data is matched successfully with the bank transaction record, the successfully associated lease contract record, bill record and bank transaction record are obtained, the successfully associated lease contract record, bill record, bank transaction record are integrated into a single association record, and stored in the unified cancellation data set.

[0025] Preferably, the multi-modal cancellation rule library comprises a basic amount matching rule, a time dimension rule and an unstructured text enhancement rule, wherein:

[0026] The basic amount matching rule is used to determine the cancellation state according to the matching relationship between the transaction amount and the bill amount, including marking as accurate cancellation when the transaction amount is equal to the bill amount; marking as error allowed cancellation when the absolute difference between the transaction amount and the bill amount is less than or equal to the product of the preset error rate and the bill amount; marking as multi-transaction cancellation of a single bill when the sum of the amounts of multiple transactions is equal to the amount of a single bill; marking as single-transaction cancellation of multiple bills when the amount of a single transaction is equal to the sum of the amounts of multiple bills.

[0027] The time dimension rule is used to determine the cancellation time limit state according to the transaction date, the bill date and the contract agreement, including marking as on-time cancellation when the transaction date is within the bill date plus the contract agreed account period; marking as early cancellation when the transaction date is earlier than the bill date; marking as overdue cancellation when the transaction date is later than the bill date plus the contract agreed account period.

[0028] The unstructured text enhancement rule is used to parse the contract terms and bill note text to enhance the cancellation decision, including term-driven cancellation and bill note analysis, the term-driven cancellation is used to identify discount, penalty or installment payment terms and calculate the amount and state of cancellation accordingly; the bill note analysis is used to identify the cancellation object, special purpose or dispute information in the note to assist in cancellation confirmation.

[0029] Preferably, the multi-modal cancellation decision engine comprises a single association record collector, a rule matching scheduler, a rule execution processor, a multi-modal information fusion arbitrator and a cancellation state and information outputter, wherein:

[0030] The single association record collector is used to obtain the single association record of the unified cancellation data set, including contract structured data, bill structured data, bank transaction structured data and associated unstructured key information;

[0031] The rule matching scheduler is configured to apply the multi-modal cancellation rules according to a preset multi-modal cancellation rule priority order; the rule execution processor is configured to perform an amount calculation or date comparison operation corresponding to the multi-modal cancellation rules and update the cancellation state label when the multi-modal cancellation rules meet a triggering condition;

[0032] The multi-modal information fusion arbitrator is configured to make a decision according to the associated unstructured key information when there is a conflict or ambiguity in the structured data during the rule execution process, and determine the final cancellation state;

[0033] The cancellation state and information outputter is configured to output the cancellation result of a single associated record.

[0034] Preferably, the multi-modal cancellation decision engine further comprises a confidence score calculator configured to perform a cancellation confidence score for each cancellation result, and evaluate the reliability of the cancellation result based on the cancellation confidence score, wherein the score of the cancellation confidence score is determined according to four factors, namely the definiteness of rule matching, the consistency of multi-modal information, the confidence of unstructured text analysis, and the accuracy of historical similar cancellation, and the cancellation confidence score is obtained by weighted calculation according to preset factor weights and factor score determination standards. If the cancellation confidence score is lower than a preset confidence threshold, the cancellation result is marked as pending review, prompting manual further confirmation.

[0035] Preferably, the multi-modal cancellation decision engine further comprises a deduction cancellation scheme generator, a deposit cancellation manager, and an excess payment processing coordinator, wherein:

[0036] The deduction cancellation scheme generator is configured to identify multiple uncancelled bills and multiple uncancelled deposits of the same customer, and when the total amount of uncancelled deposits of the customer is greater than or equal to the total amount of uncancelled bills, start a deduction cancellation process, and automatically generate a deduction scheme according to a preset business rule, and mark the cancellation state of related bills and transactions based on the deduction scheme;

[0037] The deposit cancellation manager is configured to automatically identify and mark transaction records as deposits, and record the corresponding transaction amount, and when the transaction record generates a positive bill in the subsequent transaction process, automatically check the marked deposit and apply it to the cancellation operation, and update the balance of the deposit after the cancellation operation is completed;

[0038] The excess payment processing coordinator is configured to automatically calculate the difference of each cancellation transaction after the cancellation is completed, and mark it as excess payment if the difference is positive; and determine the processing method for the excess payment according to the contract terms and enterprise regulations.

[0039] In another aspect, the present application also provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the asset bill cancellation management method of the present application when executing the program.

[0040] In another aspect, the present application also provides a computer readable storage medium having a computer program stored thereon, wherein the program is executable on a processor to implement the asset bill cancellation management method of the present application.

[0041] The present application has the following beneficial effects:

[0042] 1. The present application is an asset bill cancellation management method, electronic device and medium, which obtains bill-related data covering structured and unstructured data, establishes cross-modal association with the aid of key fields, integrates lease contracts, bills and bank transaction records into a unified cancellation data set, processes unstructured data using OCR technology and supplements it to structured records, breaks down data source barriers, improves data accuracy and availability, and provides a reliable basis for cancellation.

