Accounting data processing method and system

Abstract

The invention discloses an accounting data processing method and system, and relates to the technical field of accounting data intelligent processing, and the method comprises the steps: collecting the internal original accounting data of an enterprise, and generating an original accounting data set; performing preprocessing to generate a structured accounting data set; inputting the structured data set into an artificial intelligence classification model, and outputting the classified structured data set; automatically filing the data according to the category identifier; performing data verification, identifying abnormal data, and generating a verification result; performing abnormal cause analysis on the abnormal data, and automatically generating a corresponding processing suggestion according to an analysis result; and processing the abnormal data according to the processing suggestion, and recording the processing suggestion and an execution result. According to the invention, intelligent automation of the whole process from collection, classification, filing and verification to exception processing of the accounting data is realized, the data processing efficiency and accuracy are improved, the labor cost and risk are reduced, closed-loop management and traceability of the exception data are realized, and the method is suitable for various types of complex accounting data scenes.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent processing of accounting data, and in particular to an accounting data processing method and system. Background Art

[0002] In recent years, with the continuous advancement of information technology and the digital transformation of enterprises, the electronic and automated management of accounting data has become a key development direction for corporate financial management. Modern accounting information systems enable the collection, storage, and preliminary processing of multiple types of data, including electronic vouchers, invoices, bank statements, and financial statements, providing a data foundation for business analysis, risk control, and compliance supervision. At the same time, emerging technologies such as artificial intelligence and big data analytics have been initially applied in scenarios such as financial automation, intelligent auditing, and intelligent report generation, driving the upgrade of accounting data processing from traditional manual methods to intelligent, integrated, and automated methods. However, with the rapid increase in the scale of corporate business and the complexity of data, how to more efficiently, accurately, and securely process and manage large-scale, multi-source, and multi-structured accounting data has become a critical issue that urgently needs to be addressed.

[0003] While existing technologies have made some progress in accounting data collection, storage, and partial automation, significant shortcomings remain. First, existing methods rely heavily on rule templates and manual configuration. Given the diverse and frequently changing formats and data types of vouchers, their automated classification and structured processing capabilities are limited, making it difficult to achieve high-accuracy and efficient batch processing. Second, existing data archiving and verification processes generally rely on static rules and sample reviews, lacking dynamic intelligent verification and anomaly identification capabilities based on the full data set. This results in delayed financial risk identification and makes it difficult to promptly detect and trace data errors and anomalies. Furthermore, current processing of anomaly data relies heavily on manual judgment, lacking intelligent analysis and decision-making support, making it difficult to achieve automated anomaly attribution and closed-loop management, impacting financial compliance and data security. More importantly, existing technologies lack an end-to-end, full-process intelligent accounting data management and control system. They are unable to achieve an integrated and coordinated approach to data collection, structuring, intelligent classification, archiving, verification, anomaly analysis, and the automatic generation of action recommendations. This results in low accounting data processing efficiency and challenges in ensuring accuracy and security.

[0004] Therefore, it is urgent to propose an accounting data processing method that can achieve full-process automation, intelligence and integration to solve the above pain points. Summary of the Invention

[0005] In view of the above-mentioned problems, the present invention is proposed.

[0006] Therefore, the technical problems solved by the present invention are: the existing accounting data processing methods have limited classification and structuring capabilities, insufficient intelligent verification and anomaly identification, low levels of automation in anomaly attribution and processing, and how to achieve intelligent and automatic processing of the entire accounting data process.

[0007] In order to solve the above technical problems, the present invention provides the following technical solutions: In a first aspect, an embodiment of the present invention provides an accounting data processing method, comprising collecting original accounting data within an enterprise to generate an original accounting data set; Preprocessing the original accounting data set to generate a preprocessed structured accounting data set; Inputting the preprocessed structured accounting data set into a pre-trained artificial intelligence classification model and outputting a classified structured data set; Automatically archiving the classified structured data set according to the category identifier to obtain an archived structured data set; Perform data verification on the archived structured data set, identify abnormal data, and generate verification results; For abnormal data in the verification results, a preset intelligent analysis model is used to analyze the cause of the abnormality, and corresponding processing suggestions are automatically generated based on the analysis results; Process the abnormal data according to the processing suggestions, and record the processing suggestions and execution results.

[0008] As a preferred embodiment of the accounting data processing method of the present invention, the original accounting data includes electronic vouchers, scanned images of paper documents, electronic invoices, bank statements, contract texts, financial statements, payment details, and inventory records; The original accounting data are aggregated to form an original accounting data set.

[0009] As a preferred solution of the accounting data processing method described in the present invention, the preprocessing of the original accounting data set includes format conversion, text recognition, data deduplication, missing value processing, data consistency verification, field extraction and data standardization of the original accounting data set to obtain a preprocessed structured accounting data set.

