Cross-platform financial data automatic accounting method and system

Abstract

The present application relates to the technical field of enterprise financial automation accounting, in particular to a cross-platform financial data automatic accounting method and system, comprising: establishing connection with multiple heterogeneous financial systems, periodically acquiring original flow containing transaction note text, performing field extraction and format cleaning to generate standardized flow entries. The standardized entries are sent to a pre-trained text semantic classifier, and the economic essence category of the transaction behavior is parsed based on the note text. According to the category, the corresponding entry rule is matched from the accounting rule knowledge base, and the flow entry is decomposed into account, and a standardized accounting entry draft is generated. The hook and check relationship verification is performed on the entry draft, and the internal debit and credit balance and the logical conflict with the historical entered transaction in the subject continuity are checked. The present application realizes intelligent understanding and automatic, logical rigorous accounting processing of multi-source heterogeneous financial data.

Description

Technical Field

[0001] This invention relates to the field of automated financial accounting technology, and in particular to a cross-platform method and system for automated financial data accounting. Background Technology

[0002] Currently, when conducting financial accounting, enterprises often need to interface with multiple external, heterogeneous financial or business systems, such as banks, payment platforms, and ERP systems. The raw transaction records generated by these systems exhibit significant differences in data format, field definitions, character encoding, and monetary units. Handling these differences primarily relies on finance personnel manually cleaning, converting, and standardizing the data, or writing numerous one-off script rules. This approach is inefficient when processing multi-source, high-frequency data and is highly susceptible to errors due to rule omissions or human negligence, making the data preprocessing stage a bottleneck in automated processes.

[0003] In the transaction substance assessment stage, existing technical solutions generally employ rule-based matching methods using fixed keywords or regular expressions to categorize transaction notes. However, natural language expressions are diverse and complex; transactions with the same economic substance may correspond to a vast number of different text descriptions, making it difficult to exhaustively define rule bases and resulting in high maintenance costs. This leads to a large number of transactions failing to be accurately classified, still requiring manual review and assessment, thus limiting automation. Regarding accounting logic verification, existing automated accounting systems typically only perform the most basic debit and credit balance checks, lacking in-depth checks on the logical continuity of account usage and the correlation between inter-period transactions.

[0004] The key obstacles to improving the intelligence and reliability of the entire financial accounting process are how to accurately and automatically identify the economic substance of transactions from unstructured text, and how to implement verification beyond individual transactions after automated journal entries are generated, ensuring the rigor of the overall accounting logic. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of existing technologies by proposing a cross-platform automatic financial data calculation method and system.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a cross-platform automatic financial data calculation method, comprising: Establish communication connections with multiple heterogeneous financial systems, and periodically obtain raw transaction records from the heterogeneous financial systems. The raw transaction records include transaction timestamps, transaction party information, transaction amounts, and transaction remarks text. The obtained original transaction records are subjected to field extraction and format cleaning to eliminate differences in data format, character encoding and unit of measurement between different heterogeneous financial systems, and to generate standardized transaction records. The standardized transaction entries are fed into a pre-trained text semantic classifier, which parses the economic substance category of the transaction behavior based on the content of the transaction notes text. Based on the economic substance category of the transaction, the corresponding accounting rule is matched from the preset accounting rule knowledge base; Using the matched accounting rules, the standardized transaction entries are broken down into accounting entries to generate standardized accounting drafts. The standardized accounting entry drafts are subjected to cross-checking to examine the debit and credit balance among multiple standardized accounting entry drafts derived from the same original transaction, and to check for any logical conflicts in account continuity with historically recorded transactions.

[0007] As a further aspect of the present invention, the step of extracting fields and cleaning the format of the obtained original transaction log includes: Identify the source system identifier of each of the original transaction flows, and load the corresponding data parsing template from the preset format mapping template library according to the source system identifier; Using the data parsing template, key-value pair extraction is performed on the unstructured text or semi-structured data of the original transaction flow to extract the original strings of the transaction timestamp, transaction party information, transaction amount, and transaction remarks text; The extracted transaction timestamps are uniformly converted to Coordinated Universal Time (UTC) format, the transaction amounts are uniformly converted to the specified base currency, and the character encodings of the transaction party information and the transaction remarks text are uniformly converted to the universal character set encoding. After the format and encoding conversion is completed, the transaction party information and the transaction remarks text are processed by named entity recognition to identify and correct spelling errors, inconsistent abbreviations and multiple words with the same meaning. The data from all cleaning and conversion steps are repackaged according to a preset internal data model and output as the standardized flow entries.

[0008] As a further aspect of the present invention, the standardized transaction entries are fed into a pre-trained text semantic classifier, which parses the economic substance category of the transaction behavior based on the content of the transaction remarks text, including: A semantic understanding model based on a deep neural network architecture is constructed. The semantic understanding model is trained using a large amount of labeled historical transaction notes and corresponding accounting subjects to form the pre-trained text semantic classifier. Extract the transaction note text from the standardized transaction entries, and input the transaction note text into the pre-trained text semantic classifier; The pre-trained text semantic classifier performs word segmentation, vectorization, and context semantic encoding on the transaction notes text, and outputs a high-dimensional semantic feature vector. The high-dimensional semantic feature vector is mapped to a predefined economic substance category probability space, and the probability value of its belonging to the economic substance category of each transaction behavior is calculated. The economic substance category of the transaction with the highest probability value is selected as the final classification result of the standardized transaction entry.