[0043] 2. The present application is an asset bill cancellation management method, electronic device and medium, which defines a multi-modal cancellation rule library containing various amount matching, time dimension and unstructured text rules to meet the needs of different cancellation scenarios.

[0044] 3. The present application is an asset bill cancellation management method, electronic device and medium, which constructs a multi-modal cancellation decision engine whose components work cooperatively. In the present application, the multi-modal information fusion arbitrator in the component prioritizes the interpretation of ambiguous amounts or conflicting dates (such as fee deduction notes, holiday extension corrections, and overdue dates) based on unstructured text (such as contract terms and bill notes), solving the cancellation error problem caused by single structured data in traditional systems. Based on the four-dimensional weight confidence scoring mechanism in the confidence score calculator in the component, low confidence results are automatically identified, significantly reducing manual review costs and improving the detection rate of abnormal cancellation. BRIEF DESCRIPTION OF DRAWINGS

[0045] Figure 1 The present application is a method flowchart. DETAILED DESCRIPTION

[0046] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of the present application.

[0047] It should be understood that the step numbers used herein are only for the convenience of description, and are not intended to limit the execution sequence of the steps.

[0048] It should be understood that the terms used in the present application are only for the purpose of describing specific embodiments and are not intended to limit the present application. As used in the present application and the appended claims, "a", "an", and "the" in singular form are intended to include plural forms unless the context clearly indicates otherwise.

[0049] The terms "comprise" and "include" indicate the presence of the described features, integers, steps, operations, elements, and / or components, but do not exclude one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0050] The term "and / or" means any combination of one or more of the associated listed items and all possible combinations thereof, and includes these combinations.

[0051] Embodiment 1:

[0052] Referring to Figure 1 The present embodiment provides an asset bill cancellation management method, comprising the following steps:

[0053] S1, obtaining bill-related data including structured data and unstructured data from the enterprise's lease contract library, bill library and bank system, associating the bill-related data through key fields to generate a unified cancellation data set;

[0054] S11, obtaining structured data specifically comprises:

[0055] S111, extracting structured data of the lease contract from the enterprise's lease contract library, including contract ID, customer ID, contract specified amount, contract start date, contract end date, payment bank account and collection bank account, storing the structured data of the lease contract in the form of a table as a lease contract table, each row representing a lease contract record;

[0056] The extracted structured data of the lease contract is preliminarily cleaned, including removing duplicate records and format error data, to ensure the accuracy and integrity of the data;

[0057] S112, obtaining structured data of the bill from the enterprise's bill library, including bill ID, contract ID, customer ID, bill amount, bill date, payment bank account and collection bank account, the structured data of the bill is also stored in the form of a table as a bill table, each record corresponds to a bill;

[0058] The structured data of the bill is subjected to data quality inspection to check whether there are missing values and abnormal values, and the missing structured key information is supplemented or marked, for example, when the bill amount is detected to be missing, a chain equation multiple imputation algorithm is used to fill in the missing values, which iteratively estimates the missing values through known information (such as customer historical amount level and contract period) to be expressed in formula as:

[0059] ;

[0060] wherein, is the actual missing value to be filled in after the i th iteration, is the observation value of other complete known information in the same bill, is the estimated missing value to be filled in after the i th iteration, is the coefficient matrix based on linear regression training; For the convenience of understanding, an example is given: for example, in the equipment rental scenario of an enterprise, the amount data of a bill is lost due to system synchronization failure, and the algorithm successfully fills in 12240 according to the average rent of the enterprise in the past 12 periods, which is 12000 and the standard deviation, which is 600;

[0061] In order to facilitate understanding, an example is given: for example, in the equipment rental scenario of an enterprise, the amount data of a bill is lost due to system synchronization failure, and the algorithm successfully fills in 12240 according to the average rent of the enterprise in the past 12 periods, which is 12000 and the standard deviation, which is 600;

[0062] S113, through the interface or data import function with the bank system, the structured data of the bank statement is obtained, including transaction ID, transaction amount, transaction date, payment bank account and collection bank account, the structured data of the above bank statement records the financial transactions between the enterprise and the customer, and the structured data of the bank statement is also stored in the form of a table as a bank financial statement table;

[0063] The structured data of the bank statement is subjected to data conversion and format unification, so as to match the data format of the rental contract library and the bill library, and facilitate subsequent data association operation.

