Automatic classification and input method for intelligently identifying accounting original vouchers

Abstract

The invention relates to the technical field of financial informatization, and particularly discloses an automatic classification and input method for intelligently identifying accounting original certificates. The method comprises the following steps: extracting semantic features and context features in a voucher to form multi-dimensional feature information; inputting the multi-dimensional feature information into a business scene inference model, performing inference and constraint screening based on a pre-constructed accounting business knowledge framework, and outputting a target business scene identifier and an accounting processing rule; and automatically generating and entering an accounting entry according to the rule and the voucher value. The invention further relates to a feedback updating mechanism of the model, and self-adaptive learning is achieved by recording user correction and adjusting model parameters. And realizing hierarchical automatic processing based on a flow routing mechanism of confidence score. The method can deeply understand voucher business semantics, is adaptive to enterprise specific rules, and intelligently cooperates with manual operation, thereby improving the accuracy, efficiency and automation level of accounting voucher processing.

Description

Automatic classification and entry method for intelligent identification of accounting source documents Technical Field

[0001] This invention relates to the field of financial information technology, and more specifically, to a method for automatically classifying and entering accounting original vouchers using intelligent identification. Background Technology

[0002] With the acceleration of enterprise digitalization, the demand for automation in accounting work is becoming increasingly urgent. Among these, the intelligent identification, classification, and entry of massive amounts of original vouchers are key links in realizing the automation and intelligence of financial processes. Traditional manual processing methods are inefficient, error-prone, and costly. Therefore, developing technologies that can automatically understand the content of vouchers and generate accurate accounting entries is of great significance.

[0003] Currently, common automated processing methods mainly rely on a combination of optical character recognition (OCR) technology and predefined rule templates. These methods first extract textual information from vouchers using OCR, then map the text to preset accounting categories based on keyword matching or fixed-format templates. However, this method has shortcomings. First, it relies on a large number of pre-defined keywords and templates, resulting in poor generalization ability for situations with diverse voucher descriptions and complex business scenarios. If unrecorded keywords or new business scenarios appear, the system cannot process them or may lead to misclassification. Second, OCR can only obtain surface text, lacking the ability to understand the underlying business semantics. For example, it cannot distinguish whether "purchased equipment" is used for production or R&D, which directly determines whether it should be recorded as "inventory" or "fixed assets." Third, system rules are usually statically configured, unable to learn from processing results and adapt to changes in enterprise-specific business habits or accounting policies. Maintaining and updating the rule base requires continuous IT support. Fourth, existing solutions are mostly rigid processes, either fully automated (prone to risks due to misjudgment) or completely reliant on manual review (losing automation efficiency), lacking intelligent mechanisms for dynamic process scheduling based on processing confidence levels.

[0004] In summary, existing technologies have limitations in terms of semantic understanding, adaptability, rule maintenance costs, and flexibility of human-machine collaboration, making it difficult to meet enterprises' needs for intelligent financial processing with high accuracy, high adaptability, and security. Therefore, this paper proposes an automatic classification and entry method for intelligent identification of accounting original vouchers to address the above problems. Summary of the Invention

[0005] In order to overcome the above-mentioned defects of the prior art, embodiments of the present invention provide an automatic classification and entry method for intelligent identification of accounting original vouchers, so as to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: an automatic classification and entry method for intelligent identification of accounting original vouchers, comprising the following steps: S1, feature extraction step: extracting multi-dimensional feature information from accounting original vouchers, wherein the multi-dimensional feature information includes semantic features identified from the voucher text content and contextual features obtained from voucher attributes and related information; wherein the semantic features are used to characterize the core facts of economic matters, and the contextual features are used to define and correct the business nature of the core facts.

[0007] S2, Scenario Inference Step: Input the multi-dimensional feature information into the business scenario inference model, perform reasoning based on the pre-built accounting business knowledge framework, and output the target business scenario identifier corresponding to the voucher and the associated accounting processing rules. The accounting business knowledge framework defines the mapping relationship between business scenarios, feature information, and accounting processing rules; thus completing the mapping from voucher features to standardized accounting scenarios.