[0009] As a further aspect of the present invention, based on the economic substance category of the transaction, a corresponding accounting rule is matched from a preset accounting rule knowledge base, including: The accounting rules knowledge base contains structured accounting rule entries derived from accounting standard provisions; The accounting rule knowledge base is organized in a directed graph structure. The nodes of the graph represent specific accounting subjects, and the edges of the graph represent accounting processing logic derived from the accounting standard clauses. Each edge is accompanied by an economic substance category label of the transaction behavior to which it applies. Using the economic substance category of the transaction as the query index, retrieve all edges labeled with economic substance category in the directed graph structure of the accounting rule knowledge base; From the retrieved edges, further filter out the accounting rules that match other key attributes in the standardized transaction entries, including the transaction subject type and the country or region where the transaction occurred; From the multiple recorded entry rules selected, the highest priority recorded entry rule is selected as the matching result according to the preset rule priority strategy. The core data fields of the standardized transaction entries are bound to the matched accounting rules to form an accounting decomposition task to be executed.

[0010] As a further aspect of the present invention, the step of using the matched accounting rules to decompose the standardized transaction entries includes: The standardized accounting entry draft includes debit entries, credit entries, accounting amounts, and the corresponding accounting period; The logical structure of the accounting rules is analyzed to identify the amount calculation fields, account determination conditions, and debit / credit directions defined in the rules. The transaction amount and other relevant numerical fields are extracted from the standardized transaction entries, and the accounting amount to be included in the accounting entries is calculated according to the amount calculation logic defined in the accounting rules. Based on the attributes of the standardized transaction entries and the account determination criteria in the accounting rules, at least two accounting accounts involved in this transaction are determined. Each identified accounting item shall be designated as a debit accounting item or a credit accounting item according to the debit and credit directions defined in the accounting rules. Based on the transaction timestamps in the standardized transaction entries and in conjunction with a preset accounting period division strategy, the accounting period to which the standardized accounting entry draft belongs is determined. Summarize the debit and credit accounting entries, the accounting amounts, and the corresponding accounting periods to generate a complete draft of the standardized accounting entry.

[0011] As a further aspect of the present invention, the standardized accounting entry draft is subjected to cross-referencing verification, including: Extract all debit entries and their corresponding accounting amounts from the standardized accounting entry draft, and calculate the total debit amount; Extract all credit entries and their corresponding accounting amounts from the standardized accounting entry draft, and calculate the total credit amount. The total debit amount is compared with the total credit amount. If the difference exceeds the allowable numerical tolerance range, the draft standardized accounting entry is marked as having a debit-credit imbalance error. For accounting items related to assets, liabilities, and equity, retrieve the ending balance of the accounting item in the previous accounting period from the historical accounting database; Examine the draft standardized accounting entries containing the aforementioned accounting items to determine whether the direction and value of changes in the accounting amounts would lead to situations where the account balances are inconsistent with business logic. If such situations exist, mark them as continuous logic conflicts. Generate a verification report containing all error and conflict markers, and mark the verified standardized accounting entry drafts as pending audit entries.

[0012] As a further aspect of the present invention, it also includes a step of conducting a multi-dimensional review of the standardized accounting entry drafts marked as pending audit entries: Based on a preset set of review strategy rules, the entries to be reviewed are screened for risk dimensions. The set of review strategy rules includes a large transaction review threshold, a list of sensitive counterparties, and abnormal transaction time patterns. Trigger the business authenticity verification process for the entry to be audited, and retrieve supporting documents such as electronic invoices and scanned contracts related to the entry to be audited from the original voucher image database; The key information in the journal entry to be audited is automatically compared with the supporting documents retrieved. The key information includes amount, counterparty, date and transaction summary. For entries that fail the automatic comparison or trigger the high-risk rules in the risk dimension screening, they are pushed to the manual review queue, along with links to the original transaction records and supporting documents. For all pending entries that have passed automatic review, approve their formal posting and write the formal accounting entries into the general ledger database, while updating the real-time balances of the relevant accounts.

[0013] As a further aspect of the present invention, the process of triggering the business authenticity verification of the entry to be audited includes: The source of the entry to be audited is traced in reverse to obtain the globally unique identifier of the original transaction flow that generated the entry to be audited; Using the globally unique identifier as an index, a full-text search is performed in the original voucher image database to find the storage paths of all associated electronic invoices and contract scanned supporting documents; The document parsing service is used to identify the content of the supporting documents found and extract the structured data from the documents. The structured data includes invoice code, invoice number, invoice date, amount excluding tax, tax amount, and information of the buyer and seller. Establish a mapping relationship between the structured data from the journal entry to be audited to the supporting documents, and perform consistency verification item by item according to the preset audit logic; Record the results of each consistency check and generate a structured business authenticity verification report as part of the review conclusion of the entry to be audited.

[0014] As a further aspect of the present invention, it also includes periodic trial balancing and report pre-generation steps: At the end of each accounting period, all formal accounting entries that have been recorded in the current period are extracted from the general ledger database; All the formal accounting entries extracted are classified and summarized according to the accounting subjects, and the current period debit amount, credit amount and ending balance of each accounting subject are calculated. Based on the ending balances of all accounting items, generate a trial balance before adjustment and check whether the total assets in the trial balance are equal to the total liabilities and owners' equity. If the trial balance is unbalanced, the discrepancy tracing procedure is initiated to retrospectively examine all accounting entries and balance calculation processes for the current period until the cause of the discrepancy is found and adjusting entries are generated. After the trial balance is passed, data is extracted from the general ledger database according to the preset report template, and the main body of the balance sheet, income statement and cash flow statement are automatically filled to generate a draft financial statement.