[0064] S12, obtaining unstructured data, specifically:

[0065] Obtaining the PDF text of the rental contract in the enterprise rental contract library and bill library and the scanned bill image as unstructured data;

[0066] The bill image is denoised using a denoising algorithm such as non-local mean filtering algorithm to improve the clarity of the image; and the geometric correction of the tilted bill image is carried out, specifically when the bill image is detected to have a clockwise offset through Hough transform, the specific angle of the clockwise offset is extracted from the result of the Hough transform ; the horizontal and vertical translation amounts and ​; affine transformation matrix is constructed based on specific angle and translation amount , which is expressed in formula as

[0067] ;

[0068] Each pixel in the bill image is processed, specifically:

[0069] The pixels of the bill image are traversed one by one, and for each pixel, its new position after transformation is calculated according to the affine transformation matrix; during the calculation process, when the new position exceeds the boundary of the original data, only the pixels within the original boundary are retained, and the part exceeding the boundary is directly cropped;

[0070] After affine transformation, the new position of the pixel may not be an integer coordinate, but the pixel position is usually an integer. In order to determine the color value of the pixel at the non-integer position, the embodiment performs bilinear interpolation processing, based on the four nearest pixels around the non-integer position, and calculates the corresponding pixel color value by weighted average, so that the corrected data looks smoother;

[0071] By performing the above geometric correction processing on each pixel in the bill image, it is restored to the normal viewing angle, which is convenient for subsequent OCR recognition;

[0072] The optical character recognition (OCR) technology is used to recognize the text in the PDF contract text and the bill image after geometric correction, and convert the text in the document and image into editable text data. The natural language processing is performed on the text data after OCR recognition, and the unstructured key information is extracted, including the contract terms and the bill notes. Preferably, the structured data of the lease contract and the structured data of the bill can also be extracted from the PDF contract text and the bill image after geometric correction. Since the structured data is repeated, it will not be described here again. Based on the structured data obtained by recognition, the unstructured key information is stored in the corresponding table, including storing the contract terms in the lease contract table corresponding to the lease contract record, and storing the bill notes in the bill table corresponding to the bill record.

[0073] Preferably, the date format and the amount format in the table in the embodiment are unified formats, for example, the date is unified as "YYYY - MM - DD" format.

[0074] S12, the bill related data is data associated through the key field, specifically:

[0075] The cross-modal association is established through the joint key (Contract ID, customer ID, payment bank account, collection bank account, bill date) to establish cross-modal association A combination identifier for uniquely identifying a data record. In the corresponding leasing business in this embodiment, different types of data come from different data sources and have different formats and structures. By using the joint key, these scattered data can be associated:

[0076] S121, the lease contract table is associated with the billing table:

[0077] Since the lease contract table and the billing table both have contract ID, customer ID, payment bank account, collection bank account, and billing date (which can be considered as the date within the contract validity period in the lease contract table), this embodiment is associated by accurate matching, specifically:

[0078] Based on the lease contract table, it is traversed row by row. For each lease contract record, the contract ID, customer ID, payment bank account, and collection bank account are found in the billing table. If the billing date is also within the range of the start date and end date specified by the contract, it is considered that the lease contract record and the billing record match successfully.

[0079] For example, there is a record in the lease contract table, the contract ID is "C001", the customer ID is "U001", the payment bank account is "A123", the collection bank account is "B456", the contract start date is "2025-01-01", and the contract end date is "2025-12-31". Find the billing record with contract ID "C001", customer ID "U001", payment bank account "A123", collection bank account "B456", and billing date between "2025-01-01" and "2025-12-31" in the billing table; integrate the matching lease contract record and the billing record to obtain lease contract-billing association data;

[0080] Further, if some lease contract records are not found after accurate matching, fuzzy matching is performed:

[0081] For example, for the payment bank account and the collection bank account, there may be slight differences in entry, such as one or more spaces, etc. After simple cleaning (removing spaces, special characters, etc.) of the payment bank account and the collection bank account fields, matching is attempted again; for the billing date, according to the actual business situation, the date range is relaxed, and the corresponding data within the relaxed date range is found; the data obtained by fuzzy matching is combined with the data obtained by accurate matching.

[0082] S122, the lease contract-billing association data is associated with the bank fund collection and disbursement table:

[0083] Since the bank fund income and expenditure form does not have the contract ID, the customer ID and the billing date, the embodiment uses the payer bank account and the payee bank account for association:

[0084] In the bank fund income and expenditure form, find the transaction record whose payer bank account and payee bank account are consistent with those in the lease contract-billing association data. For example, the payer bank account in the lease contract-billing association data is “A123” and the payee bank account is “B456”. In the bank fund income and expenditure form, find the bank transaction record whose payer bank account is “A123” and payee bank account is “B456”. Based on the bank transaction record, further compare the relationship between the transaction amount and the billing amount (or the contract specified amount). If the transaction amount is equal to the billing amount or within a certain error range (the error range is set according to business conditions, such as ±5% in the embodiment), and the transaction date is within the contract start date and end date range, it is considered that the lease contract-billing association data matches the bank transaction record successfully, and the successfully associated lease contract record, billing record and bank transaction record are obtained.

[0085] The successfully associated lease contract record, billing record and bank transaction record are integrated into a complete record and stored in the unified cancellation data set. This record contains detailed information of the contract, billing and transaction, which can be used for subsequent financial cancellation and business analysis.

