Automatic electronic accounting archive filing system based on finance and tax data intelligent association

Abstract

The invention discloses an electronic accounting archive automatic filing system based on finance and tax data intelligent association, and relates to the technical field of electronic accounting archive management. The system comprises the following components: a voucher data acquisition module, a multi-source original document storage module, an AI intelligent association matching module, a voucher-document binding and compliance filing module and a system control and operation module. According to the invention, through integration of the voucher data acquisition module, the multi-source original document storage module and the AI intelligent association matching module, automatic association and archiving of the electronic accounting voucher and the original document are realized, especially through a multi-dimensional matching algorithm adopted by the AI intelligent association matching module, the matching precision of the amount and the text feature is significantly improved, and the matching efficiency is improved. And human errors are reduced, so that the archiving process is more efficient and accurate.

Description

Technical Field

[0001] This invention relates to the field of electronic accounting record management technology, specifically to an automatic electronic accounting record archiving system based on intelligent association of financial and tax data. Background Technology

[0002] With the acceleration of enterprise digital transformation, the number of electronic accounting vouchers and original documents has increased dramatically, and traditional manual filing methods can no longer meet the needs of efficient and accurate management.

[0003] Traditional electronic accounting record archiving methods mainly rely on manual operation, which has problems such as low efficiency and easy error. Specifically, when processing a large number of electronic accounting vouchers and original documents, traditional methods often require a lot of manpower and resources for manual matching and archiving. This is not only time-consuming and labor-intensive, but also prone to human error, such as data entry errors and inaccurate matching. In addition, traditional methods often fall short when dealing with complex business scenarios, such as those involving multiple transaction types and large amounts of unstructured data, and are unable to meet the needs of enterprises for efficient and accurate financial and tax management.

[0004] The electronic accounting archive automatic archiving system based on intelligent correlation of financial and tax data is an innovative solution developed to address the shortcomings of traditional archiving methods. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of existing technologies and provide an automatic electronic accounting archive archiving system based on intelligent association of financial and tax data. By integrating advanced components such as a voucher data acquisition module, a multi-source original document storage module, and an AI intelligent association matching module, it realizes the automated acquisition, intelligent association, and compliant archiving of electronic accounting vouchers and original documents. This system not only significantly improves archiving efficiency and accuracy and reduces human error, but also effectively enhances the enterprise's compliance management and risk control capabilities through differentiated integrity verification rules and anomaly warning mechanisms.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: an automatic electronic accounting archive archiving system based on intelligent association of financial and tax data, which includes the following components: a voucher data acquisition module, a multi-source original document storage module, an AI intelligent association matching module, a voucher-document binding and compliance archiving module, and a system control and maintenance module;

[0007] The voucher data acquisition module establishes a real-time data interface with the enterprise's financial software, captures all electronic accounting voucher data, and extracts key feature information, including at least the voucher number, transaction amount, transaction date, counterparty name, transaction type, and account code.

[0008] The multi-source original document storage module integrates three major data sources: an invoice box, a contract database, and a bank-enterprise direct connection database. It stores the original document structured data and electronic files from each data source and performs format standardization processing on unstructured data to form a document dataset with unified field format.

[0009] The AI ​​intelligent association matching module receives key feature information output by the voucher data collection module, completes targeted search, multi-dimensional matching and result verification in the document dataset of the multi-source original document storage module, and outputs a unique matching document list.

[0010] The voucher-document binding and compliance archiving module binds electronic accounting voucher files with original document files in the unique matching document list to generate voucher-document association packages, performs compliance verification on the association packages, encrypts and stores the verified association packages in the electronic archive and establishes a retrieval index;

[0011] The system control and maintenance module schedules the work sequence of the above four modules, provides a parameter configuration interface, records system operation logs, triggers abnormal warnings, and performs data backup, permission management, and system maintenance operations. The above five modules interact in real time through data interfaces to realize the full automation of the process from voucher collection to document matching, binding and archiving, and system maintenance. Each module can independently adapt to the financial and tax management scenarios of different enterprises.