[0008] S3, Journal Entry Generation and Entry Steps: Based on the accounting processing rules and the numerical information extracted from the vouchers, accounting entries are automatically generated and entered; this automates accounting recognition, measurement, and recording.

[0009] Furthermore, step S1 specifically includes: identifying and constructing a structured semantic relationship representing the subject, action, and object of an economic event by parsing the descriptive text of the voucher, as the semantic feature; the structured semantic relationship is represented in the form of a triple: the logic of subject-object-operation.

[0010] By identifying the voucher type, querying the basic information of the transacting party, and obtaining the associated business process metadata, auxiliary information is obtained to define the nature of the semantic relationship business, which serves as the context feature. The voucher type is determined by layout matching or image classification, the transacting party information is obtained by querying through an external interface, and the business process metadata is obtained by querying the internal system through the associated business order number.

[0011] Furthermore, in step S2, the accounting business knowledge framework is implemented through a feature-scenario mapping model; this model can be a rule engine or a machine learning classifier, which encodes the correspondence between feature combinations and business scenarios.

[0012] The reasoning process includes: combining the structured semantic relationship with the auxiliary information to match at least one initial business scenario in the feature-scenario mapping model; filtering and confirming the initial business scenario based on the business constraints contained in the auxiliary information to obtain the target business scenario identifier; and storing the business constraints and the scenario identifier together to exclude candidate scenarios that do not conform to the current business background.

[0013] Furthermore, the method also includes a model update step S4, which constitutes a feedback learning closed loop: S41, storage step: storing the multi-dimensional feature information, the output target business scenario identifier, and subsequent user correction operations on the inference result based on each execution of step S2; S42, comparative analysis step: comparing and analyzing the correction operation with the corresponding inference result, and inferring the correct business scenario based on the corrected entry to generate a model optimization sample; S43, adjustment step: using the model optimization sample, adjusting the feature-scenario mapping relationship or decision parameters in the accounting business knowledge framework so that the model output tends to the user's correction result.

[0014] Furthermore, step S43 employs a differentiated adjustment strategy, including: analyzing the correction operation, and determining it as a critical correction if it results in changes to primary items in the accounting entries that affect the balance sheet or income statement; critical corrections reflect a fundamental misjudgment of the substance of the economic transaction.

[0015] For the mapping relationship or parameters corresponding to the feature combination that triggers the key correction, reinforcement learning or weight updates are performed with priority over other adjustments; this strategy ensures that learning resources are focused on correcting the most important classification errors.

[0016] Furthermore, the generation of accounting entries in step S3 also includes an accounting policy adaptation sub-step: in the feature-scenario mapping model, configurable accounting policy options are bound to the target business scenario identifier; the policy options include measurement methods, amortization periods, etc.

[0017] Based on the accounting policies pre-configured for the company, select a specific policy from the configurable accounting policy options and apply the policy to calculate the voucher values ​​to determine the journal entry account and amount; for example, automatically calculate the monthly depreciation of fixed assets and create asset cards based on the selected depreciation policy.

[0018] Furthermore, step S3 also includes an internal control verification sub-step, embedding the control point into the automated process: automatically triggering the internal control verification program associated with the target business scenario identifier; executing the internal control verification program, which includes: verifying whether the voucher exists in a completed electronic approval process, or verifying whether the cumulative value of this business transaction exceeds the authorized limit of the associated budget project; the verification result determines whether the process continues or an additional warning is issued.

[0019] Furthermore, in step S2, when the business scenario inference model outputs the target business scenario identifier, it simultaneously outputs a confidence score calculated based on feature matching degree and constraint satisfaction degree; the matching degree and satisfaction degree together reflect the degree of certainty of this inference.

[0020] Step S3 automatically selects to perform fully automatic data entry, generate a task to be reviewed and entered after manual confirmation, or transfer it to fully manual processing based on the different threshold ranges in which the confidence score is located. This mechanism realizes hierarchical automation based on determinism, balancing efficiency and risk.

[0021] Furthermore, the information stored in step S41 also includes the intermediate candidate scenario set and its evaluation data during the inference process; these data record the alternative paths and their evaluations before the model makes the final decision.