[0086] Preferably, the lease contract record, billing record and bank transaction record without matching items are marked as abnormal records, and the reasons for unassociated are analyzed, including data entry errors or system synchronization problems, etc. The abnormal records are further manually checked and processed to ensure the completeness and accuracy of the data.

[0087] Through the above steps, the embodiment completes the process from data association to generation of the unified cancellation data set, and provides accurate and complete data basis for the cancellation operation of the lease business.

[0088] S2, define a multi-modal cancellation rule library and construct a multi-modal cancellation decision engine. The multi-modal cancellation decision engine takes the unified cancellation data set as input, combines the multi-modal cancellation rule library, and outputs the cancellation result.

[0089] S21, the multi-modal cancellation rule library includes basic amount matching rules, time dimension rules, unstructured text enhancement rules and account consistency rules, wherein:

[0090] S211, the base amount matching rule includes accurate verification, error allowed verification, multi-transaction verification single bill and single-transaction verification multi-bills, when the transaction amount is completely consistent with the bill amount, it is marked as accurate verification; when the absolute value of the difference between the transaction amount and the bill amount is less than or equal to the product of the bill amount and a preset error rate, it is marked as error allowed verification, at this time the difference between the transaction amount and the bill amount needs to be recorded, and the preset error rate is set according to the business tolerance; when the sum of the amounts of multiple transactions under the same bill ID is equal to the bill amount, it is marked as multi-transaction verification single bill; when multiple bills correspond to the same transaction ID, and the transaction amount is equal to the sum of the bill amounts, the bills are marked as single-transaction verification multi-bills (at this time it is a shared transaction);

[0091] S212, the time dimension rule includes on-time verification, early verification and overdue verification, when the transaction date is within the bill date plus the contract agreed billing period, it is marked as on-time verification, which is set in this embodiment because for the payer, it often needs a certain time to prepare funds after receiving the bill, for example, after an enterprise purchases goods or services, its sales return, fund allocation and the like need a process, and the contract agreed billing period is a buffer period for the payer to arrange funds reasonably, assuming that a small enterprise receives a 500,000 yuan bill for raw material purchase, it may need to wait for its product to be sold and the funds to be returned before paying the bill, if the contract agreed billing period is 30 days, then within 30 days from the bill date, the payer has enough time to complete the fund raising and allocation and ensure the normal operation of the business; when the transaction date is earlier than the bill date, it is marked as early verification, and the contract terms need to be associated to see whether there are discount or penalty rules; when the transaction date is later than the bill date plus the contract agreed billing period, it is marked as overdue verification, and the contract terms need to be associated to calculate the late fees or penalty interest;

[0092] S213, the unstructured text enhancement rule includes clause-driven verification and bill note analysis, wherein the clause-driven verification further includes discount or exemption identification, penalty or late fee calculation and installment or special payment plan, wherein:

[0093] The discount or exemption identification is specifically to analyze the associated contract terms (such as “2% discount for early payment”) through NLP, when early verification is detected, the early verification amount to be verified is automatically calculated, which is expressed in the formula as early verification amount to be verified = bill amount * (1 - discount rate), and the early verification amount to be verified is compared with the transaction amount, if they match, it is marked as complete verification (including discount); if the transaction amount is the bill amount, the difference is automatically marked as discount to be returned or customer credit;

[0094] The calculation of the penalty or late payment fee is specifically performed by analyzing the contract terms through NLP (e.g., "a late payment fee of 0.05% per day overdue") and automatically calculating the total amount to be written off for the overdue write-off when an overdue write-off is detected, using the formula: total amount to be written off for the overdue write-off = bill amount + late payment fee. This amount is then compared with the transaction amount. If the transaction amount is equal to the total amount to be written off for the overdue write-off, it is marked as a full write-off (including late payment fee); if the transaction amount is only equal to the bill amount, it is marked as a partial write-off (owing late payment fee); if the transaction amount is greater than the total amount to be written off for the overdue write-off, it is marked as a full write-off and an overpayment record is automatically generated;

[0095] The installment or special payment plan specifically identifies special payment terms in the contract (e.g., "30% down payment, 60% after acceptance, and 10% final payment after the warranty period") and intelligently determines the current write-off stage based on the bill amount, bill date, transaction amount, and transaction date. The write-off status is then determined accordingly, including first-stage write-off, first pending write-off, mid-term write-off, mid-term pending write-off, final-stage write-off, and final pending write-off.

[0096] Specifically, the bill note parsing involves using NLP to identify the intent of bill notes (e.g., "Payment of rent for May 2025 for contract C001," "Payment of final payment for customer U001"), including assisting in confirming the write-off object (contract ID, bill ID, customer ID), especially providing strong evidence for fuzzy matching or partial write-offs; identifying special purposes (e.g., "deposit," "security deposit," "advance payment"), triggering different write-off logic processing; or identifying dispute information (e.g., "deduction of XXX yuan for damage compensation"), marking it as a dispute write-off, and associating the deduction amount.