[0012] Furthermore, the key feature extraction process of the voucher data acquisition module adopts multimodal data fusion verification technology, specifically including the following steps: First, obtain structured data of electronic accounting vouchers through the financial software API interface, and simultaneously identify unstructured text in the voucher scan or original electronic voucher using OCR technology; Second, use the BERT-BiLSTM-CRF hybrid model to extract entities from the unstructured text identified by OCR, extracting potential counterparty names, supplementing transaction types, and aligning the fields with the structured data obtained from the API interface; Third, establish feature consistency verification rules: if the potential counterparty extracted by OCR... If the similarity between the unit name and the counterparty's unit name in the structured data is ≥95%, the field is considered to have passed the verification. If the absolute error between the OCR recognition result of the transaction amount and the amount in the structured data is ≤0.01 yuan, the amount field is considered to have passed the verification. In the fourth step, the fields that have passed the verification are integrated into a voucher feature dataset, and the fields that have failed the verification are automatically marked as pending confirmation, triggering an early warning from the system control and maintenance module, prompting manual review. This technology can reduce the error rate of key feature extraction to below 0.3%, solving the problem of feature loss or error caused by traditional single data source extraction, and is especially suitable for scenarios such as manufacturing and retail industries where there is handwritten supplementary information on vouchers.

[0013] Furthermore, the format standardization process of the multi-source original document storage module is designed with a hierarchical field mapping algorithm for unstructured contract text. The specific implementation is as follows: First, the unstructured contract text is segmented into paragraphs according to the contract header, transaction terms, and signature page, and the core keywords of each paragraph are extracted using the TextRank algorithm. Second, an industry-specific field mapping library is established. Third, a field matching model based on an attention mechanism is used to associate paragraph keywords with fields in the mapping library. Fourth, the data types of the mapped fields are unified. This algorithm can improve the accuracy of field extraction from unstructured contracts to over 98%, solving the problem of field misalignment caused by differences in contract templates among different enterprises, and providing a unified data foundation for the subsequent AI intelligent association and matching module.

[0014] Furthermore, the feature weight configuration unit of the AI ​​intelligent association matching module adopts a business-data dual-driven weight determination algorithm to generate the matching weights of each key feature, as shown in the formula: W i =α×S i +(1-α)×C i W i Let S be the final matching weight for the i-th key feature, with a value range of [0, 1], and α be an adjustment coefficient with a value range of [0.4, 0.6]. i The business importance score for the i-th feature is determined, with a value range of [0, 1], C i The contribution of the i-th feature to historical data is defined, with a value range of [0, 1]. This algorithm solves the problem that traditional fixed weights cannot adapt to different business scenarios. By combining business needs with historical data, the weight configuration becomes more reasonable. According to tests, the matching accuracy of the algorithm is more than 25% higher than that of fixed weights, and it is especially suitable for group enterprises with complex business types.

[0015] Furthermore, the intelligent targeted search unit of the AI ​​intelligent association matching module adopts a transaction type-data source correlation ranking algorithm to achieve accurate data source retrieval. The specific steps and parameter definitions are as follows: First, establish a transaction type-data source correlation matrix. The row dimension of the matrix represents all transaction types of the enterprise, and the column dimension represents the three major data sources. The matrix element R ij This represents the correlation between the i-th transaction type and the j-th data source, with a value range of [0, 1]; Second step, R... ij The determination method is as follows: count the number of times (N) the j-th data source provides matching documents when the i-th transaction type is archived within the past 24 months. ij ) and the total number of times this transaction type is archived (N) i The ratio of ) to the third step, based on the correlation degree R ij Sort the data sources according to the following rule: prioritize searching R.ij For data sources with a value ≥ 0.9, then search for values ​​≤ 0.7. ij For data sources with a value <0.9, the final search result is R. ij The fourth step involves using a recent update time priority strategy within each data source to prioritize retrieving documents updated within the past 30 days, thereby reducing invalid retrieval of outdated historical data. This algorithm can shorten the search time to 1 / 3 of the traditional full-volume search, making it particularly suitable for medium and large enterprises with more than 100,000 documents, thus solving the problem of low data source retrieval efficiency.

[0016] Furthermore, the multi-dimensional matching unit of the AI ​​intelligent association matching module designs a time-decay error rate algorithm for numerical features to accurately determine the validity of monetary matching. The formula is as follows: Where E ′ τ is the time decay error rate of the amount matching, with a value range of [0, +∞). A is the transaction amount of the voucher extracted by the voucher data acquisition module, B is the amount of the document to be matched in the multi-source original document storage module, β is the time decay coefficient, with a value range of [0.1, 0.3], T is the difference between the date of the document to be matched and the date of the voucher, and τ is the base time span.