[0010] As a preferred solution of the accounting data processing method of the present invention, the pre-trained artificial intelligence classification model is a deep neural network classification model based on multimodal data feature fusion, specifically including: Inputting the text information and layout features of each piece of data in the preprocessed structured accounting data set into the embedding layer to generate corresponding text feature vectors and layout feature vectors; fusing the text feature vector and the layout feature vector into a unified multimodal feature vector; Inputting the multimodal feature vector into multiple convolutional neural network layers and attention mechanism layers in series to perform deep feature extraction and output a fused deep feature representation; The fused deep feature representation is input into the fully connected classification layer, the category probability distribution is calculated by the classification layer, the category with the largest probability is selected as the classification result of the corresponding data, and the classified structured data set is output.

[0011] As a preferred solution of the accounting data processing method described in the present invention, the archived structured data set includes automatically allocating and storing each classified structured accounting data in the classified structured data set to a corresponding archive group according to its corresponding category identifier, dividing the archive group according to the category identifier, and aggregating the data of each archive group to form an archived structured data set that can be indexed, retrieved and managed based on category.

[0012] As a preferred solution of the accounting data processing method described in the present invention, the data verification of the archived structured data set includes automatically comparing the data of the archived structured data set with historical data based on preset verification rules. The verification rules include amount consistency verification, date legitimacy verification, account correspondence verification, voucher number uniqueness verification and cross-checking with external accounting data. If data inconsistency, missing, duplication and abnormal fluctuation are detected, they are marked as abnormal data, and a verification result containing abnormal type, abnormal location and abnormal description information is generated.

[0013] As a preferred solution of the accounting data processing method described in the present invention, the generating of corresponding processing suggestions includes extracting the associated accounting subjects, voucher contents, business types and historical data features of the data marked as abnormal in the verification results, inputting the extracted information into an intelligent analysis model constructed based on a decision tree and anomaly detection algorithm, automatically classifying the causes of the abnormal data, and outputting anomaly cause labels; according to the anomaly cause labels and the knowledge rule base built into the model, automatically matching and generating processing suggestions for different anomaly types.

[0014] As a preferred solution of the accounting data processing method described in the present invention, the recording of processing suggestions and execution results includes automatically performing corresponding processing operations on abnormal data according to the processing suggestions generated by the intelligent analysis model, and recording the adopted processing suggestions, processing time, processing person and system identification and processing result information in a structured manner after the processing is completed, forming a complete processing log that can be used for subsequent tracing, auditing and statistical analysis.

[0015] In a second aspect, an embodiment of the present invention provides an accounting data processing system, comprising: Original data collection module: collects original accounting data within the enterprise and generates original accounting data sets; Data preprocessing module: preprocessing the original accounting data set to generate a preprocessed structured accounting data set; Intelligent classification module: inputs the pre-processed structured accounting data set into a pre-trained artificial intelligence classification model and outputs a classified structured data set; Automatic archiving module: automatically archiving the classified structured data set according to the category identifier to obtain an archived structured data set; Data verification module: performs data verification on the archived structured data set, identifies abnormal data, and generates verification results; Intelligent anomaly analysis module: For abnormal data in the verification results, a preset intelligent analysis model is used to analyze the cause of the anomaly and automatically generate corresponding treatment suggestions based on the analysis results; Exception processing and recording module: processes the exception data according to the processing suggestions, and records the processing suggestions and execution results.

[0016] Beneficial effects of the present invention: The present invention realizes the full-process automated processing of accounting data from collection, preprocessing, intelligent classification, automatic archiving, intelligent verification to exception analysis and generation of processing suggestions by integrating artificial intelligence classification models and intelligent analysis models. Compared with the existing technology, the present invention significantly improves the automation level and intelligent capabilities of accounting data processing, effectively reduces the intensity of manual operation and the risk of errors, and greatly improves the efficiency and accuracy of data archiving, verification and exception processing. At the same time, the intelligent analysis model is used to automatically attribute abnormal data and generate processing suggestions, realizing closed-loop management of abnormal data and enhancing the compliance, security and traceability of financial data. The present invention can adapt to multi-source, multi-type and complex structure accounting data environments, has a high degree of versatility and scalability, and provides efficient, intelligent and secure data support for corporate financial management and risk prevention and control. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. Those skilled in the art can also derive other drawings based on these drawings without inventive work, among which: Figure 1 An overall flow chart of an accounting data processing method provided for the first embodiment of the present invention. DETAILED DESCRIPTION

[0018] To make the above-mentioned objects, features, and advantages of the present invention more clearly understood, the following detailed description of the specific embodiments of the present invention is given in conjunction with the accompanying drawings. It is obvious that the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in this field without creative work should fall within the scope of protection of the present invention.