[0097] S214: Since the payment bank account and the receiving bank account have been matched during the association in S1, the account consistency rule is used as the final check here. If there is any inconsistency, a serious exception is marked.

[0098] S22. The multimodal write-off decision engine takes a single associated record of the unified write-off dataset as input, combines it with the multimodal write-off rule base, and outputs detailed write-off status and operation results. Its core lies in rule priority scheduling and multimodal information fusion decision-making. The multimodal write-off decision engine includes a single associated record collector, a rule matching scheduler, a rule execution processor, a multimodal information fusion arbitrator, a confidence score calculator, and a write-off status and information outputter, wherein:

[0099] S221. The single associated record collector is used to obtain a single associated record of the unified write-off data set, including contract structured data, bill structured data, bank transaction structured data, and associated unstructured key information;

[0100] S222, the rule matching scheduler is used to apply rules according to a preset multi-modal verification rule priority order, in the embodiment, the preset multi-modal verification rule priority is account consistency rule > basic amount matching rule > time dimension rule > unstructured text enhancement rule; when the rule trigger condition is met, the corresponding verification calculation and state marking are performed; for example, when a single associated record is received, the account consistency rule is first checked, if the rule trigger condition is met, the subsequent basic amount matching rule is continued to be executed; if it is not met, it is directly determined as an exception-account inconsistency;

[0101] S223, when the trigger condition of a certain rule is met, the rule execution processor is used to perform corresponding verification calculation and state marking, for example, in the basic amount matching rule, if the transaction amount and the bill amount are equal, the record is marked as accurate verification, and corresponding amount calculation and state update are performed;

[0102] S224, the multi-modal information fusion arbitrator comprehensively considers various types of data, solves the contradiction between the data, makes a more reasonable decision, and determines the final verification state, specifically, when the structured data (amount or date) is ambiguous or conflicting, the unstructured key information (contract terms, bill notes) is preferred for interpretation and arbitration; for example, a transaction amount is slightly lower than the bill amount, but the bill note clearly states "deduct water and electricity fee XXX yuan", the multi-modal information fusion arbitrator will apply part of the verification rules and record the deduction details; or the transaction date is slightly later than the bill date, but the contract terms have "holiday extension" provisions and the transaction date is within the contract period, then mark the on-time verification instead of the overdue verification;

[0103] S225, the confidence score calculator is used to score the verification confidence (0-1) for each verification result, the score is based on the definiteness of rule matching, the consistency of multi-modal information, the confidence of unstructured text analysis, and the accuracy rate of historical similar verification, wherein:

[0104] The definiteness of rule matching is specifically that if the rule matching is very clear, for example, in the basic amount matching rule, the transaction amount and the bill amount are completely equal, without any error or ambiguity, in this case, the definiteness of rule matching is high, which positively contributes to the confidence score; when the rule matching has certain ambiguity, such as the transaction amount is close to the bill amount within the allowed error range, but not completely consistent, or the date is in a critical state in the time rule, the ambiguous matching will reduce the definiteness of rule matching, thereby reducing the confidence score;

[0105] The consistency of the multi-modal information is specifically that structured data (such as contract amount, transaction date) and unstructured text information (such as contract terms, bill notes) are highly consistent, without contradictions or conflicts, for example, the payment method, amount and bank transaction record and bill notes specified in the contract are completely consistent, which indicates that the multi-modal information consistency is high, and the confidence score is improved; if there are contradictions between structured data and unstructured key information, such as the payment date specified in the contract and the bank transaction date differ greatly, or the transaction amount does not match the amount explained in the bill notes, such inconsistency will reduce the consistency of multi-modal information, and in turn reduce the confidence score.

[0106] The confidence of unstructured text analysis is specifically that when using natural language processing (NLP) technology to analyze unstructured text (such as contract terms, bill notes), if the NLP model gives a high confidence in the analysis result, for example, the model accurately understands the semantics of the text and can clearly interpret the key information in the text, then the confidence score of the entire verification result will be improved; if the NLP model has difficulty in analyzing unstructured text, the confidence of the analysis result is low, for example, the understanding of some ambiguous expressions is uncertain, or the key information cannot be accurately extracted, which will lead to a decrease in the confidence score.

[0107] The accuracy of historical similar verification is specifically that if the accuracy of similar verification in the past is high, it indicates that the processing method and result of the current verification scene have high reliability, for example, the results of processing the same customer and the same type of business in the past are relatively accurate, then the confidence score of the current similar verification will be relatively high; on the contrary, if the accuracy of historical similar verification is low, it indicates that this type of verification may have some complex factors or places prone to errors, and the confidence score of the current verification result will be correspondingly reduced.