[0017] Furthermore, the multi-dimensional matching unit of the AI ​​intelligent association matching module designs an industry entity-enhanced similarity algorithm for text-based features to address matching errors caused by company abbreviations and aliases. The formula is: Sim = γ × Sim c +(1-γ)×Sim i Where Sim is the final text similarity of the other party's unit name, with a value ranging from [0, 1], and γ is the core entity weight coefficient, with a value ranging from [0.7, 0.8]. c Sim represents the core entity similarity, with values ​​ranging from [0, 1]. i The industry feature similarity is defined as [0, 1]. This algorithm solves the problem that traditional text similarity algorithms cannot identify company abbreviations and aliases, and meets the actual business needs for matching company abbreviations. According to tests, the text matching accuracy of this algorithm is more than 30% higher than that of traditional algorithms, and it is especially suitable for industries such as Internet and Finance where company abbreviations are frequently used.

[0018] Furthermore, the matching result verification unit of the AI ​​intelligent association matching module adopts a two-layer filtering algorithm to exclude duplicate matches and invalid documents. The specific implementation process is as follows: The first layer is duplicate match filtering: a document-voucher association mapping table is established to record the voucher number and association time associated with each original document. When a document is to be associated with a new voucher, the matching score between the document and the new voucher is calculated. If the difference in transaction date between the historical voucher associated with the document and the new voucher is ≤15 days and the difference in transaction amount is ≤5%, it is determined to be a duplicate match, and the document is automatically filtered. If the transaction date difference between the historical associated voucher and the new voucher is greater than 30 days, or the amount difference is greater than 10%, then re-association is allowed, and the associated records in the mapping table are updated. The second layer is invalid document filtering: invalid judgment rules are set for different types of documents, and the filtered duplicate and invalid document information is recorded in the log of the system control and maintenance module for easy traceability later. This algorithm can reduce the false judgment rate of invalid documents to below 0.2%, solving the problem of missing duplicate and invalid documents in traditional manual verification, and is especially suitable for the retail and service industries with frequent invoice circulation.

[0019] Furthermore, the compliance verification unit of the voucher-document binding and compliance archiving module adopts transaction type-risk level mapping verification rule generation technology to generate differentiated integrity verification rules for different transaction types. The specific implementation steps are as follows: First, establish a three-dimensional mapping library of transaction type-risk level-verification item, where transaction types cover procurement, sales, expense reimbursement, etc., and risk levels are divided into high, medium, and low levels according to the historical compliance risk occurrence rate. Verification items are associated with the mandatory / optional items corresponding to the risk level; Second, use the analytic hierarchy process (AHP) to determine the weight of verification items at each level, with high-risk verification items having a weight of [missing information]. The value range is [0.4, 0.6], with medium risk being [0.2, 0.4] and low risk being [0.1, 0.2]. The third step involves matching the risk level in the mapping library to the transaction type of the voucher to be verified, automatically calling the corresponding verification items and weights, and generating differentiated verification rules. The fourth step involves updating the risk level and verification items in the mapping library quarterly based on newly added compliance risk events to ensure that the rules dynamically adapt to business changes. This technology can improve the verification accuracy of different transaction types by more than 35%, avoiding inefficiency caused by excessive verification of low-risk transactions, while ensuring the compliance integrity of high-risk transactions.

[0020] Furthermore, the anomaly warning unit of the system control and operation module adopts a three-dimensional dynamic scheduling technology of warning event-resource occupancy-processing timeliness to rationally allocate manual intervention resources. The specific implementation steps are as follows: First, construct a correlation library of warning event type-resource occupancy coefficient-standard processing timeliness. Warning event types include matching failure, verification failure, etc. The resource occupancy coefficient is divided into 0.1-0.9 according to the manual processing time required, and the standard processing timeliness refers to regulatory requirements and enterprise process settings. Second, calculate the warning scheduling coefficient: scheduling coefficient = resource occupancy coefficient × (1 + expired timeliness / standard processing timeliness). The expired timeliness is the difference between the current time and the warning trigger time minus the standard processing timeliness. Third, classify priorities according to the scheduling coefficient: ≥0.8 is level one, 0.4-0.8 is level two, and <0.4 is level three. Fourth, update the resource occupancy coefficient every half month based on the processing time data to ensure scheduling accuracy. This technology can improve the response speed of high-priority warning processing by 50% and avoid misallocation of manual resources.