[0019] Example 1, reference Figure 1 , as an embodiment of the present invention, provides an accounting data processing method, comprising: S1: Collect the original accounting data within the enterprise and generate the original accounting data set.

[0020] Original accounting data includes electronic vouchers, scanned images of paper documents, electronic invoices, bank statements, contract texts, financial statements, payment details, and inventory records; The original accounting data are aggregated to form an original accounting data set.

[0021] In the embodiments of the present invention, "raw accounting data" refers to various types of business data generated, circulated, and archived during an enterprise's daily operations, financial accounting, and management activities. Specifically, it includes, but is not limited to, electronic vouchers (such as system-generated accounting voucher XML / JSON files), scanned images of paper documents (including scans or photos of contracts, invoices, and expense reimbursement documents), electronic invoices (invoice files stored in standard formats such as PDF, OFD, and XML), bank transaction records (electronic details of bank statements or standard interface data), contract texts (electronic contracts in PDF or Word format), financial statements (standard table data such as balance sheets, income statements, and cash flow statements), payment details (various collection and payment business records exported from ERP systems), and inventory records (including barcode scan data, inventory count forms, etc.). In the present invention, the above-mentioned multiple types of "raw accounting data" are based on the objective needs of financial scenario data and the compliance requirements of current accounting regulations to ensure comprehensive coverage of data at all stages and traceability of data sources.

[0022] During specific implementation, the S1 step adopts an automated collection method, utilizing the integration with multi-source information systems such as the enterprise ERP, financial sharing system, bank interface, image archiving platform, and electronic invoice management system to realize batch automatic pulling and real-time synchronization of data. The system unifies the original data of different sources and formats through a preset standardized data interface, and automatically aggregates them into a dedicated data aggregation area. For example, for paper documents, high-resolution scanning and OCR recognition modules are used to extract the document content and images and convert them into structured electronic data; for bank statements, electronic invoices, etc., API interfaces or data import scripts are called to regularly collect the latest business details and automatically archive them. This execution detail not only ensures the integrity and timeliness of the original accounting data, but also significantly reduces the risks of loss, misrecording, and delays caused by human operations.

[0023] S2: Preprocessing the original accounting data set to generate a preprocessed structured accounting data set.

[0024] The original accounting data set is subjected to format conversion, text recognition, data deduplication, missing value processing, data consistency verification, field extraction and data standardization to obtain a preprocessed structured accounting data set.

[0025] In the embodiments of the present invention, "preprocessing of raw accounting data sets" refers to systematically and automatically processing various types of collected raw accounting data in accordance with preset data quality standards and structural requirements, thereby transforming the raw data into high-quality structured data suitable for subsequent intelligent analysis, archiving, and verification. S2 specifically includes the following steps: First, format conversion addresses the heterogeneous nature of raw accounting data. Through automated format parsing and conversion scripts, data from various sources (such as PDF, OFD, Word, Excel, XML, and images) is unified into recognizable structured intermediate formats (such as JSON, CSV, and database forms), ensuring seamless access and analysis in downstream processing. Second, text recognition primarily targets accounting data in the form of scanned images or pictures. By integrating OCR (optical character recognition) technology, it automatically extracts text from images and, combined with layout analysis, identifies key information such as bill fields, amounts, and dates by partition, enhancing the digitization of raw data.

[0026] Data deduplication and missing value processing are key steps in improving data quality. To address the potential duplication of data introduced by batch collection and multi-channel aggregation, the system automatically compares each piece of data's unique identifiers (such as voucher numbers, invoice codes, contract numbers, etc.) and key information to achieve efficient deduplication. Missing items detected are filled using rules, referenced by historical data, or marked as missing to ensure the integrity and availability of structured data. Data consistency verification automatically compares multiple tables and fields to verify the consistency of different data sources for the same business on fields such as amounts, dates, and related parties, promptly identifying and alerting potential errors.

[0027] Field extraction automatically extracts key information from raw data (such as voucher type, amount, date, payee / payee, and transaction type) into a standardized field structure, enabling the attributed representation of various data types. Finally, data standardization processes standardize the extracted fields according to the company's pre-set data caliber, accounting subject code, and date format. This eliminates inconsistent calibers and format differences caused by different sources, establishing a unified data baseline for subsequent intelligent classification and archiving.

[0028] Furthermore, by integrating multiple technical paths (such as automatic format conversion, OCR recognition, intelligent deduplication verification, regularized field extraction and standardization) to form an end-to-end pre-processing process, all types of accounting data can be automatically structured and cleaned with high quality in a short period of time, greatly reducing manual intervention and processing time. On the other hand, this link provides a high-quality input foundation for subsequent intelligent processes such as AI classification, automatic archiving, and data verification, ensuring the accuracy, continuity and traceability of the entire data flow chain. In particular, for image-based, unstructured or multi-format raw accounting data, the solution of the present invention automatically converts it into structured data, breaking through the limitation of traditional systems that can only process standard electronic data, and effectively improving the system's adaptability and intelligence level to complex and dynamic financial business scenarios.