[0022] The objective of adjusting the mapping relationship or decision parameters in step S43 includes making the evaluation data of the intermediate candidate scenario set consistent with the correct reasoning path implied by the correction operation; this optimization improves the rationality of the internal reasoning logic of the model, rather than just the correctness of the final result.

[0023] Furthermore, the method also includes a historical data processing optimization step S5, which uses an evolved model to improve the quality of historical data: S51, re-inference step: after the business scenario inference model is updated, the storage feature information of historically processed vouchers is re-inferred based on the updated model; S52, comparison step: the new business scenario identifier obtained by re-inference is compared with the business scenario identifier of the original record; S53, prompt generation step: when the two are inconsistent and the confidence score of the new identifier is higher than the preset confidence threshold of the original record identifier, a data optimization prompt containing details of the difference is generated; this step supports continuous review and optimization of historical accounting records.

[0024] The technical effects and advantages of this invention are as follows: Compared with existing technologies, this invention constructs and extracts feature combinations containing structured semantic relationships and multidimensional contextual information, and uses this to drive a dynamically evolving accounting business knowledge framework for reasoning. Specifically, the system not only recognizes text, but also constructs triples such as "(subject, action, object)" through semantic analysis to represent core business facts, while integrating contextual information such as voucher type, transaction party attributes, and business processes as qualitative basis. This composite feature is input into a feature-scenario mapping model, which outputs a defined business scenario through matching and filtering based on business constraints. This approach enables the system to understand the substance of economic matters, thereby improving the accuracy and robustness of classification when faced with diverse voucher descriptions and complex business scenarios, and reducing reliance on fixed keywords and templates.

[0025] Compared to existing technologies, this invention introduces a model update with a feedback learning mechanism, enabling the system to continuously self-optimize. The system automatically records the features used in each inference, the output results, and the user's final corrections. Through comparative analysis, it generates optimized samples and adjusts the mapping relationships or decision parameters in the knowledge framework accordingly. The system can identify critical corrections involving changes to core accounting subjects and prioritize and reinforce such errors. This process allows the system to learn from historical experience, gradually adapting to the accounting habits, business characteristics, and accounting policy updates of different enterprises, reducing the cost of manual rule maintenance in the later stages.

[0026] Compared to existing technologies, this invention generates a quantified confidence score simultaneously during the inference process and intelligently routes the processing flow based on a preset threshold. The confidence score integrates feature matching and business constraint satisfaction, objectively reflecting the reliability of a single judgment. Based on this, the system automatically makes decisions: high-confidence results are automatically entered; medium-confidence results generate suggested tasks for manual review; and low-confidence results are transferred to fully manual processing. Dynamic risk assessment and control nodes are embedded in the process, ensuring processing efficiency under high determinism while allowing for timely manual intervention when the system is uncertain. Overall, this achieves an optimal balance between efficiency and risk control, enhancing the practicality and security of the automated system. Attached Figure Description

[0027] Figure 1 is a flowchart of the automatic classification and entry method for intelligent identification of accounting original vouchers according to the present invention.

[0028] Figure 2 is a flowchart of the model feedback update and adaptive learning process of the present invention.

[0029] Figure 3 is a diagram of the confidence score-driven process routing branching mechanism of the present invention. Detailed Implementation

[0030] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0031] Example 1 illustrates the automatic classification and entry method for intelligent identification of accounting source documents, as shown in Figures 1 to 3. The following description is based on the implementation process. This method extracts the structured semantic relationships and contextual features of the documents to construct multi-dimensional feature information, and performs reasoning and constraint filtering based on a pre-built accounting business knowledge framework to achieve automatic processing from source documents to accounting entries.

[0032] S100, Feature extraction step.

[0033] This step transforms the unstructured data of accounting source documents into structured multidimensional feature information. The input is a digital image of the source document. The output is multidimensional feature information, which consists of semantic features and contextual features.

[0034] S110, extract semantic features.