[0015] As a further aspect of the present invention, the present invention also includes a cross-platform automatic financial data accounting system, the system including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein when the processor executes the computer program, it implements the steps of the cross-platform automatic financial data accounting method described above.

[0016] Compared with the prior art, the advantages and positive effects of the present invention are as follows: By directly parsing the content of transaction notes using a pre-trained text semantic classifier, the model can understand the transaction scenarios and purposes expressed in natural language. Trained on a large amount of labeled financial corpus, this classifier learns a deep mapping relationship between text features and economic substance categories, thus eliminating absolute reliance on fixed keyword rules. Faced with unfamiliar expressions or complex descriptions, the model can infer based on semantic similarity and classify the transaction into the correct economic substance category. This achieves high-precision automatic understanding of diverse and non-standardized transaction texts, reduces the need for manual intervention due to incomplete rule coverage, and truly automates the transaction classification process.

[0017] The system performs cross-checking on standardized accounting drafts, including debit / credit balance and account continuity checks, constructing a multi-layered automated audit logic. The system not only verifies the balance of debits and credits in multiple entries derived from the same transaction, but also compares newly generated entries with historically confirmed transaction data. This cross-time dimension check can detect whether account usage follows consistency principles and identify logical contradictions or abnormal deviations in the processing of new transactions compared to similar historical transactions. This method adds a deep verification checkpoint before accounting data enters the formal ledger, proactively intercepting potential accounting logic errors and strengthening the overall consistency and reliability of financial data during multi-period, multi-batch automated processing. Attached Figure Description

[0018] Figure 1 This is a flowchart of a cross-platform automatic financial data calculation method according to the present invention; Figure 2 A flowchart for extracting and formatting the original transaction log fields; Figure 3 A flowchart for matching accounting rules to the economic substance category of a transaction; Figure 4 A chart showing the monthly consistency verification results for business authenticity checks; Figure 5 This is a trend chart of changes in the balance of accounting items. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0020] In the description of this invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, in the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0021] See Figure 1 The original transaction logs contain transaction timestamps, transacting party information, transaction amounts, and transaction remarks. After acquiring the data, field extraction and format cleaning operations are performed on the original transaction logs to eliminate differences in data format, character encoding, and units of measurement from different source systems, generating standardized transaction entries. Subsequently, the standardized transaction entries are fed into a pre-trained text semantic classifier, which parses the economic substance category of the transaction behavior based on the content of the transaction remarks text. Based on the identified economic substance category, the corresponding accounting rule is matched from a pre-set accounting rule knowledge base. Using the matched accounting rule, the standardized transaction entries are decomposed into accounting entries to generate standardized accounting entry drafts. The generated standardized accounting entry drafts are then subjected to cross-checking to examine the debit and credit balance relationships between multiple accounting entry drafts decomposed from the same original transaction log, and to check for any logical conflicts in account continuity with historically recorded transactions.

[0022] See Figure 2In one embodiment of the present invention, the raw transaction records periodically obtained from heterogeneous financial systems have diverse structures. A transaction record from a bank system might appear as "2026-01-15, Merchant A, -550.00, CNY, POS Consumption - Office Supplies Purchase," while another transaction record from an internal reimbursement system might appear as "Date: 2026 / 01 / 16, Payer: Employee B, Amt: 1200.00, Desc: Travel Expenses - High-Speed ​​Rail Ticket." The data processing module identifies the source system identifier for each raw transaction record. The source system identifier can be a system name, interface number, or specific code, and loads the corresponding data parsing template from a pre-set format mapping template library based on the source system identifier. The data parsing template defines how to extract key information from unstructured text or semi-structured data, such as extracting key-value pairs through regular expression matching, delimiter splitting, or JSON path parsing, to extract the raw strings of transaction timestamp, transacting party information, transaction amount, and transaction remarks text. After extraction, the conversion program will uniformly convert the transaction timestamp to Coordinated Universal Time (UTC) format, uniformly convert the transaction amount to the specified base currency, and uniformly convert the character encoding of the transaction party information and transaction remarks text to the universal character set encoding. The conversion process is performed according to the predefined exchange rate table and encoding comparison table.

[0023] In some embodiments, named entity recognition (NAME) further cleanses the transaction party information and transaction remarks text after format and encoding conversion. The NAME model identifies entities in the text, for example, mapping different expressions such as "Merchant A," "Co.Ltd.", and "Company A" to the internally preset standardized entity name "Customer A Limited Company," correcting spelling errors such as "office supplies" to "office supplies," and expanding the abbreviation "HR fees" to "human resources service fees" based on context. The data that has undergone all cleaning and conversion steps is then repackaged according to a preset internal data model, outputting standardized transaction entries. The internal data model defines fixed field names, data types, and length constraints.