[0029] S3: Input the preprocessed structured accounting data set into a pre-trained artificial intelligence classification model, and output the classified structured data set.

[0030] The pre-trained AI classification model is a deep neural network classification model based on multimodal data feature fusion, specifically including: Inputting the text information and layout features of each piece of data in the preprocessed structured accounting data set into the embedding layer to generate corresponding text feature vectors and layout feature vectors; fusing the text feature vector and the layout feature vector into a unified multimodal feature vector; Inputting the multimodal feature vector into multiple convolutional neural network layers and attention mechanism layers in series to perform deep feature extraction and output a fused deep feature representation; The fused deep feature representation is input into the fully connected classification layer, the category probability distribution is calculated by the classification layer, the category with the largest probability is selected as the classification result of the corresponding data, and the classified structured data set is output.

[0031] In this embodiment of the present invention, the structured accounting data set pre-processed by S2 is input into a "pre-trained artificial intelligence classification model" to achieve high-precision automatic classification of accounting data. The artificial intelligence classification model uses a deep neural network structure that integrates multimodal data features. Its innovations and feasibility are mainly reflected in the following aspects: First, to address the complexity and diversity of accounting data, this model leverages textual information (such as voucher summaries, amounts, dates, and transaction details) and layout features (such as the spatial distribution of fields within the document, font size, and table structure) for joint modeling. Specifically, the textual content and layout information within each piece of structured accounting data are separately encoded and converted into low-dimensional, dense text and layout feature vectors through an embedding layer. This process not only captures the semantic relationships in the data but also reflects layout variations across different document templates, providing multi-dimensional support for intelligent classification in complex scenarios.

[0032] Secondly, the model combines text features with layout features through feature fusion mechanisms (such as feature concatenation and attention weighting) to form a unified multimodal feature vector, fully preserving the complementary nature of different information sources. These multimodal features are then fed into a multi-layer convolutional neural network (CNN) and a self-attention mechanism (such as the Transformer architecture) to achieve deep feature extraction. The CNN layer helps automatically learn the local structure and patterns of document content, while the self-attention mechanism captures global dependencies and long-range interactions, significantly improving the ability to distinguish complex accounting document types.

[0033] During the classification phase, the model inputs the fused deep features into a fully connected classification layer. Using activation functions such as softmax or sigmoid, the model calculates the class probability distribution and ultimately selects the class with the highest probability as the classification result. This process not only supports the automatic recognition of multi-category and multi-level accounting data, but also demonstrates a certain degree of generalization for newly emerged or rare categories. The entire training process utilizes supervised learning based on the company's historically annotated data, with model parameters continuously and adaptively optimized.

[0034] Compared with traditional template matching or manual rule classification, deep neural network models have higher adaptability and fault tolerance for accounting data with diverse formats and complex fields, and can significantly improve the accuracy and coverage of automatic classification; multimodal fusion improves the robustness of the model to actual business scenarios such as abnormal templates and mixed data sources, and reduces misclassification caused by the lack of a single feature; intelligent classification output provides a high-quality, standardized data foundation for subsequent automatic archiving, anomaly detection and other links, making the entire accounting data processing chain more efficient, intelligent and automated.

[0035] S4: Automatically archiving the classified structured data set according to the category identifier to obtain an archived structured data set.

[0036] Each piece of classified structured accounting data in the classified structured data set is automatically assigned and stored to the corresponding archiving group according to its corresponding category identifier. The archiving groups are divided according to the category identifier, and the data of each archiving group is aggregated to form an archived structured data set that can be indexed, retrieved and managed based on the category.

[0037] In an embodiment of the present invention, automatic archiving uses "category identification" as the primary method, enabling precise grouping and efficient management of intelligently classified structured accounting data. Specifically, for each piece of data in the classified structured dataset, the system first identifies the label based on the category identifier output by the model (such as "travel reimbursement voucher," "supplier payment slip," "contract receipt," "balance sheet details," etc.), and then automatically assigns the data to the corresponding archiving group. The archiving grouping method supports multi-level indexing, for example, with voucher type as the primary classification, and sub-grouping can be further refined according to conditions such as accounting period, business department, and amount range, achieving flexible hierarchical aggregation.

[0038] Automatic archiving utilizes a storage structure based on a rules engine or dynamic archiving strategies, dynamically adapting to the actual needs of business development. For example, for newly emerging voucher categories, the system automatically generates and archives new groups based on the output labels of the classification model, eliminating the need for manual intervention and significantly enhancing the system's adaptability. After archiving, each grouped data supports multi-dimensional attribute indexing, allowing enterprises to quickly retrieve and access archived data by multiple criteria, such as category, time, department, and amount, enabling precise data location and rapid access.