[0021] Compared with existing technologies, this electronic accounting record automatic archiving system based on intelligent correlation of financial and tax data has the following advantages:

[0022] I. This system integrates a voucher data acquisition module, a multi-source original document storage module, and an AI intelligent association matching module to achieve automated association and archiving of electronic accounting vouchers and original documents. In particular, the multi-dimensional matching algorithm used by the AI ​​intelligent association matching module significantly improves the matching accuracy of amount and text features, reduces human error, and makes the archiving process more efficient and accurate.

[0023] Second, this system, through its voucher-document binding and compliance archiving module, employs transaction type-risk level mapping verification rule generation technology. This technology automatically generates differentiated integrity verification rules for different transaction types. This function ensures that high-risk transactions receive strict compliance audits while avoiding excessive verification of low-risk transactions, effectively improving the overall level of compliance management. In addition, the anomaly warning unit of the system control and maintenance module uses three-dimensional dynamic scheduling technology, which can reasonably allocate human intervention resources, quickly respond to and handle compliance anomalies, further enhancing the enterprise's risk control capabilities. Through these measures, the system helps enterprises achieve refined management of electronic accounting archives and reduce compliance risks.

[0024] Other advantages, objectives and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination or study, or may be learned from the practice of the invention. Attached Figure Description

[0025] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0026] Figure 1 The flowchart shows the operation of an automated electronic accounting record archiving system based on intelligent correlation of financial and tax data.

[0027] Figure 2 Layered workflow diagram for the AI ​​intelligent association matching module;

[0028] Figure 3 This is a schematic diagram of the core modules of an electronic accounting archive automatic archiving system based on intelligent association of financial and tax data. Detailed Implementation

[0029] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structure, features and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0030] Example 1

[0031] After a manufacturing company completes a raw material procurement transaction with a supplier, if Figure 1 As shown, firstly, the voucher data acquisition module captures the full amount of electronic accounting voucher data for the procurement transaction through a real-time data interface with the enterprise's financial software. It extracts key feature information such as voucher number, transaction amount, transaction date, counterparty name, transaction type, and account code. Simultaneously, it employs multimodal data fusion verification technology to perform feature extraction verification: structured data is obtained through the financial software API interface; unstructured text in the voucher scan is identified using OCR technology; a BERT-BiLSTM-CRF hybrid model is used to extract entities and supplement relevant information; and the data is aligned with the structured data obtained from the API. Then, it is verified according to feature consistency verification rules. It confirms that the similarity between the potential counterparty name extracted by OCR and the counterparty name text in the structured data reaches over 95%, and the absolute error between the OCR recognition result of the transaction amount and the amount in the structured data does not exceed 0.01 yuan. The verified fields are integrated into a voucher feature dataset. If any fields fail verification, they are marked as pending confirmation, triggering an early warning prompt for manual review.

[0032] Next, the multi-source original document storage module integrates three major data sources: invoice box, contract database, and bank-enterprise direct connection database. It stores the structured data of purchase invoices, electronic documents of purchase contracts, and bank payment records corresponding to the purchase business. For the unstructured text of the purchase contract, a hierarchical field mapping algorithm is used for format standardization: the contract text is divided into paragraphs according to the contract header, transaction terms, and signature page. The core keywords of each paragraph are extracted by the TextRank algorithm. The industry-specific field mapping library is called. A field matching model based on the attention mechanism is used to associate the paragraph keywords with the fields in the mapping library. After completion, the data types of the mapped fields are unified to form a document dataset with a unified field format.

[0033] Next, the AI ​​intelligent association and matching module receives the voucher feature dataset, such as Figure 2 As shown, the matching weights for each key feature are first generated using a business-data dual-driven weight determination algorithm, with the formula: W i =α×S i +(1-α)×C i W i S is the final matching weight for the i-th key feature, α is the adjustment coefficient, and S i To score the business importance of the i-th feature, C i To assess the contribution of historical data to the i-th feature, an intelligent targeted search is conducted using a transaction type-data source correlation ranking algorithm: a correlation matrix between transaction type and the three major data sources is established. The correlation is determined by the ratio of the number of times each data source provided matching documents when the procurement business was archived in the past 24 months to the total number of archives for that business. Data sources with a correlation ≥ 0.9 are searched first, followed by data sources with a correlation ≤ 0.7 and < 0.9, and finally data sources with a correlation < 0.7. Within each data source, documents updated in the past 30 days are searched first to reduce invalid searches.