[0108] The confidence score calculator considers the above-mentioned factors, quantifies each factor into a certain weight, and then calculates the final confidence score by weighted average method. For the convenience of understanding, an example is given as follows:

[0109] The confidence score is calculated according to the four factors of rule matching definiteness, multi-modal information consistency, unstructured text analysis confidence and historical similar verification accuracy, and the corresponding weight of each factor is set as shown in Table 1:

[0110] Table 1 Confidence score basis weight table

[0111]

[0112] Further, the scores of each scoring basis are shown in Tables 2-4:

[0113] Table 2. Explicitness criterion score table for rule matching

[0114]

[0115] Table 3. Consistency criterion score table for multi-modal information

[0116]

[0117] Table 4. Accuracy criterion score table for historical similar verifications

[0118]

[0119] Since the unstructured text analysis is usually completed by an NLP model, which itself outputs a confidence score, this embodiment directly uses this score as the confidence criterion score for structured text analysis, but needs to be converted to the 0-1 interval. Assuming that the confidence output by the NLP model is (valued in the range of 0-100), then the score of unstructured text analysis is

[0120] Based on the weight and score calculation of the above four factors, the confidence score is obtained , which is expressed in the formula as:

[0121]

[0122] In the formula, , , and are the weights of explicitness of rule matching, consistency of multi-modal information, confidence of unstructured text analysis, and accuracy of historical similar verifications, respectively; , , and are the scores of explicitness of rule matching, consistency of multi-modal information, confidence of unstructured text analysis, and accuracy of historical similar verifications, respectively. Assuming that the scores of each factor of a certain verification result are: since the transaction amount and the bill amount are exactly equal, it is an exact match, with a score of 0.9; the structured data and the unstructured text information are completely consistent, with a score of 0.95; the confidence output by the NLP model is 80%, with a score of 0.8 after conversion; the accuracy of historical similar verification is 85%, with a score of 0.8; then the confidence score of the verification result is .

[0123] It should be noted that the confidence score calculation described in this embodiment is only an example, and the specific calculation method is different according to the business needs and system design of the enterprise. Through the confidence score, the reliability of the verification result can be evaluated.​​

[0124] For low-confidence results (i.e. results below the confidence threshold), marked for review, prompting manual further confirmation, and for high-confidence results (i.e. results below the confidence threshold), the result is considered reliable, for example, the confidence score of the result of the cancellation The confidence threshold is set to 0.5 in this embodiment, and the result of the cancellation is reliable.

[0125] S226, the cancellation state and information output device for outputting the cancellation result (such as complete cancellation, partial cancellation (arrears XXX), overdue cancellation (late payment XXX), etc.) of a single associated record, and the corresponding cancellation related data, including cancellation amount, cancellation related date, cancellation basis rule and contract clause index, etc. These information provides comprehensive cancellation result details for users.

[0126] S227, the multi-modal cancellation decision engine aims to handle various business scenarios related to cancellation, to achieve accurate and efficient cancellation management. Simple one-to-one cancellation scenarios are relatively basic cases, and steps S221-S226 can solve them. However, in actual business, there are many complex cancellation situations. In order to adapt to the processing of complex cancellation scenarios, the multi-modal cancellation decision engine of the embodiment further includes a offset cancellation scheme generator, a deposit cancellation manager and an overpayment processing coordinator, wherein:

[0127] The offset cancellation scheme generator is used to identify multiple un-canceled bills and multiple un-canceled deposits under the same customer (customer ID). Only when the total amount of un-canceled deposits of the customer is greater than or equal to the total amount of un-canceled bills, the offset cancellation process will be started. On the basis of meeting the above conditions, the offset scheme is automatically generated according to the business rules (such as first-in-first-out FIFO, specified bills, etc.). For example, if the first-in-first-out rule is used, the offset cancellation scheme generator preferentially uses the earliest deposit to cancel the earliest bill;

[0128] According to the offset cancellation scheme generated by the offset cancellation scheme generator, the cancellation status of the related bills and transactions is marked to ensure the accuracy and consistency of the financial data.

[0129] Preferably, the offset cancellation scheme generator can also use the debt settlement order clause in the contract (extracted by NLP) to optimize the generated offset scheme, so that the cancellation scheme is more in line with the actual business situation.

[0130] The deposit cancellation manager can automatically identify transaction records marked as "prepayment" or "deposit", record the corresponding transaction amount and related information, and when the transaction record subsequently generates a positive bill, automatically check the available prepayment or deposit and apply it to cancellation, for example, the customer has pre-paid a deposit of 1000 yuan, and subsequently generates a bill of 800 yuan, the deposit cancellation manager will automatically deduct 800 yuan from the deposit for cancellation; after completing the cancellation operation, update the balance of the prepayment or deposit in a timely manner, such as in the above example, the balance of the deposit will be updated to 200 yuan after cancellation.

[0131] The overpayment processing coordinator is used to automatically calculate the difference of each cancellation transaction after cancellation is completed, and if the difference is positive, i.e. the customer has overpaid, it is marked as overpayment; according to the contract terms and enterprise regulations, determine the processing method for the overpayment, in this embodiment, there are two processing methods: transfer the overpayment to the customer's credit limit, which can be directly used by the customer to pay in subsequent transactions; if the contract stipulates or the customer requires, return the overpayment to the customer, the overpayment processing coordinator will automatically trigger the refund process, including generating a refund application, interfacing with the financial system to return funds, etc.