[0039] Furthermore, it has achieved the following effects: (1) The whole archiving process is automated, which significantly improves the efficiency and standardization of large-scale accounting data processing and reduces the risks of grouping errors, information omissions and delayed sorting caused by manual archiving; (2) The archiving structure is highly standardized, which not only facilitates the subsequent tracing of data, compliance inspections and internal control audits, but also provides a convenient data foundation for data mining and financial analysis; (3) The automatic grouping and indexing mechanism realizes the adaptation to multi-category and dynamically changing business scenarios, and effectively solves the technical bottlenecks of the existing accounting information system in archiving grouping flexibility and batch archiving capabilities.

[0040] In addition, the automatically archived structured data sets have a complete data security and permission management mechanism. The data in each archive group can achieve access control and operation records, ensuring the security and traceability of data during archiving and subsequent circulation, providing solid support for enterprises to achieve compliant financial management and efficient data operations.

[0041] S5: Perform data verification on the archived structured data set, identify abnormal data, and generate verification results.

[0042] The data in the archived structured data set is automatically compared with historical data based on preset verification rules. The verification rules include amount consistency verification, date legitimacy verification, account correspondence verification, voucher number uniqueness verification and cross-checking with external accounting data. If data inconsistency, missing, duplication and abnormal fluctuation are detected, they will be marked as abnormal data and a verification result containing abnormal type, abnormal location and abnormal description information will be generated.

[0043] In an embodiment of the present invention, archived structured data sets undergo systematic, automated data validation to efficiently identify abnormal data and output accurate validation results. First, the system automatically applies a series of pre-set validation rules to perform multi-dimensional checks on each piece of archived accounting data. These validation rules include, but are not limited to: amount consistency verification (e.g., consistency between voucher amounts and report and sub-account amounts); date legitimacy verification (e.g., matching date format, accounting cycle, and financial period); account correspondence verification (e.g., matching transaction type with accounting subject); voucher number uniqueness verification (to prevent duplicate entries or omissions); and cross-checking with external accounting data (e.g., comparing bank statements, tax platform data, and account book records), ensuring data quality from multiple perspectives.

[0044] During implementation, the system establishes verification thresholds based on historical business data statistics and intelligently compares newly received data. To address anomalies potentially caused by seasonality or business fluctuations, the system uses a dynamic historical baseline approach to determine fluctuation ranges for indicators such as amounts and quantities, automatically identifying significant deviations from historical data. The system generates detailed verification logs for all verification processes, including verification pass / fail status, anomaly type (e.g., amount discrepancy, missing field, duplicate number, time anomaly), anomaly location (specific data entry, field, and archiving group), and a detailed description of the anomaly.

[0045] Furthermore, its effects are mainly reflected in the following aspects: (1) the verification process is fully automated and batched, which can cover the entire amount of data in real time, significantly superior to the efficiency and breadth of traditional sampling review or single-rule static verification; (2) the intelligent judgment method of multi-rule fusion and dynamic comparison with historical data effectively reduces the "blind spots" of manual rules and improves the comprehensiveness and sensitivity of anomaly identification; (3) through automatic interface comparison with external systems, cross-system and cross-business account verification is achieved, greatly improving the credibility and compliance of financial data.

[0046] Furthermore, it can not only detect and locate anomalies in the first place, reducing financial risks and decision-making errors caused by data errors, but also provide accurate input and traceability basis for subsequent intelligent anomaly analysis, automatic processing and closed-loop control, laying a solid foundation for enterprises to build a comprehensive, rigorous and efficient accounting data quality control system.

[0047] S6: For the abnormal data in the verification result, a preset intelligent analysis model is used to analyze the cause of the abnormality, and corresponding processing suggestions are automatically generated according to the analysis results.

[0048] For data marked as abnormal in the verification results, its associated accounting subjects, voucher content, business type and historical data features are extracted, and the extracted information is input into the intelligent analysis model built based on the decision tree and anomaly detection algorithm. The causes of the abnormal data are automatically classified and the abnormal cause labels are output; based on the abnormal cause labels and the knowledge rule base built into the model, processing suggestions for different abnormal types are automatically matched and generated.

[0049] In this embodiment of the present invention, for data identified as abnormal in step S5, the system further utilizes an intelligent analysis model to automatically attribute the abnormality and generate treatment suggestions. This intelligent analysis model integrates a decision tree and anomaly detection algorithm, specifically including the following key processes: First, the system automatically extracts the associated attributes of each piece of abnormal data, including feature vectors such as the accounting subject, voucher content, transaction type, amount, date, business department involved, and historical frequency of occurrence. This information not only fully reflects the business context of the abnormal data but also provides a rich foundation for highly accurate identification of the cause of the anomaly.