[0034] In the multi-dimensional matching process, a time-decaying error rate algorithm is used to determine the matching validity for the numerical feature of transaction amount. The formula is as follows: Where E ′ Let β be the time decay error rate for matching amounts, A be the transaction amount extracted by the voucher data acquisition module, B be the amount of the document to be matched in the multi-source original document storage module, β be the time decay coefficient, T be the difference between the date of the document to be matched and the date of the voucher, and τ be the baseline time span. For the textual feature of the counterparty's name, an industry entity-enhanced similarity algorithm is used to calculate the similarity, with the formula: Sim = γ × Sim c +(1-γ)×Sim i Where Sim is the final text similarity of the other party's unit name, γ is the core entity weight coefficient, and Sim c For core entity similarity, Simi To ensure industry feature similarity and accurate feature matching, a two-layer filtering algorithm is used to verify the results after matching: a document-voucher association mapping table is established. If the difference in transaction date between a document's associated historical voucher and the new voucher is ≤15 days and the difference in transaction amount is ≤5%, it is determined to be a duplicate match and automatically filtered. If the difference in transaction date between the historical associated voucher and the new voucher is >30 days or the difference in amount is >10%, re-association is allowed and the mapping table is updated. At the same time, invalid documents are filtered according to the set rules, and finally a unique matching document list is output.

[0035] Subsequently, the voucher-document binding and compliance archiving module binds electronic accounting voucher files with original document files such as purchase invoices, purchase contracts, and bank payment records from the uniquely matched document list, generating voucher-document association packages. Compliance verification is performed using transaction type-risk level mapping verification rule generation technology: calling the three-dimensional mapping library of transaction type-risk level-verification item, matching the corresponding mandatory / optional verification items according to the risk level of the procurement business, determining the weight of verification items through the analytic hierarchy process, generating differentiated compliance verification rules, and comprehensively verifying the association packages. The verified association packages will be encrypted and stored in the electronic archive, and a retrieval index will be established for subsequent queries.

[0036] Finally, as Figure 3 As shown, the system control and operation module schedules the working sequence of the above four modules, provides a parameter configuration interface for staff to adjust relevant settings, and records system operation logs in real time. If early warning events such as matching failure or verification failure occur during the process, a three-dimensional dynamic scheduling technology of early warning event-resource occupancy-processing timeliness is used to allocate human intervention resources: calling the early warning event type-resource occupancy coefficient-standard processing timeliness association library, calculating the scheduling coefficient, dividing priorities according to the scheduling coefficient, and reasonably allocating manpower to handle early warning events. At the same time, data backup, permission management and system operation and maintenance operations are performed regularly. The resource occupancy coefficient is updated every half month, and the risk level and verification items in the three-dimensional mapping library are updated every quarter based on newly added compliance risk events to ensure that the system can stably adapt to business changes.

[0037] Example 2

[0038] An employee of an internet company submitted a travel expense reimbursement application after completing a remote project coordination, such as... Figure 1As shown, firstly, the voucher data acquisition module captures the full amount of electronic accounting voucher data for the expense reimbursement through a real-time data interface with the enterprise's financial reimbursement software. It extracts key feature information such as voucher number, reimbursement amount, reimbursement date, the department of the employee making the reimbursement, transaction type, and account code. Multimodal data fusion verification technology is used for feature extraction and verification: structured data such as reimbursement amount and reimbursement date are obtained through the financial software API interface; OCR technology is used to identify unstructured text in scanned travel invoices and reimbursement form photos; and a BERT-BiLSTM-CRF hybrid model is used to extract potential entity information such as passenger information and invoice issuing unit. Field alignment is performed with the structured data obtained from the API, and verification is conducted according to feature consistency verification rules. This ensures that the similarity between the invoice issuing unit name extracted by OCR and the related name text recorded in the structured data is ≥95%, and the absolute error between the OCR recognition result of the reimbursement amount and the amount in the structured data is ≤0.01 yuan. Fields that pass verification are integrated into a voucher feature dataset, while fields that fail are marked as pending confirmation and trigger system alerts, prompting financial personnel to conduct manual review.