[0024] It is understandable that to analyze the economic substance category of transactions, a pre-trained text semantic classifier capable of understanding natural language descriptions is needed. A semantic understanding model based on a deep neural network architecture is constructed, and trained using a massive amount of labeled historical transaction notes and corresponding accounting entries to form a pre-trained text semantic classifier. The massive amount of labeled data comes from historical manually reviewed and archived accounting records. Each data entry contains transaction notes and its final determined economic substance category label. For example, the note "Pay annual software service fee to XX Technology" corresponds to the label "Purchase of intangible assets," and the note "Received sales proceeds from YY Company" corresponds to the label "Sales revenue."

[0025] In practice, the process for parsing economic substance categories is defined. Transaction notes are extracted from standardized transaction entries and input into a pre-trained text semantic classifier. The pre-trained classifier performs word segmentation, vectorization, and contextual semantic encoding on the transaction notes, converting the text into a high-dimensional semantic feature vector that can be processed by a computer. This high-dimensional semantic feature vector is then mapped to a predefined probability space for economic substance categories, calculating the probability of it belonging to the economic substance category of each transaction. This mapping process can be implemented using a fully connected layer and a Softmax function, expressed by the following formula:

[0026] in: This represents the high-dimensional semantic feature vector of the input. and These are the weight matrix and bias vector of the classification layer, respectively. This is the output probability distribution vector, where each element represents the probability that the input text belongs to the corresponding economic substance category. The economic substance category of the transaction with the highest probability value is selected as the final classification result of the standardized transaction item. For example, "POS consumption - catering and entertainment" is classified as "business entertainment expenses", and "online banking transfer - payment of rent" is classified as "rental fee".

[0027] See Figure 3In one embodiment of the present invention, the accounting rule knowledge base is a structured knowledge set embedded in the system, containing structured accounting rule entries derived from accounting standard provisions. For example, the provisions in "Enterprise Accounting Standard No. 14—Revenue" regarding the recognition of revenue for fulfilling performance obligations within a certain period are transformed into structured rules containing accounts such as "Contract Assets" and "Main Business Revenue" and specific amount calculation logic. The accounting rule knowledge base is organized in a directed graph structure. The nodes of the graph represent specific accounting accounts, such as "Cash on Hand," "Bank Deposits," "Accounts Receivable," "Main Business Revenue," and "Selling Expenses." The edges of the graph represent accounting processing logic derived from accounting standard provisions, and each edge is accompanied by an economic substance category label for the applicable transaction behavior. In the graph structure, an edge can connect two or more nodes, and the logic on the edge defines how to decompose one or more input fields to the debit and credit directions of the connected nodes under a specific economic substance category. In some embodiments, the steps of retrieving and filtering accounting rules are standardized. Using the economic substance category of the transaction behavior as the query index, all edges with economic substance category labels are retrieved in the directed graph structure of the accounting rule knowledge base. For example, for a standardized transaction entry with the economic substance category of "employee salary payment," the search process will find all edges whose tags contain "employee salary payment." From the retrieved edges, further filtering is performed to identify accounting rules that match other key attributes in the standardized transaction entry. Key attributes include the transaction entity type and the country or region where the transaction occurred. The transaction entity type attribute might be "domestic parent company," "overseas subsidiary," or "R&D center," while the country or region attribute might be "China," "United States," or "EU." For example, for "equipment procurement," a transaction with the transaction entity type of "R&D center" and the region of "China" might match a specific accounting rule that supports R&D expense deduction, while a transaction with the entity type of "domestic parent company" would match the ordinary fixed asset procurement rule.

[0028] Understandably, a deterministic strategy is needed to select from the multiple filtering entry rules. From these rules, the highest priority rule is chosen as the matching result based on a pre-defined rule priority strategy. This strategy can be defined based on the rule's effective date, the granularity of its applicable scope, or manually specified weights. A common strategy is to select the rule with the most specific description of its applicable scope. The core data fields of the standardized transaction entries are then bound to the matched entry rules, forming an accounting decomposition task to be executed. The binding relationship between the data fields and the entry rules is explicit.

[0029] In practice, the preset rule priority strategy can be quantified and ranked using a defined calculation method. For each selected accounting rule, a priority score is calculated. ,Score The calculation formula is defined as follows:

[0030] Where: symbol This represents the rule specificity score, which is calculated based on the number and specificity of matching attributes in the rule. The more and more specific the matching attributes, the higher the score; (symbol) This represents the timeliness score of the rule; the more recent the rule's effective date, the higher the score. and symbols These are preset weighting coefficients, and they satisfy... Final selection based on score The rule with the highest value is used as the matching result.