[0132] S3, generating a cancellation report based on the cancellation result:

[0133] In the case of ensuring compatibility with data interaction and interface of other system modules of the enterprise at present, ensuring the stability and smoothness of the entire system, the multi-modal cancellation decision engine constructed is integrated into the existing financial management system or related business system of the enterprise, and a detailed cancellation management report is generated regularly, the cancellation management report adds the fields of cancellation status, cancellation amount, difference and reason, cancellation date, rules / terms, confidence, pending review flag, etc. output by the multi-modal cancellation decision engine based on the unified cancellation data set, so as to facilitate the enterprise to understand the specific cancellation situation and make targeted management decisions.

[0134] Embodiment 2:

[0135] The embodiment provides an electronic device, including a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to realize the asset bill cancellation management method of any one of the embodiments.

[0136] Embodiment 3:

[0137] The embodiment provides a computer readable storage medium, which stores a computer program, and the program is executed by a processor to realize the asset bill cancellation management method of any one of the embodiments.

[0138] In the embodiments of the present application, “at least one” means one or more, and “multiple” means two or more. “And / or” describes the association relationship of the associated objects, which means that there can be three kinds of relationships, for example, A and / or B can represent the cases of A alone, A and B together, and B alone. Wherein A and B can be singular or plural. The character “ / ” generally represents an “or” relationship between the front and rear associated objects. “At least one of the following” and the like means any combination of these items, including any combination of single or multiple items. For example, at least one of a, b and c can represent: a, b, c, a and b, a and c, b and c, or a and b and c, wherein a, b, and c can be single or multiple.

[0139] Those skilled in the art can appreciate that the units and algorithm steps described in the embodiments disclosed herein can be realized by electronic hardware, computer software and a combination of electronic hardware and computer software. Whether the functions are realized in hardware or software depends on the specific application and design constraints of the technical solution. A person skilled in the art can use different methods to realize the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0140] Those skilled in the art can clearly understand that, for the convenience and brevity of the description, the specific working processes of the above-described system, device and unit can refer to the corresponding processes in the foregoing method embodiments, which will not be described here.

[0141] In several embodiments provided by the present application, any function realized in the form of a software function unit and sold or used as an independent product can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application essentially or the parts that make contributions to the prior art or parts of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium, and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the embodiments of the present application. The foregoing storage medium includes: a U disk, a mobile hard disk, a read-only memory (Read-Only Memory, ROM for short), a random access memory (Random Access Memory, RAM for short), a magnetic disk or an optical disk, and various media that can store program codes.

[0142] The above is only an embodiment of the present application, and does not limit the patent scope of the present application. Any equivalent structure or equivalent flow transformation made by using the content of the specification and drawings, or direct or indirect application in other related technical fields, is also included in the patent protection scope of the present application.

Claims

1. An asset bill write-off management method, characterized in that: The method comprises: Obtain bill-related data, including structured and unstructured data, and associate the bill-related data using key fields to generate a unified write-off data set; A multimodal write-off rule base is defined. The multimodal write-off rule base includes a basic amount matching rule for providing amount matching for write-off decisions, a time dimension rule for time determination, an unstructured text enhancement rule for unstructured text parsing, and an account consistency rule, wherein: The basic amount matching rule is used to determine the write-off status based on the matching relationship between the transaction amount and the invoice amount, including marking as accurate write-off when the transaction amount is equal to the invoice amount; marking as error-allowed write-off when the absolute difference between the transaction amount and the invoice amount is less than or equal to the product of a preset error rate and the invoice amount; marking as multiple transactions writing off a single invoice when the sum of multiple transaction amounts is equal to the amount of a single invoice; and marking as single transaction writing off multiple invoices when the single transaction amount is equal to the sum of multiple invoice amounts. The time dimension rule is used to determine the write-off time status based on the transaction date, billing date, and contractual agreement. It includes marking the transaction date as on-time write-off when it falls within the billing date plus the contractually agreed payment period; marking the transaction date as early write-off when it is earlier than the billing date; and marking the transaction date as overdue write-off when it is later than the billing date plus the contractually agreed payment period. The unstructured text enhancement rules are used to parse contract terms and bill notes to enhance write-off decisions, including clause-driven write-offs and bill note parsing. Clause-driven write-offs are used to identify discounts, penalties, or installment payment terms and calculate the write-off amount and status accordingly. Bill note parsing is used to identify write-off objects, special uses, or dispute information in the notes to assist in write-off confirmation. The account consistency rule is used to match the paying bank account and the receiving bank account, and if the matching results are inconsistent, it is marked as a serious anomaly; A multimodal write-off decision engine is constructed based on rule priority scheduling and multimodal information fusion decision-making. The multimodal write-off decision engine takes a unified write-off dataset as input, applies the rules in the multimodal write-off rule library to perform write-off decision calculations and status marking, and outputs write-off results. The rule priority scheduling applies rules in the order of account consistency rule > basic amount matching rule > time dimension rule > unstructured text enhancement rule. Generate corresponding write-off management reports based on the write-off results.