[0035] The system receives the original voucher image and uses optical character recognition (OCR) technology to convert the text regions in the image into machine-readable text character sequences. Subsequently, natural language processing is performed on the acquired text to parse the fields describing the economic transactions.

[0036] The specific implementation process involves analyzing text statements using a semantic role labeling model. This model identifies the core predicate in the statement, which describes the action of economic behavior. Simultaneously, it identifies the core arguments associated with this predicate, including the agent, receiver, and object of the action.

[0037] For example, in the text “Pay network service fee to Company A”, the predicate “pay” is identified, the agent is “this company”, the recipient is “network service fee”, and the target is “Company A”.

[0038] The system will construct these identified elements into one or more triples of the form "(entity, relation, entity)".

[0039] For example, construct triples (this company, payment, network service fee) and (payment, target, company A). This set of triples constitutes a structured semantic relationship representing the content of the economic transaction, serving as a semantic feature for subsequent processing.

[0040] S120, extract contextual features.

[0041] Contextual features are used to define and correct the business facts described by semantic features. Their extraction is achieved through the following methods: First, voucher type identification. The system compares the layout features of the voucher image with a pre-stored voucher template library, or uses a trained image classification model to determine whether the voucher belongs to a specific type such as a VAT invoice, taxi receipt, or bank statement.

[0042] Second, transaction information query. The system extracts the unified social credit code or full name of the counterparty from the identified text, using it as the query key to call the application programming interface (API) of an external business information database. This API returns structured data, from which the system extracts attribute tags such as the industry classification of the transaction party.

[0043] Third, business process linkage. The system queries the enterprise's internal business management system or financial reimbursement system database using the business order number associated with the voucher. This query retrieves metadata such as the organizational department where the transaction occurred, the project number, and the budget type.

[0044] The information on bill type, industry attributes of the transacting party, business department, and budget type generated by the above methods together constitutes the contextual features.

[0045] S200, Scene Inference Steps.

[0046] This step determines the accounting business scenario corresponding to the voucher based on the multi-dimensional feature information output by S100. The core of the processing is a business scenario inference model and the accounting business knowledge framework it relies on. The output is a specific target business scenario identifier and a set of accounting processing rules bound to that identifier.

[0047] S210, Knowledge Framework and Reasoning Execution.

[0048] The accounting business knowledge framework is a structured knowledge representation pre-installed in the system. This framework defines the mapping relationships between business scenarios, judgment characteristics, and accounting rules.

[0049] In a specific implementation, the framework exists in the form of a feature-scene mapping model. This model can be a rule engine that stores a large number of production rules in the form of "if-then" statements; or it can be a machine learning classification model trained through supervised learning.

[0050] At the start of the reasoning process, the system combines and vectorizes the structured semantic relations generated in S110 with the contextual features generated in S120 to form a unified feature vector. .

[0051] Subsequently, the feature vector The input feature-scene mapping model is used for matching. The matching process calculates the correlation between the vector and each predefined business scenario and generates one or more initial business scenario candidates.

[0052] S220, Constraint Filtering and Scenario Confirmation.

[0053] The reasoning process includes a screening and confirmation phase based on business constraints. In the accounting business knowledge framework, in addition to the feature-scenario mapping relationship, contextual constraints that must be met are defined for each business scenario. These constraints are stored as logical expressions associated with scenario identifiers.

[0054] The system examines each initial candidate scenario obtained from S210 and determines whether the contextual features of the current credential satisfy all the constraints required by the scenario.

[0055] For example, a scenario called "Procurement of Fixed Assets for R&D" is subject to the constraints that "the department to which the expense belongs belongs to the R&D category" and "the budget type is capital expenditure".

[0056] The system compares the contextual characteristics of the current voucher with these constraints. If the contextual characteristics of the current voucher indicate that its department is "Sales Department" or its budget type is "Expense", then the candidate scenario is excluded.

[0057] After constraint filtering, scenarios that meet all constraints are identified as target business scenario identifiers. If more than one scenario meets the conditions, one is selected as the final output based on preset priority rules or initial matching score.

[0058] S230, Confidence Assessment and Process Routing.