[0031] In one embodiment of the present invention, the accounting rules define the specific logic of accounting decomposition. The standardized accounting entry draft is the formal carrier of the decomposition result, including debit accounting subjects, credit accounting subjects, accounting amounts, and the accounting period to which they belong. The logical structure of the accounting rules is parsed to identify the amount calculation fields, subject determination conditions, and debit / credit directions defined in the rules. The amount calculation fields may point to the original amount field or may involve composite calculation logic including parameters such as tax rates and exchange rates. The subject determination conditions are Boolean expressions or mapping tables that determine which specific accounting subject to use. Transaction amounts and other relevant numerical fields are extracted from the standardized transaction entries. Based on the amount calculation logic defined in the accounting rules, the accounting amount to be included in the accounting entry is calculated. For example, for a purchase transaction including tax with a transaction amount of 11,300 yuan, if the matching accounting rule defines the accounting amount calculation logic as "price and tax separation (tax rate 13%)", then the calculated accounting amount included in the "Raw Materials" account is 10,000 yuan, and the accounting amount included in the "Taxes Payable - VAT Payable - Input Tax" account is 1,300 yuan. Based on the attributes of standardized transaction entries and the account determination criteria in the accounting rules, at least two accounting accounts involved in this transaction are identified. Standardized transaction entry attributes include the counterparty's industry, payment method, and product / service type. Account determination criteria may stipulate that the "Management Expenses - Regulatory Fees" account should be used when the counterparty is a "government department" and the transaction includes "administrative fees." Each identified accounting account is designated as a debit or credit account according to the debit / credit direction defined in the accounting rules. The rules define asset accounts such as "Raw Materials" and "Fixed Assets" as debit entries when increases are made, and liability or revenue accounts such as "Accounts Payable" and "Main Business Revenue" as credit entries when increases are made. Based on the transaction timestamps in the standardized transaction entries and combined with the preset accounting period division strategy, the accounting period to which the standardized accounting entry draft belongs is determined. The accounting period division strategy may be a calendar month, a calendar quarter, or a user-defined financial cycle. The debit accounts, credit accounts, accounting amounts, and accounting periods are summarized to generate a complete standardized accounting entry draft.

[0032] In some embodiments, reconciliation verification ensures the mathematical and logical consistency of the draft accounting entries. All debit entries and their corresponding amounts are extracted from the standardized draft accounting entries, and the total debit amount is calculated. For example, if a draft entry contains a debit of 100 yuan in "Bank Deposits" and 200 yuan in "Accounts Receivable," the total debit amount is 300 yuan. All credit entries and their corresponding amounts are also extracted from the standardized draft accounting entries, and the total credit amount is calculated. In the example above, the credit amount for the "Main Business Revenue" account is 300 yuan. The total debit amount is compared with the total credit amount. If the difference exceeds the allowable numerical tolerance range, the draft accounting entry is marked as having a debit / credit imbalance error. The numerical tolerance range is an absolute value threshold set to handle extremely small floating-point rounding errors, for example, 0.01 yuan. For accounting items related to assets, liabilities, and equity, retrieve the ending balance of the accounting item in the previous accounting period from the historical accounting database. For example, the ending balance of "Bank Deposits - RMB Account" was 500,000 yuan in the previous period, and the ending balance of "Accounts Receivable - Customer A" was 100,000 yuan in the previous period. Examine the draft standardized accounting entries containing the accounting items to see if the direction and value of the changes in the accounting amount would lead to situations where the account balance is inconsistent with business logic. If so, mark it as a continuity logic conflict. Inconsistent business logic includes, but is not limited to: a credit balance in the "Cash on Hand" account and a balance in the "Accumulated Depreciation" account exceeding the original value of the corresponding fixed asset.

[0033] It is understandable that the algorithmic expression of reconciliation verification can be described using deterministic formulas. The judgment of whether changes in the accounting entries in a standardized draft account result in logical conflicts in the account balances can be achieved using the following logical expression:

[0034] Where: symbol Represents the ending balance of an accounting item in the previous accounting period; symbol This represents the sum of all accounting amounts recorded on the debit side of this account in a standardized accounting entry draft; symbol This represents the sum of all accounting amounts credited to this account in a standardized accounting entry draft; symbol This represents a predefined set of invalid values ​​that defines the unacceptable range of balance values ​​for a specific accounting item, such as asset accounts. It can contain all values ​​less than zero; sign When the system automatically reviews accounting entries, it determines whether the entry will lead to unreasonable or invalid balances in related accounting accounts, such as a logical flag indicating a negative cash account balance. It generates a verification report containing all error and conflict flags and marks the verified standardized accounting entry drafts as pending review entries.

[0035] In one embodiment of the present invention, risk dimensions are screened for the journal entries to be reviewed based on a preset set of review strategy rules. The set of review strategy rules includes a large transaction review threshold, a list of sensitive counterparties, and abnormal transaction time patterns. The large transaction review threshold is set for different accounting subjects. For example, a single transfer exceeding RMB 500,000 in the "Bank Deposits" subject triggers a review, and a single transfer exceeding RMB 10,000 in the "Management Expenses - Entertainment Expenses" subject triggers a review. The list of sensitive counterparties includes parties requiring special attention, such as affiliated companies, enterprises or individuals listed on regulatory lists. Abnormal transaction time patterns refer to transactions outside of working hours, transactions occurring at the end of the month or quarter, and transactions with amounts close to certain thresholds, as shown in Table 1.

[0036] Table 1: Application Table of Review Strategy Rule Set

[0037] The process of verifying the authenticity of the business transactions to be audited is triggered. Supporting documents, such as scanned copies of electronic invoices and contracts, related to the transactions to be audited are retrieved from the original voucher image database. The origin of the transactions to be audited is traced back to obtain a globally unique identifier (GUID) of the original transaction flow that generated the transactions. This GUID is assigned and bound to standardized flow entries during the field extraction and format cleaning stages. Using the GUID as an index, a full-text search is performed in the original voucher image database to find the storage paths of all associated electronic invoices and scanned copies of contracts. Through document parsing services, the content of the found supporting documents is recognized, and structured data is extracted. This structured data includes invoice code, invoice number, invoice date, amount excluding tax, tax amount, and buyer and seller information. The document parsing service can use OCR technology combined with template recognition.