[0050] These features are then fed into a decision tree model trained on historical anomaly attribution cases. The decision tree model automatically learns attribution rules based on the different paths of anomaly types. For example, by assessing the context of the amount discrepancy, the matching relationship between the transaction type and account, and the timing of the anomaly, it can quickly locate the likely cause of the anomaly. For example, an anomaly may be caused by duplicate data entry, cross-period vouchers, account mapping errors, document format mismatches, or delayed system integration. The model's self-learning capabilities allow it to continuously incorporate new anomaly cases, improving the generalization and practicality of the model's attribution.

[0051] To address the limitations of decision trees in detecting new anomalies and those with weak features, the system integrates unsupervised anomaly detection algorithms such as Isolation Forest and LOF, enabling complementary identification of hidden and complex anomalies in historical data that are difficult to detect using rules. By calculating the degree of deviation between anomalies and the normal data distribution, the model achieves highly sensitive identification and dynamic early warning in label-free scenarios.

[0052] After obtaining the anomaly cause label, the system automatically calls upon its built-in knowledge rule base (which can be constructed from industry standards, corporate management systems, and expert experience) to generate optimal handling recommendations based on the specific anomaly type. Recommendations include, but are not limited to, data correction, supplemental entry, manual review, freezing related accounts, and risk warnings. Specific measures automatically adapt to the anomaly type and business scenario, ensuring an efficient, closed-loop exception handling process.

[0053] On the one hand, through the integration of decision trees and anomaly detection algorithms, the automation and intelligence level of anomaly attribution has been significantly improved, and high-accuracy anomaly tracing and classification has been achieved, providing strong technical support for financial management risk prevention and control; on the other hand, the automatic suggestion generation mechanism has greatly reduced the subjectivity and response lag of manual judgment, and effectively improved the standardization, batching and timeliness of accounting anomaly processing.

[0054] S7: Process the abnormal data according to the processing suggestions, and record the processing suggestions and execution results.

[0055] According to the processing suggestions generated by the intelligent analysis model, the corresponding processing operations are automatically performed on the abnormal data. After the processing is completed, the adopted processing suggestions, processing time, processing person and system identification and processing result information are recorded in a structured manner to form a complete processing log that can be used for subsequent traceability, auditing and statistical analysis.

[0056] In an embodiment of the present invention, the system forms a complete closed-loop management and traceability chain around the final disposal of abnormal data and the recording of the entire process. Specifically, the system automatically selects the corresponding exception handling action based on the processing suggestions output by the intelligent analysis model in step S6. For example, in response to data entry errors, the system can automatically correct the data or initiate a re-entry process; for complex exceptions that require manual confirmation, it automatically generates a review task and assigns it to the designated responsible person, and tracks the progress of the processing; for systemic or high-risk exceptions, the system can automatically freeze the relevant accounts or trigger risk warnings, and push processing notifications to management.

[0057] The entire processing process is rigorously documented. The system records the recommended treatment, actual action taken, processing time, responsible individual (or automated processing identifier), data changes before and after processing, and associated exception numbers in a structured data format, all in real time in a dedicated log module. This log not only facilitates internal audits and external regulatory bodies for retrospective review but also provides a rich historical data foundation for subsequent financial data statistical analysis, exception handling strategy optimization, and model iteration.

[0058] It should also be noted that exception handling has achieved an organic combination of automation and artificial intelligence assistance, significantly reducing the risks of subjective misjudgment, operational delays and information omissions brought about by traditional manual intervention, and improving the efficiency and accuracy of accounting exception handling; the structured processing log design realizes the full life cycle management of each abnormal data, including full records of all links such as suggestion generation, task assignment, execution and result feedback, greatly improving the compliance, transparency and traceability of data management; the system has the ability of self-evolution and strategy closure, and through the long-term accumulation and analysis of processing results, it continuously optimizes exception handling rules and model parameters to realize intelligent exception handling and continuous evolution of corporate financial management.

[0059] Example 2 is the second embodiment of the present invention, which is different from the previous embodiment in that: If the functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art or the portion of the current technical solution, can be embodied in the form of a software product. The current computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes various media that can store program code, such as USB flash drives, mobile hard drives, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical disks.

[0060] The logic and / or steps represented in the flowcharts or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing the logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (e.g., a computer-based system, a system including a processor, or other system that can fetch and execute instructions from an instruction execution system, apparatus, or device). For purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport a program for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0061] More specific examples (a non-exhaustive list) of computer-readable media include the following: an electrical connection with one or more wires (electronic devices), a portable computer disk cartridge (magnetic devices), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disc read-only memory (CDROM). In addition, the computer-readable medium may even be paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, and then editing, interpreting, or processing in another suitable manner as necessary, and then storing it in a computer memory.