[0059] This step includes confidence assessment and process routing sub-steps. When outputting the target business scenario identifier, the business scenario inference model simultaneously calculates and outputs a quantified confidence score (Conf).

[0060] The confidence score Conf is the feature matching degree. With constraint satisfaction The synthesis function, where It reflects the degree of matching between the feature vector F and the winning scenario pattern. Both reflect the proportion of business constraints satisfied, and their value range is [0,1]. In one example... .

[0061] The system presets two thresholds: a high confidence threshold. (e.g., 0.90) and low confidence threshold (For example, 0.65). The system automatically selects the subsequent processing path based on the Conf value: if If the process proceeds to the subsequent S300 step, fully automated entry generation and input will be performed; if If the system generates a task to be reviewed, this task includes the current voucher image, extracted multi-dimensional feature information, inferred target business scenario identifier, confidence level (Conf), and system-generated suggested accounting entries. This task is sent to the manual review queue, awaiting confirmation before triggering step S300; if If so, the system will transfer all information of the voucher to the manual processing workbench.

[0062] S240, Internal control rule verification.

[0063] This step also includes an internal control rule verification sub-step. After determining the target business scenario identifier, the system automatically triggers the internal control verification procedure associated with that identifier.

[0064] For example, in the scenario of "large outward payment", the system calls the application programming interface of the approval process system to verify whether the electronic approval process associated with the payment has been fully executed and finally approved. The verification result will determine whether the process continues or whether a warning mark is added to the generated accounting voucher.

[0065] S300, Journal Entry Generation and Entry Steps.

[0066] This step executes accounting record operations based on the target business scenario identifier and associated accounting rules output by S200. Inputs also include numerical information extracted from original vouchers. Output is the standard accounting entry already recorded in the accounting information system.

[0067] S310, Accounting Policy Adaptation and Journal Entries Generation.

[0068] This step includes an accounting policy adaptation sub-step.

[0069] Within the accounting business knowledge framework, each target business scenario identifier is bound to a set of configurable accounting policy options and parameters. The system reads the global accounting policy configuration file preset for the enterprise and determines the specific policy parameters to be used in the current scenario.

[0070] For example, in the scenario of "purchasing fixed assets for production and operation," the policy options bound within the framework include depreciation methods and residual value rate ranges. The company configuration file specifies the use of the "straight-line method," a residual value rate of 0.05, and an expected useful life of 5 years for this type of asset.

[0071] The system first generates basic accounting entries based on the standard account template corresponding to the scenario and the tax-exclusive amount on the voucher.

[0072] Subsequently, the system calls the internal asset measurement module, based on the original value of the assets. (Amount excluding tax), selected depreciation method, and depreciation period (5 years) and residual value rate (0.05), calculate monthly depreciation. .

[0073] The calculation formula is: The system automatically creates a fixed asset card, recording the asset information and depreciation calculation rules.

[0074] S400, Model Update Steps.

[0075] This step is a background learning process independent of the main processing flow, enabling the business scenario inference model to adaptively adjust and optimize based on user feedback.

[0076] S410, Feedback Data Recording.

[0077] After each execution of steps S200 and S300, the system associates and stores the key data of this processing.

[0078] These data include at least: the complete multidimensional feature information upon which this inference is based. The target business scenario identifier output by the model Accounting entries automatically generated by the system .

[0079] When subsequent users conduct verification or payment processing... After the modifications are made and finally confirmed, the system records the final accounting entries confirmed by the user. .

[0080] S420, Comparative Analysis and Model Adjustment.

[0081] The system uses a background process to periodically check stored records and filter them. and For each inconsistent entry, the system bases its approach on... The accounting logic it embodies leads, in turn, to the user-accepted and correct business scenario identifier. .

[0082] Therefore, the system generates a form of Supervised learning samples.

[0083] The system uses newly accumulated supervised learning samples to incrementally adjust the business scenario inference model. If the model is a machine learning-based classifier, these new samples are added to the training set to initiate an incremental training or fine-tuning round. If the model is a rule-based engine, existing rules may be added or modified based on the samples.

[0084] S430, a differentiated adjustment strategy.

[0085] The model adjustment process employs a differentiated adjustment strategy.