2. The asset bill write-off management method according to claim 1, characterized in that: Obtain bill-related data specifically as follows: Extract the structured data of the lease contract and store it as a lease contract form; obtain the structured data of the bill and store it as a bill form; obtain the structured data of the bank statement and store it as a bank funds receipt and payment form; The PDF text of the lease contract and the bill image are treated as unstructured data, and text recognition is performed on the unstructured data. The recognized text data is subjected to natural language processing to extract unstructured key information, including contract terms and bill notes. The contract terms are stored in the lease contract record corresponding to the lease contract form, and the bill notes are stored in the bill record corresponding to the bill form.

3. The asset bill write-off management method according to claim 2, characterized in that: The bill-related data is associated through key fields, specifically through joint keys (Contract ID, Customer ID, Paying Bank Account, Receiving Bank Account, Billing Date) establishes a cross-modal association, where: The structured data of the lease contract includes the contract ID, customer ID, contract amount, contract start date, contract end date, paying bank account, and receiving bank account; the structured data of the bill includes the bill ID, contract ID, customer ID, bill amount, bill date, paying bank account, and receiving bank account; the structured data of the bank statement includes the transaction ID, transaction amount, transaction date, paying bank account, and receiving bank account; Based on the lease contract table, we traverse each row one by one. For each lease contract record, we search the billing table for a billing record with the same contract ID, customer ID, paying bank account, and receiving bank account. If the billing date is within the start and end dates specified in the contract, the lease contract record is considered to be a match with the billing record. The matching lease contract record and billing record are then integrated to obtain the lease contract-bill association data. In the bank fund income and expenditure table, search for bank transaction records that match both the paying bank account and the receiving bank account in the lease contract-invoice association data, and where the transaction date falls within the contract start and end dates. This indicates a successful match between the lease contract-invoice association data and the bank transaction record. The successfully associated lease contract record, invoice record, and bank transaction record are then obtained. These successfully associated lease contract record, invoice record, and bank transaction record are then integrated into a single associated record and stored in a unified write-off dataset.

4. The asset bill write-off management method according to claim 3, characterized in that: The multimodal write-off decision engine includes a single associated record collector, a rule matching scheduler, a rule execution processor, a multimodal information fusion arbitrator, and a write-off status and information outputter, wherein: The single associated record collector is used to obtain a single associated record of the unified write-off data set, including contract structured data, bill structured data, bank transaction structured data and associated unstructured key information; The rule matching scheduler is used to apply the multimodal write-off rules according to the preset multimodal write-off rule priority order; the rule execution processor is used to perform the amount calculation or date comparison operation corresponding to the multimodal write-off rule when the multimodal write-off rule meets the trigger condition, and update the write-off status mark; The multimodal information fusion arbitrator is used to make a ruling based on the associated unstructured key information when there is conflict or ambiguity in the structured data during the rule execution process to determine the final write-off status; The write-off status and information outputter is used to output the write-off result of a single associated record.

5. The asset bill write-off management method according to claim 4, characterized in that: The multimodal write-off decision engine also includes a confidence score calculator, which is used to perform a write-off confidence score for each write-off result and evaluate the reliability of the write-off result based on the write-off confidence score. The write-off confidence score is scored based on four factors, namely the clarity of rule matching, the consistency of multimodal information, the confidence of unstructured text parsing, and the accuracy of historical similar write-offs. The write-off confidence score is obtained by weighted calculation based on preset factor weights and factor score judgment criteria. If the write-off confidence score is lower than the preset confidence threshold, the write-off result will be marked as pending review to remind manual further confirmation.

6. The asset bill write-off management method according to claim 4, characterized in that: The multimodal write-off decision engine also includes an offset write-off scheme generator, a deposit write-off manager, and an overpayment processing coordinator, wherein: The offset and write-off scheme generator is used to identify multiple unreconciled bills and multiple unreconciled deposits under the same customer. When the total amount of the customer's unreconciled deposits is greater than or equal to the total amount of unreconciled bills, the offset and write-off process is initiated. At this time, an offset scheme is automatically generated according to preset business rules, and the write-off status of the relevant bills and transactions is marked based on the offset scheme; The deposit write-off manager is used to automatically identify transaction records marked as deposits and record the corresponding transaction amounts. When a positive bill is generated during a subsequent transaction, the marked deposit is automatically checked and applied to the write-off operation. After the write-off operation is completed, the deposit balance is updated. The overpayment processing coordinator is used to automatically calculate the difference of each write-off transaction after the write-off is completed. If the difference is positive, it is marked as an overpayment; and the processing method of the overpayment is determined according to the contract terms and corporate regulations.

7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, an asset bill write-off management method as described in any one of claims 1 to 6 is implemented.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, an asset bill write-off management method as described in any one of claims 1 to 6 is implemented.

Images