[0062] Example 3 is an embodiment of the present invention, which provides an accounting data processing system, including a raw data acquisition module, a data preprocessing module, an intelligent classification module, an automatic archiving module, a data verification module, an intelligent anomaly analysis module and an anomaly processing and recording module.

[0063] Original data collection module: collects original accounting data within the enterprise and generates original accounting data sets; Data preprocessing module: preprocessing the original accounting data set to generate a preprocessed structured accounting data set; Intelligent classification module: inputs the pre-processed structured accounting data set into a pre-trained artificial intelligence classification model and outputs a classified structured data set; Automatic archiving module: automatically archiving the classified structured data set according to the category identifier to obtain an archived structured data set; Data verification module: performs data verification on the archived structured data set, identifies abnormal data, and generates verification results; Intelligent anomaly analysis module: For abnormal data in the verification results, a preset intelligent analysis model is used to analyze the cause of the anomaly and automatically generate corresponding treatment suggestions based on the analysis results; Exception processing and recording module: processes the exception data according to the processing suggestions, and records the processing suggestions and execution results.

[0064] Example 4 is an embodiment of the present invention, which provides an accounting data processing method. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through economic benefit calculation and simulation / comparative experiments.

[0065] This example uses typical accounting data from a manufacturing enterprise in May 2024 for experimentation. First, through integration with the enterprise's ERP system, bank transaction interface, electronic invoice collection platform, and contract image archiving system, it automatically collects raw accounting data from 10 business categories, including purchase payments, sales receipts, daily reimbursements, and inventory counts. This raw data includes electronic vouchers, scanned contract images, bank transaction details, financial statements, and electronic invoices, truly reflecting the enterprise's daily accounting scenarios.

[0066] After collection, the raw data undergoes format conversion, OCR text recognition, deduplication, and missing value correction to generate a standardized, pre-processed, structured accounting dataset. A deep neural network model, based on the fusion of text semantics and layout features, automatically and intelligently categorizes this structured data, outputting multiple labels for categories such as vouchers, reimbursements, inventory checks, and invoices. The system then archives the structured data into pre-set groups based on these categorized labels, enabling automated partitioned storage for business operations such as vouchers, reimbursements, and inventory checks.

[0067] Based on the archived data, the system automatically conducts full verification of dimensions such as amount, date, account, and unique number. The preset verification rules cover amount consistency, number uniqueness, date range, business account matching, and cross-verification with bank statements, and found that some data had anomalies in terms of amount, number, or ownership. For abnormal data, it is further input into the intelligent analysis model that integrates the decision tree and isolation forest algorithm to identify the causes of anomalies such as amount entry errors, duplicate reimbursements, contract number matching anomalies, etc., and automatically generates processing suggestions such as re-entry, review, and data correction. The system automatically corrects some data according to the suggestions, and issues review tasks for anomalies that require manual intervention. All processing processes and results are recorded in a structured manner to provide a detailed basis for subsequent audits and tracing. For specific reference data, please refer to Table 1.

[0068] Table 1 Data reference table

[0069] The experimental data table covers many types of typical accounting business such as procurement, sales, reimbursement, inventory and contracts, reflecting the adaptability of the present invention in actual complex business scenarios. First, through automated collection and structured preprocessing, all 10 original business data in the table have achieved information standardization and comprehensive collection, solving the problems of traditional manual entry of information that is prone to omissions, scattered data sources, and poor normalization. Based on multimodal features, the classification model automatically and accurately classifies business data into archiving categories such as vouchers, reimbursements, inventory, and contracts, realizing accurate and automatic archiving of different types of business, significantly improving data management efficiency and the convenience of subsequent retrieval.

[0070] From the perspective of verification and anomaly detection, A04 and A08 in the table are typical examples of duplicate number anomalies, while A05 and A10 reflect amount errors. Compared to traditional sampling verification, which is prone to missing details or requires high manpower input, the full verification mechanism of this invention can identify all anomalies at once, achieving high-density and high-accuracy data risk control. Furthermore, after integrating an intelligent analysis model, the system not only clearly determines the type of anomaly (such as an abnormal amount or duplicate number), but also automatically generates reasonable handling suggestions based on historical cases and business rules. For example, the A05 anomaly was automatically corrected through intelligent suggestions, significantly reducing human intervention time and the probability of misjudgment. Data such as A04, A08, and A10, which require manual judgment, has been automatically dispatched to the review stage by the system, and structured records of the entire processing process are recorded to meet subsequent compliance audit and traceability requirements.