[0038] In some embodiments, establishing mapping relationships and performing consistency checks is automated. A mapping relationship is established between the structured data of the entries to be audited and supporting documents, and consistency checks are performed item by item according to a preset audit logic. The mapping relationship is explicit; for example, comparing the counterparty "Shanghai ABC Technology Co., Ltd." in the entries to be audited with the seller's name field in the structured data of the electronic invoice, and comparing the breakdown of the accounting amount in the entries to be audited with the tax-exclusive amount and tax amount fields in the structured data of the electronic invoice. The result of each consistency check is recorded, and a structured business authenticity verification report is generated as part of the review conclusion of the entries to be audited.

[0039] It is understandable that the quantitative assessment of consistency verification can be achieved through deterministic rules. The key information in the journal entries to be audited is automatically compared with the supporting documents. Key information includes amount, counterparty, date, and transaction summary. The amount comparison checks whether the journal entry amount matches the total amount in the supporting documents; the counterparty comparison checks whether the counterparty recorded in the journal entry matches the name of the counterparty on the invoice or contract; the date comparison checks whether the transaction date and the invoice date are within a reasonable time window; and the transaction summary comparison checks whether there is semantic relevance between the journal entry summary and the invoice summary. For journal entries to be audited that fail the automatic comparison, or that trigger high-risk rules in the risk dimension screening, they are pushed to the manual review queue, along with links to the relevant original transaction records and supporting documents.

[0040] In practical implementation, the formula for judging the comprehensive result of automatic comparison can be expressed as:

[0041] Where: symbol The amount comparison passed (Boolean value), symbol Represents counterparty comparison passed (Boolean value), symbol Date comparison passed (Boolean value), symbol Represents the logical AND operation; symbol This represents the risk score for this transaction, calculated based on the number and level of high-risk rules matched; [symbol] This represents the risk threshold that allows for automatic approval. The logical expression only applies when all key information comparisons pass and the risk score is below the threshold. Only then is it considered true. For all pending entries that have passed automatic review, approve their formal posting and write the formal accounting entries into the general ledger database, while simultaneously updating the real-time balances of the relevant accounts.

[0042] See Figure 4This chart shows the monthly consistency verification results for business authenticity checks, illustrating the performance of financial transactions during the business authenticity verification phase from January to June. The blue line represents the pass rate, which shows a steady upward trend overall, with only a slight drop in April (86%) before continuing to rise, indicating a continuous improvement in overall verification quality. Amount mismatch failures were the most common type of failure, peaking in January (18 cases) and decreasing to 8 cases in June, showing a clear downward trend. Counterparty mismatch failures were the second most common, decreasing from 10 in January to 4 in June. Date mismatch failures were the fewest, decreasing from 7 in January to 2 in June. The increase in pass rate and the decrease in failures are highly synchronized, indicating that optimization of amount mismatch is the core factor driving the overall pass rate improvement. June saw the lowest number of failures across all categories for the year, indicating the best consistency verification effect. The pass rate slightly declined in April, while the number of amount mismatch failures rebounded slightly; therefore, the quality of transaction data in April requires close monitoring.

[0043] In one embodiment of the present invention, at the end of each accounting period, all formal accounting entries recorded in the current period are extracted from the general ledger database. The end of the accounting period is preset by the system, such as the last day of each month, the last day of each quarter, or the last day of the year. The extracted formal accounting entries are data records that have undergone all the processing steps of the aforementioned embodiments and have been finally approved for recording, including standard fields such as accounting period, accounting subject, debit / credit direction, and accounting amount. All extracted formal accounting entries are classified and summarized according to accounting subject, and the current period debit amount, credit amount, and ending balance of each accounting subject are calculated. The classification and summarization are based on the accounting subject code for grouping and aggregation operations. The calculation logic follows the accounting identity "ending balance = beginning balance + current period increase amount - current period decrease amount". For asset, cost, and expense accounts, the increase is debit and the decrease is credit; for liability, owner's equity, and revenue accounts, the opposite is true. Based on the ending balances of all accounting subjects, a trial balance is generated before adjustments. The total assets in the trial balance are checked to see if they equal the total liabilities and owner's equity. The trial balance is a worksheet that lists the ending balances of all general ledger subjects and is used to preliminarily verify the mathematical correctness of the journal entries.

[0044] In some embodiments, the discrepancy tracing procedure is a systematic method for locating trial balance imbalances. If the trial balance is unbalanced, the discrepancy tracing procedure is initiated to retrospectively check all accounting entries and balance calculations for the current period until the cause of the discrepancy is found and adjusting entries are generated. The discrepancy tracing procedure first verifies the completeness of the data extracted from the general ledger database, confirming that no entries have been omitted. The procedure then recalculates the amount and balance of each account, checking for errors in the grouping and summarizing logic and calculation process. The procedure further sums all accounting entries for the current period separately for debit and credit entries, checking for any entries with debit / credit imbalances that have been incorrectly entered. After the trial balance passes, data is extracted from the general ledger database according to a preset report template, automatically populating the main body of the balance sheet, income statement, and cash flow statement, generating a draft financial statement. The preset report template defines the data retrieval rules and calculation formulas for report items. For example, the "Cash and Cash Equivalents" item in the balance sheet is retrievaled by the sum of the ending balances of the three accounts: "Cash on Hand," "Bank Deposits," and "Other Monetary Funds."