[0039] Next, the multi-source original document storage module integrates three major data sources: the invoice box, the contract database, and the bank-enterprise direct connection database. It stores structured data such as travel invoices, electronic files of project cooperation contracts, and bank transfer records of reimbursements paid by enterprises to employees. For unstructured text such as project cooperation contracts, a hierarchical field mapping algorithm is used for format standardization: the contract is divided into paragraphs according to the contract header, transaction terms, and signature page. The core keywords of each paragraph are extracted using the TextRank algorithm and matched with fields in the industry-specific field mapping library. After completion, the data types of each field are unified to form a document dataset with a unified format.

[0040] Next, the AI ​​intelligent association and matching module receives the voucher feature dataset, such as Figure 2 As shown, the algorithm first assigns matching weights to key features such as reimbursement amount, reimbursement date, and invoice issuing unit through a business-data dual-drive weight determination algorithm. Then, a targeted search is performed using a transaction type-data source correlation ranking algorithm: a correlation matrix is ​​constructed between the expense reimbursement transaction type and the three major data sources. The ratio of the number of times each data source provides matching documents to the total number of times the expense reimbursement business is archived in the past 24 months is calculated to determine the correlation of each data source. Data sources are searched in the order of correlation ≥ 0.9, 0.7 ≤ correlation < 0.9, and correlation < 0.7. Within each data source, documents updated in the past 30 days are queried first to avoid invalid searches of outdated data.

[0041] When performing multi-dimensional matching, a time-decaying error rate algorithm is used to determine the validity of the match for numerical features such as reimbursement amount, while an industry entity-enhanced similarity algorithm is used to calculate similarity for textual features such as invoice issuing unit and employee name to ensure matching accuracy. After matching, the results are verified by a two-layer filtering algorithm: query the document-voucher association mapping table. If the transaction date difference between the historical voucher associated with a certain original document and the current reimbursement voucher is ≤15 days and the amount difference is ≤5%, it is determined to be a duplicate match and automatically filtered. If the transaction date difference is >30 days or the amount difference is >10%, it is allowed to re-associate and update the mapping table. At the same time, invalid documents are filtered according to preset rules, and finally a unique matching document list is output.

[0042] Subsequently, the voucher-document binding and compliance archiving module binds the electronic accounting vouchers for expense reimbursement with the original document files in the uniquely matched document list, generating a voucher-document association package. Compliance verification is then performed using a transaction type-risk level mapping verification rule generation technology: a three-dimensional mapping library of transaction type-risk level-verification item is invoked; the risk level is determined based on the historical compliance risk occurrence rate of expense reimbursement transactions; corresponding mandatory and optional verification items are matched; the weight of each verification item is determined using the analytic hierarchy process; differentiated compliance verification rules are generated; and the association package is comprehensively verified. Verified association packages are encrypted and stored in the electronic archive, and a retrieval index is established to facilitate subsequent financial audits, document searches, and other operations.

[0043] Finally, as Figure 3 As shown, the system control and operation module coordinates the workflow of the above four modules, provides a parameter configuration interface for staff to adjust algorithm parameters, verification rules, etc., and records real-time log information such as module interaction, data processing, and early warning triggering during system operation. When early warning events such as matching failure or verification failure occur, a three-dimensional dynamic scheduling technology of early warning event-resource occupancy-processing time is used to allocate human resources: based on the association library of early warning event type-resource occupancy coefficient-standard processing time, the scheduling coefficient is calculated, and priorities are divided according to scheduling coefficient ≥0.8, 0.4-0.8, and <0.4, with high-priority early warning events being processed first. The system regularly performs data backups, strictly controls the file access permissions of personnel in different positions, updates the resource occupancy coefficient every half month based on the processing time data of early warning events, and updates the risk level and verification items in the three-dimensional mapping library every quarter based on newly added expense reimbursement compliance risk events, ensuring that the system continuously adapts to the needs of enterprise business development and compliance management.