[0086] The system first analyzes the user's correction operation: if the correction only involves auxiliary accounting information in the accounting entries and does not affect the main primary accounting subjects, it is classified as a normal correction; if the correction results in changes to the primary accounts of the balance sheet or income statement in the accounting entries, it is determined to be a critical correction.

[0087] During the model tuning phase, for supervised learning samples triggered by key corrections, the system assigns higher learning weights or applies larger adjustments. For example, in the training loss function of a machine learning model... In the middle, the loss term of the key corrected sample Multiplied by a coefficient greater than 1 ,Right now .

[0088] S440, inference process optimization.

[0089] In the S410 data record, the system also saves the intermediate data generated by the business scenario inference model during this inference process, specifically including the top-ranked data generated by the matching in the S210 stage. Each initial candidate scenario and its respective evaluation score.

[0090] During the model tuning phase of S420, the optimization objective is not limited to making the model more accurate to the samples. The output is correct, and it also extends to making the reasoning process inside the model more reasonable.

[0091] Specifically, the adjustment process attempts to optimize the model so that the input features... At that time, the correct scenario The scenarios in the intermediate candidate scenario list should be ranked as high as possible, and their evaluation scores should be higher than those of other erroneous scenarios. This can be achieved by introducing a ranking loss term into the objective function.

[0092] S500, historical data optimization steps.

[0093] After the business scenario inference model completes multiple rounds of updates according to the S400 steps, a historical data quality optimization process can be initiated, which is an offline batch processing job.

[0094] S510, re-inferred.

[0095] The system reads the characteristic information of all processed vouchers stored in S410 during a specific accounting period in the past. .

[0096] Using the latest updated business scenario inference model, this historical feature information is analyzed. Perform batch re-inference to obtain a new business scenario identifier for each voucher. and their corresponding confidence scores .

[0097] S520, comparison and optimization suggestion generation.

[0098] The system will compare the re-inferred results with the original records. For each voucher, its new business scenario identifier will be checked. Does it match the business scenario identifier in the original record? Inconsistent, and new confidence scores Is it higher than the original confidence level? With a preset stability threshold The sum of (e.g., 0.15) is the judgment condition. and Whether it is valid or not.

[0099] For all voucher records that meet the above conditions, the system automatically generates a structured data quality optimization suggestion report. The report lists the voucher number, accounting period, original scenario identifier and entry, new suggested scenario identifier and entry, and a comparison of the old and new confidence levels.

[0100] This report was pushed to the review interface to provide a basis for decision-making regarding retrospective adjustments to historical accounts.

[0101] The above description is merely a specific implementation of this solution. Any equivalent transformations or modifications based on this technical solution and concept should be included within the declared scope of protection.

[0102] Finally, several points should be noted: First, in the description of this application, it should be noted that, unless otherwise specified and limited, the terms "installation," "connection," and "linkage" should be interpreted broadly, and can refer to mechanical or electrical connections, or internal connections between two components, or direct connections. Terms such as "upper," "lower," "left," and "right" are only used to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may change. Second, the accompanying drawings of the embodiments disclosed in this invention only involve structures relevant to the embodiments disclosed in this invention; other structures can refer to common designs. Where there is no conflict, the same embodiment and different embodiments of this invention can be combined with each other. Finally, the above descriptions are merely preferred embodiments of this invention and are not intended to limit this invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A method for automatically classifying and entering accounting source documents using intelligent identification, characterized in that: Includes the following steps: S1, Feature Extraction Step: Extract multi-dimensional feature information from the original accounting vouchers. The multi-dimensional feature information includes semantic features identified from the voucher text content and contextual features obtained from voucher attributes and related information. S2, Scenario Inference Step: Input the multi-dimensional feature information into the business scenario inference model. Based on a pre-built accounting business knowledge framework, perform inference and output the target business scenario identifier corresponding to the voucher and the associated accounting processing rules. The accounting business knowledge framework defines the mapping relationship between business scenarios, feature information, and accounting processing rules. S3, Journal Entry Generation and Input Step: Automatically generate and input accounting entries based on the accounting processing rules and the numerical information extracted from the vouchers.