[0045] It is understandable that the automatic filling of financial statement items is accomplished based on deterministic mapping and calculation rules. The filling of the main parts of the balance sheet, income statement, and cash flow statement relies on the mapping relationship between accounting items and statement items, as well as the reconciliation calculations between items. The mapping relationship is stored in the statement template. For example, the "Accounts Receivable" account maps to the "Accounts Receivable" item under "Current Assets" on the balance sheet, and the "Main Business Revenue" account maps to the "Operating Revenue" item on the income statement. The filling of the cash flow statement may be based on the direct method or the indirect method. Under the indirect method, the "Net Cash Flow from Operating Activities" item needs to be calculated from net profit by adding back depreciation, amortization, and adjusting changes in working capital. The entire statement pre-generation process is a series of automated operations of data query, mapping, summarization, and calculation.

[0046] See Figure 5 This is a chart showing the trend of accounting item balances, displaying the changes in the balances of the company's three core current assets from January to June. Cash and cash equivalents fluctuated significantly, exhibiting an "up-down-up" trend, indicating that the company's cash flow was generally ample in the second quarter, but significant pressure to collect funds emerged in March. Accounts receivable showed a steady upward trend overall, with only a slight pullback in April. The continued rise in accounts receivable may indicate a relaxation of credit sales policies or a longer collection cycle, requiring attention to the risk of bad debts. Inventory showed a slow upward trend overall, with the smallest fluctuation range. The continuous increase in inventory balance may reflect increased stockpiling or a slowdown in sales turnover, requiring attention to the risk of inventory backlog. Cash and cash equivalents consistently constitute the largest proportion of current assets, ensuring the company's short-term solvency; however, the simultaneous increase in accounts receivable and inventory may have tied up more working capital.

[0047] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications 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 protection scope of the present invention.

Claims

1. A cross-platform automatic accounting method for financial data, characterized in that, The method includes: Establish communication connections with multiple heterogeneous financial systems, and periodically obtain raw transaction records from the heterogeneous financial systems. The raw transaction records include transaction timestamps, transaction party information, transaction amounts, and transaction remarks text. The obtained original transaction records are subjected to field extraction and format cleaning to eliminate differences in data format, character encoding and unit of measurement between different heterogeneous financial systems, and to generate standardized transaction records. The standardized transaction entries are fed into a pre-trained text semantic classifier, which parses the economic substance category of the transaction behavior based on the content of the transaction notes text. Based on the economic substance category of the transaction, the corresponding accounting rule is matched from the preset accounting rule knowledge base. Using the economic substance category of the transaction as the query index, all edges with economic substance category labels are retrieved in the directed graph structure of the accounting rule knowledge base. The accounting rule that matches other key attributes in the standardized transaction entry is further filtered out. From the filtered accounting rule, the accounting rule with the highest priority is selected as the matching result according to the preset rule priority strategy. The core data fields of the standardized transaction entry are bound to the matched accounting rule to form an accounting decomposition task to be executed. Using the matched accounting rules, the standardized transaction entries are broken down into accounting entries to generate standardized accounting drafts. Perform cross-checking on the standardized accounting entry drafts, including: Extract all debit entries and their corresponding accounting amounts from the standardized accounting entry draft, and calculate the total debit amount; Extract all credit entries and their corresponding accounting amounts from the standardized accounting entry draft, and calculate the total credit amount. The total debit amount is compared with the total credit amount. If the difference exceeds the allowable numerical tolerance range, the draft standardized accounting entry is marked as having a debit-credit imbalance error. For accounting items related to assets, liabilities, and equity, retrieve the ending balance of the accounting item in the previous accounting period from the historical accounting database; Examine the draft standardized accounting entries containing the aforementioned accounting items to determine whether the direction and value of changes in the accounting amounts would lead to situations where the account balances are inconsistent with business logic. If such situations exist, mark them as continuous logic conflicts. Generate a verification report containing all error and conflict markers, and mark the verified standardized accounting entry drafts as pending audit entries; Examine the debit and credit balance among the multiple standardized accounting entry drafts derived from the same original transaction, and check for any logical conflicts in account continuity with historically recorded transactions.

2. The cross-platform automatic financial data accounting method according to claim 1, characterized in that, The process of extracting fields and cleaning the format of the obtained original transaction log includes: Identify the source system identifier of each of the original transaction flows, and load the corresponding data parsing template from the preset format mapping template library according to the source system identifier; Using the data parsing template, key-value pair extraction is performed on the unstructured text or semi-structured data of the original transaction flow to extract the original strings of the transaction timestamp, transaction party information, transaction amount, and transaction remarks text; The extracted transaction timestamps are uniformly converted to Coordinated Universal Time (UTC) format, the transaction amounts are uniformly converted to the specified base currency, and the character encodings of the transaction party information and the transaction remarks text are uniformly converted to the universal character set encoding. After the format and encoding conversion is completed, the transaction party information and the transaction remarks text are processed by named entity recognition to identify and correct spelling errors, inconsistent abbreviations and multiple words with the same meaning. The data from all cleaning and conversion steps are repackaged according to a preset internal data model and output as the standardized flow entries.