[0044] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. An automatic electronic accounting record archiving system based on intelligent association of financial and tax data, characterized in that: The system comprises the following components: a voucher data acquisition module, a multi-source original document storage module, an AI intelligent association and matching module, a voucher-document binding and compliance archiving module, and a system control and maintenance module; The voucher data acquisition module establishes a real-time data interface with the enterprise's financial software, captures all electronic accounting voucher data, and extracts key feature information, including at least the voucher number, transaction amount, transaction date, counterparty name, transaction type, and account code. The multi-source original document storage module integrates three major data sources: an invoice box, a contract database, and a bank-enterprise direct connection database. It stores the original document structured data and electronic files from each data source and performs format standardization processing on unstructured data to form a document dataset with unified field format. The AI ​​intelligent association matching module receives key feature information output by the voucher data collection module, completes targeted search, multi-dimensional matching and result verification in the document dataset of the multi-source original document storage module, and outputs a unique matching document list. The voucher-document binding and compliance archiving module binds electronic accounting voucher files with original document files in the unique matching document list to generate voucher-document association packages, performs compliance verification on the association packages, encrypts and stores the verified association packages in the electronic archive and establishes a retrieval index; The system control and maintenance module schedules the working sequence of the above four modules, provides a parameter configuration interface, records system operation logs, triggers abnormal warnings, and performs data backup, permission management, and system maintenance operations; the above five modules interact in real time through data interfaces.

2. The electronic accounting archive automatic archiving system based on intelligent association of financial and tax data as described in claim 1, characterized in that, The key feature extraction process of the voucher data acquisition module adopts multimodal data fusion verification technology, specifically including the following steps: First, obtain structured data of electronic accounting vouchers through the financial software API interface, and simultaneously identify unstructured text in the voucher scan or original electronic voucher using OCR technology; Second, use the BERT-BiLSTM-CRF hybrid model to extract entities from the unstructured text identified by OCR, extract potential counterparty names and supplement information such as transaction type, and align the fields with the structured data obtained from the API interface; Third, establish feature consistency verification rules: if the similarity between the potential counterparty name extracted by OCR and the counterparty name text in the structured data is ≥95%, the field is deemed to have passed verification; if the absolute error between the OCR recognition result of the transaction amount and the amount in the structured data is ≤0.01 yuan, the amount field is deemed to have passed verification; Fourth, integrate the fields that have passed verification into a voucher feature dataset, and automatically mark the fields that have failed verification as pending confirmation, triggering an early warning from the system control and maintenance module, prompting manual review.

3. The electronic accounting archive automatic archiving system based on intelligent association of financial and tax data as described in claim 1, characterized in that, The format standardization process of the multi-source original document storage module is designed with a hierarchical field mapping algorithm for unstructured contract text. The specific implementation is as follows: First, the unstructured contract text is divided into paragraphs according to the contract header, transaction terms, and signature page, and the core keywords of each paragraph are extracted using the TextRank algorithm; Second, an industry-specific field mapping library is established; Third, a field matching model based on an attention mechanism is used to associate paragraph keywords with fields in the mapping library; Fourth, the data types of the mapped fields are unified.

4. The electronic accounting archive automatic archiving system based on intelligent association of financial and tax data as described in claim 1, characterized in that, The feature weight configuration unit of the AI ​​intelligent association matching module adopts a business-data dual-driven weight determination algorithm to generate the matching weight of each key feature, using the formula: W i =α×S i +(1-α)×C i W i S is the final matching weight for the i-th key feature, α is the adjustment coefficient, and S i To score the business importance of the i-th feature, C i The contribution of the historical data to the i-th feature.

5. The electronic accounting archive automatic archiving system based on intelligent association of financial and tax data as described in claim 1, characterized in that, The intelligent targeted search unit of the AI ​​intelligent association matching module adopts a transaction type-data source correlation ranking algorithm. The specific steps and parameter definitions are as follows: First, establish a transaction type-data source correlation matrix. The matrix row dimension represents all transaction types of the enterprise, and the column dimension represents the three major data sources. The matrix element R ij This represents the correlation between the i-th transaction type and the j-th data source, with a value range of [0, 1]; Second step, R... ij The determination method is as follows: count the number of times (N) the j-th data source provides matching documents when the i-th transaction type is archived within the past 24 months. ij ) and the total number of times this transaction type is archived (N) i The ratio of ) to the third step, based on the correlation degree R ij Sort the data sources according to the following rule: prioritize searching R. ij For data sources with a value ≥ 0.9, then search for values ​​≤ 0.

7. ij For data sources with a value <0.9, the final search result is R. ij For data sources with a value of <0.7; the fourth step is to adopt a recent update time priority strategy within each data source, prioritizing the retrieval of documents updated within the past 30 days to reduce invalid retrieval of outdated historical data.