2. The method for automatic classification and entry of intelligent accounting source vouchers according to claim 1, characterized in that, Step S1 specifically includes: identifying and constructing a structured semantic relationship representing the subject, action, and object of an economic transaction by parsing the descriptive text of the voucher, as the semantic feature; and obtaining auxiliary information for defining the business nature of the semantic relationship by identifying the voucher type, querying the basic information of the transacting parties, and related business process metadata, as the context feature.

3. The method for automatic classification and entry of intelligent accounting source documents according to claim 2, characterized in that, In step S2, the accounting business knowledge framework is implemented through a feature-scenario mapping model; the reasoning process includes: combining the structured semantic relationship with the auxiliary information, matching at least one initial business scenario in the feature-scenario mapping model; and filtering and confirming the initial business scenario based on the business constraints contained in the auxiliary information to obtain the target business scenario identifier.

4. The method for automatic classification and entry of intelligent accounting source documents according to claim 1, characterized in that, The method further includes a model update step S4: S41, and a storage step: storing the multi-dimensional feature information on which step S2 is based each time, the output target business scenario identifier, and the subsequent user correction operation for the inference result; S42, Comparative analysis step: Compare and analyze the correction operation with the corresponding inference result to generate model optimization sample; S43, Adjustment step: Use the model optimization sample to adjust the feature-scenario mapping relationship or decision parameters in the accounting business knowledge framework.

5. The method for automatic classification and entry of intelligent accounting source documents according to claim 4, characterized in that, Step S43 employs a differentiated adjustment strategy, including: analyzing the correction operation, and determining it as a critical correction if it results in changes to the primary items of the balance sheet or income statement in the accounting entries; and implementing reinforcement learning or weight updates that take precedence over other adjustments for the mapping relationship or parameters corresponding to the feature combination that triggers the critical correction.

6. The method for automatic classification and entry of intelligent accounting source vouchers according to claim 3, characterized in that, The step S3 of generating accounting entries also includes an accounting policy adaptation sub-step: in the feature-scenario mapping model, a configurable accounting policy option is bound to the target business scenario identifier; according to the accounting policy pre-configured for the enterprise, a specific policy is selected from the configurable accounting policy option, and the policy is applied to calculate the voucher value to determine the entry account and amount.

7. The method for automatic classification and entry of intelligent accounting source documents according to claim 1, characterized in that, Step S3 further includes an internal control verification sub-step: automatically triggering an internal control verification program associated with the target business scenario identifier; executing the internal control verification program, which includes: verifying whether the voucher exists in a completed electronic approval process, or verifying whether the cumulative value of this business transaction exceeds the authorized limit of the associated budget project.

8. The method for automatic classification and entry of intelligent accounting source documents according to claim 1, characterized in that, In step S2, when the business scenario inference model outputs the target business scenario identifier, it simultaneously outputs a confidence score calculated based on feature matching degree and constraint satisfaction degree; in step S3, depending on the different threshold ranges in which the confidence score is located, it automatically selects to perform fully automatic data entry, generate a task to be reviewed and manually confirmed before data entry, or transfer it to fully manual processing.

9. The method for automatic classification and entry of intelligent accounting source documents according to claim 4, characterized in that, The information stored in step S41 also includes the intermediate candidate scenario set and its evaluation data in the inference process; the goal of adjusting the mapping relationship or decision parameters in step S43 includes making the evaluation data of the intermediate candidate scenario set consistent with the correct reasoning path implied by the correction operation.

10. The method for automatic classification and entry of intelligent accounting source documents according to claim 4, characterized in that, The method further includes a historical data processing optimization step S5: S51, re-inference step: after the business scenario inference model is updated, the storage feature information of the historically processed vouchers is re-inferred based on the updated model; S52, comparison step: the new business scenario identifier obtained by re-inference is compared with the business scenario identifier of the original record; S53, prompt generation step: when the two are inconsistent and the confidence score of the new identifier is higher than the preset confidence threshold of the original record identifier, a data optimization prompt containing details of the difference is generated.

Images