3. The cross-platform automatic financial data accounting method according to claim 2, characterized in that, The standardized transaction entries are fed into a pre-trained text semantic classifier, which parses the economic substance category of the transaction behavior based on the content of the transaction notes text, including: A semantic understanding model based on a deep neural network architecture is constructed. The semantic understanding model is trained using a large amount of labeled historical transaction notes and corresponding accounting subjects to form the pre-trained text semantic classifier. Extract the transaction note text from the standardized transaction entries, and input the transaction note text into the pre-trained text semantic classifier; The pre-trained text semantic classifier performs word segmentation, vectorization, and context semantic encoding on the transaction notes text, and outputs a high-dimensional semantic feature vector. The high-dimensional semantic feature vector is mapped to a predefined economic substance category probability space, and the probability value of its belonging to the economic substance category of each transaction behavior is calculated. The economic substance category of the transaction with the highest probability value is selected as the final classification result of the standardized transaction entry.

4. The cross-platform automatic financial data accounting method according to claim 3, characterized in that, Based on the economic substance category of the transaction, the corresponding accounting rule is matched from a preset accounting rule knowledge base, including: The accounting rules knowledge base contains structured accounting rule entries derived from accounting standard provisions; The accounting rule knowledge base is organized in a directed graph structure. The nodes of the graph represent specific accounting subjects, and the edges of the graph represent accounting processing logic derived from the accounting standard clauses. Each edge is accompanied by an economic substance category label of the transaction behavior to which it applies. The key attributes include the type of the transaction entity and the country or region where the transaction takes place.

5. The cross-platform automatic financial data accounting method according to claim 4, characterized in that, The step of using the matched accounting rules to decompose the standardized transaction entries includes: The standardized accounting entry draft includes debit entries, credit entries, accounting amounts, and the corresponding accounting period; The logical structure of the accounting rules is analyzed to identify the amount calculation fields, account determination conditions, and debit / credit directions defined in the rules. The transaction amount and other relevant numerical fields are extracted from the standardized transaction entries, and the accounting amount to be included in the accounting entries is calculated according to the amount calculation logic defined in the accounting rules. Based on the attributes of the standardized transaction entries and the account determination criteria in the accounting rules, at least two accounting accounts involved in this transaction are determined. Each identified accounting item shall be designated as a debit accounting item or a credit accounting item according to the debit and credit directions defined in the accounting rules. Based on the transaction timestamps in the standardized transaction entries and in conjunction with a preset accounting period division strategy, the accounting period to which the standardized accounting entry draft belongs is determined. Summarize the debit and credit accounting entries, the accounting amounts, and the corresponding accounting periods to generate a complete draft of the standardized accounting entry.

6. The cross-platform automatic financial data accounting method according to claim 5, characterized in that, It also includes a step of multi-dimensional review of standardized accounting entry drafts marked as pending audit: Based on a preset set of review strategy rules, the entries to be reviewed are screened for risk dimensions. The set of review strategy rules includes a large transaction review threshold, a list of sensitive counterparties, and abnormal transaction time patterns. Trigger the business authenticity verification process for the entry to be audited, and retrieve supporting documents such as electronic invoices and scanned contracts related to the entry to be audited from the original voucher image database; The key information in the journal entry to be audited is automatically compared with the supporting documents retrieved. The key information includes amount, counterparty, date and transaction summary. For entries that fail the automatic comparison or trigger the high-risk rules in the risk dimension screening, they are pushed to the manual review queue, along with links to the original transaction records and supporting documents. For all pending entries that have passed automatic review, approve their formal posting and write the formal accounting entries into the general ledger database, while updating the real-time balances of the relevant accounts.

7. The cross-platform automatic financial data accounting method according to claim 6, characterized in that, The process for triggering the verification of the authenticity of the entry to be audited includes: The source of the entry to be audited is traced in reverse to obtain the globally unique identifier of the original transaction flow that generated the entry to be audited; Using the globally unique identifier as an index, a full-text search is performed in the original voucher image database to find the storage paths of all associated electronic invoices and contract scanned supporting documents; The document parsing service is used to identify the content of the supporting documents found and extract the structured data from the documents. The structured data includes invoice code, invoice number, invoice date, amount excluding tax, tax amount, and information of the buyer and seller. Establish a mapping relationship between the structured data from the journal entry to be audited to the supporting documents, and perform consistency verification item by item according to the preset audit logic; Record the results of each consistency check and generate a structured business authenticity verification report as part of the review conclusion of the entry to be audited.

8. The cross-platform automatic financial data accounting method according to claim 7, characterized in that, It also includes periodic trial balancing and report pre-generation steps: At the end of each accounting period, all formal accounting entries that have been recorded in the current period are extracted from the general ledger database; All the formal accounting entries extracted are classified and summarized according to the accounting subjects, and the current period debit amount, credit amount and ending balance of each accounting subject are calculated. Based on the ending balances of all accounting items, generate a trial balance before adjustment and check whether the total assets in the trial balance are equal to the total liabilities and owners' equity. If the trial balance is unbalanced, the discrepancy tracing procedure is initiated to retrospectively examine all accounting entries and balance calculation processes for the current period until the cause of the discrepancy is found and adjusting entries are generated. After the trial balance is passed, data is extracted from the general ledger database according to the preset report template, and the main body of the balance sheet, income statement and cash flow statement are automatically filled to generate a draft financial statement.

9. A cross-platform automatic financial data accounting system, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the cross-platform automatic financial data accounting method according to any one of claims 1 to 8.

Images