6. The automatic electronic accounting record archiving system based on intelligent association of financial and tax data as described in claim 1, characterized in that, The multi-dimensional matching unit of the AI ​​intelligent association matching module designs a time-decay error rate algorithm for numerical features to accurately determine the validity of monetary matching. The formula is as follows: Where E ′ τ is the time decay error rate for matching amounts, A is the transaction amount of the voucher extracted by the voucher data acquisition module, B is the amount of the document to be matched in the multi-source original document storage module, β is the time decay coefficient, T is the difference between the date of the document to be matched and the date of the voucher, and τ is the base time span.

7. The electronic accounting archive automatic archiving system based on intelligent association of financial and tax data as described in claim 1, characterized in that, The multi-dimensional matching unit of the AI ​​intelligent association matching module designs an industry entity-enhanced similarity algorithm for text-based features, with the formula: Sim = γ × Sim c +1-γ)×Sim i Where Sim is the final text similarity of the other party's unit name, γ is the core entity weight coefficient, and Sim c For core entity similarity, Sim i This represents the similarity in industry characteristics.

8. The automatic electronic accounting archive archiving system based on intelligent association of financial and tax data as described in claim 1, characterized in that, The matching result verification unit of the AI ​​intelligent association matching module adopts a two-layer filtering algorithm to exclude duplicate matches and invalid documents. The specific implementation process is as follows: The first layer is duplicate match filtering: a document-voucher association mapping table is established to record the voucher number and association time associated with each original document. When a document is to be associated with a new voucher, the matching score between the document and the new voucher is calculated. If the difference in transaction date between the historical voucher associated with the document and the new voucher is ≤15 days and the difference in transaction amount is ≤5%, it is judged as a duplicate match and the document is automatically filtered. If the difference in transaction date between the historical associated voucher and the new voucher is >30 days or the difference in amount is >10%, re-association is allowed and the association record in the mapping table is updated. The second layer is invalid document filtering: invalid judgment rules are set for different types of documents, and the information of the filtered duplicate and invalid documents is recorded in the log of the system control and maintenance module.

9. The automatic electronic accounting archive archiving system based on intelligent association of financial and tax data as described in claim 1, characterized in that, The compliance verification unit of the voucher-document binding and compliance archiving module adopts transaction type-risk level mapping verification rule generation technology to generate differentiated integrity verification rules for different transaction types. The specific implementation steps are as follows: First, establish a three-dimensional mapping library of transaction type-risk level-verification item. The transaction types cover procurement, sales, expense reimbursement, etc., and the risk levels are divided into high, medium, and low levels according to the historical compliance risk occurrence rate. Verification items are associated with the mandatory / optional items corresponding to the risk level. Second, use the analytic hierarchy process (AHP) to determine the weight of each level of verification item. The weight of high-risk verification items ranges from [0.4, 0.6], medium-risk from [0.2, 0.4], and low-risk from [0.1, 0.2]. Third, match the transaction type of the voucher to be verified with the risk level in the mapping library, automatically call the corresponding level of verification items and weights, and generate differentiated verification rules. Fourth, update the risk level and verification items in the mapping library every quarter based on newly added compliance risk events to ensure that the rules dynamically adapt to business changes.

10. The automatic electronic accounting record archiving system based on intelligent association of financial and tax data according to claim 1, characterized in that, The abnormal early warning unit of the system control and operation and maintenance module adopts a three-dimensional dynamic scheduling technology of early warning event-resource occupancy-processing timeliness to rationally allocate manual intervention resources. The specific implementation steps are as follows: First, construct a correlation library of early warning event type-resource occupancy coefficient-standard processing timeliness. The early warning event type includes matching failure, verification failure, etc. The resource occupancy coefficient is divided into 0.1-0.9 according to the manual time required for processing. The standard processing timeliness refers to the regulatory requirements and enterprise process settings. The second step is to calculate the early warning scheduling coefficient: scheduling coefficient = resource occupancy coefficient × (1 + timeout duration / standard processing time), where the timeout duration is the difference between the current time and the early warning trigger time minus the standard processing time; The third step is to prioritize tasks based on scheduling coefficients: ≥0.8 is Level 1, 0.4-0.8 is Level 2, and <0.4 is Level 3. The fourth step is to update the resource utilization coefficients every half month based on processing time data.

Images