[0071] In general, experiments have shown that the present invention can achieve end-to-end automated management in key links such as data collection, archiving, verification, anomaly analysis and processing. Compared with the existing technology, it avoids the inefficiency and high error rate of manual data entry, the data lag of batch archiving, the limitations of sampling review, and the arbitrariness and non-traceability of anomaly handling. Especially in the closed-loop link of data archiving, anomaly detection and suggestion execution, the present invention, through the collaboration of structured data and intelligent decision-making models, gives the accounting information system the ability to be highly automated, intelligent, traceable and self-evolving. As a result, the compliance, security and business response speed of corporate financial data governance are improved, fully reflecting the innovation and significant technological progress of the present invention in the field of intelligent accounting data processing.

[0072] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for processing accounting data, characterized in that: include: Collect original accounting data within the enterprise and generate original accounting data sets; Preprocessing the original accounting data set to generate a preprocessed structured accounting data set; Inputting the preprocessed structured accounting data set into a pre-trained artificial intelligence classification model and outputting a classified structured data set; Automatically archiving the classified structured data set according to the category identifier to obtain an archived structured data set; Perform data verification on the archived structured data set, identify abnormal data, and generate verification results; For abnormal data in the verification results, a preset intelligent analysis model is used to analyze the cause of the abnormality, and corresponding processing suggestions are automatically generated based on the analysis results; Process the abnormal data according to the processing suggestions, and record the processing suggestions and execution results; Generating corresponding processing suggestions includes extracting the associated accounting subjects, voucher content, business type, and historical data features for data marked as abnormal in the verification results, inputting the extracted information into an intelligent analysis model built based on a decision tree and anomaly detection algorithm, automatically classifying the causes of the abnormal data, and outputting an abnormality cause label; Automatically match and generate processing suggestions for different exception types based on the exception cause labels and the model's built-in knowledge rule base; The recording of processing suggestions and execution results includes automatically performing corresponding processing operations on abnormal data according to the processing suggestions generated by the intelligent analysis model, and recording the adopted processing suggestions, processing time, processing person and system identification and processing result information in a structured manner after the processing is completed, forming a complete processing log that can be used for subsequent traceability, auditing and statistical analysis.

2. The accounting data processing method according to claim 1, wherein: The original accounting data includes electronic vouchers, scanned images of paper documents, electronic invoices, bank transaction records, contract texts, financial statements, payment details, and inventory records; The original accounting data are aggregated to form an original accounting data set.

3. The accounting data processing method according to claim 2, wherein: The preprocessing of the original accounting data set includes format conversion, text recognition, data deduplication, missing value processing, data consistency verification, field extraction and data standardization to obtain a preprocessed structured accounting data set.

4. The accounting data processing method according to claim 3, wherein: The pre-trained artificial intelligence classification model is a deep neural network classification model based on multimodal data feature fusion, specifically including: Inputting the text information and layout features of each piece of data in the preprocessed structured accounting data set into the embedding layer to generate corresponding text feature vectors and layout feature vectors; fusing the text feature vector and the layout feature vector into a unified multimodal feature vector; Inputting the multimodal feature vector into multiple convolutional neural network layers and attention mechanism layers in series to perform deep feature extraction and output a fused deep feature representation; The fused deep feature representation is input into the fully connected classification layer, the category probability distribution is calculated by the classification layer, the category with the largest probability is selected as the classification result of the corresponding data, and the classified structured data set is output.

5. The accounting data processing method according to claim 4, wherein: The archived structured data set includes automatically allocating and storing each classified structured accounting data in the classified structured data set to a corresponding archiving group according to its corresponding category identifier, dividing the archiving groups according to the category identifier, and aggregating the data of each archiving group to form an archived structured data set that can be indexed, retrieved and managed based on the category.

6. The accounting data processing method according to claim 5, wherein: The data verification of the archived structured data set includes automatically comparing each data of the archived structured data set with historical data based on preset verification rules. The verification rules include amount consistency verification, date legitimacy verification, account correspondence verification, voucher number uniqueness verification and cross-checking with external accounting data. If data inconsistency, missing, duplication and abnormal fluctuation are detected, they are marked as abnormal data, and a verification result including abnormal type, abnormal location and abnormal description information is generated.

7. An accounting data processing system for implementing the accounting data processing method according to any one of claims 1 to 6, characterized in that: include: Original data collection module: collects original accounting data within the enterprise and generates original accounting data sets; Data preprocessing module: preprocessing the original accounting data set to generate a preprocessed structured accounting data set; Intelligent classification module: inputs the pre-processed structured accounting data set into a pre-trained artificial intelligence classification model and outputs a classified structured data set; Automatic archiving module: automatically archiving the classified structured data set according to the category identifier to obtain an archived structured data set; Data verification module: performs data verification on the archived structured data set, identifies abnormal data, and generates verification results; Intelligent anomaly analysis module: For abnormal data in the verification results, a preset intelligent analysis model is used to analyze the cause of the anomaly and automatically generate corresponding treatment suggestions based on the analysis results; Exception processing and recording module: processes the exception data according to the processing suggestions, and records the processing suggestions and execution results.

Images