A method for automatic process processing of accounting data

Abstract

The present invention provides a method and system for automatic process processing of accounting data. Data is collected by connecting to the internal business system of an enterprise through an adapter interface, and the data is cleaned by combining OCR technology and a professional vocabulary library. An improved random forest algorithm is used to construct a classification model based on attribute weights according to accounting professional knowledge, so as to achieve accurate classification and standardization of data. According to preset financial process rules, reimbursement, procurement payment and other processes are automatically processed and monitored in real time, and timely warnings are issued in case of abnormalities. The processed data is encrypted and stored, and a backup and authority management mechanism is established. Financial statements can also be automatically generated and displayed visually. This invention solves the problems of low efficiency and easy errors in traditional accounting data processing, improves data processing efficiency and accuracy, ensures data security, and provides efficient support for corporate financial management.

Description

Technical Field

[0001] The present invention relates to data processing related fields, and in particular to an automatic process processing method for accounting data. Background Art

[0002] In traditional accounting data processing, large amounts of accounting data, such as daily expense reports, invoice information, and bank statement data, require manual entry into financial systems by accountants. For example, in the case of expense report processing, accountants must individually verify information such as employee name, department, expense item, amount, and invoice number on the report and accurately enter this information into the financial software system. This process is not only time-consuming and labor-intensive, but also prone to data entry errors due to the inevitable oversight inherent in manual operations. Data entry errors can impact subsequent tasks such as financial report generation, cost accounting, and tax filing, potentially leading to inaccurate financial data and posing potential risks to business decision-making and operations. Furthermore, the inefficiency of manual data entry makes it difficult for accounting departments to quickly process large amounts of data, impacting the overall timeliness of financial work. Therefore, there is an urgent need for a streamlined approach that can automatically, accurately, and efficiently process accounting data. Summary of the Invention

[0003] The purpose of the present invention is to provide a method for automatic process processing of accounting data to solve the problems raised in the above background technology.

[0004] To achieve the above-mentioned purpose, the present invention provides the following technical solution: a method for automatically processing accounting data, comprising the following steps:

[0005] Step 1: Data collection: Use the adapter interface to connect to the company's OA, procurement, and sales internal business systems. Daily, electronic expense reimbursement forms submitted by employees are captured from the OA system. Purchase orders and corresponding invoice data are obtained from the procurement system. Then, optical character recognition technology is used to scan paper invoices and bank statements. After scanning, incorrect characters recognized by OCR are automatically verified and corrected based on the accounting professional vocabulary and grammatical rules. At the same time, the data format is unified, duplicate data is removed, and missing values ​​are filled. This completes data cleaning and obtains the collected data.

[0006] Step 2: Data classification and standardization: Using an improved machine learning algorithm, build a classification model based on the company's accounting system. Input the cleaned data collected in step 1 into the classification model and accurately classify it into the corresponding accounting categories.

[0007] Step 3: Automated Processing: The reimbursement data, purchase data, and sales data categorized in Step 2 are processed according to the company's financial approval and accounting processes. The processing process is based on pre-set process rules in the system.

[0008] Step 4: Data Storage and Analysis: The accounting data processed in Step 3 is encrypted and stored in the enterprise database. A regular backup mechanism is established to prevent data loss. Access permissions are set for different data based on personnel responsibilities to ensure data security. At the same time, data analysis tools and algorithms are used to automatically generate financial statement data, including but not limited to balance sheets, income statements, and cash flow statements, based on the stored data. Financial data is presented in intuitive charts using data visualization technology.

[0009] In step 2, an improved machine learning algorithm is used. Specifically, an improved random forest algorithm is used that assigns different weights to different attributes based on accounting expertise. A classification model is constructed through the improved random forest algorithm, and the collected data is automatically and accurately classified through the classification model.

[0010] Preferably, the improved random forest algorithm is specifically implemented as follows:

[0011] Data preparation stage: From the historical accounting data of the enterprise, a sample data set of accounting subjects is constructed by using stratified sampling combined with dynamic proportion adjustment. Suppose the historical accounting data set is , data set Contains various accounting data samples in different proportions, including but not limited to expense reimbursement data , procurement data , sales data The corresponding accounting subject samples; in stratified sampling, first Divided by business type non-overlapping layers, i.e. , for each layer , the data volume is recorded as and the importance is recorded as , importance It is determined by accounting experts based on a comprehensive assessment of the impact of different business type data on accounting subject classification, with a value range of ,and , dynamically adjust the extraction ratio according to the amount and importance of each layer of data After calculating the extraction ratio of each layer according to the above formula, randomly extract data from each layer according to the corresponding ratio to construct a diverse training subset ;

[0012] Decision tree training phase: for each constructed training subset , respectively carry out the decision tree training work; in the process of node splitting of the decision tree, based on the hierarchical analysis method combined with the experience of accounting experts, automatically assign weights to the decision tree node splitting attributes, and according to the calculated weight distribution method, for each training subset Perform recursive partitioning and construct multiple decision trees;

[0013] Random forest integration stage: the above-trained multiple decision trees are combined into a random forest classification model, i.e., a classification model;

[0014] Data classification execution phase: The collected data cleaned in step 1 is input into the constructed classification model. Each decision tree in the classification model independently classifies the input data. Each decision tree analyzes the various features of the input data based on its own training rules and attribute weights, and then generates a classification result. After all decision trees have completed the classification, a voting mechanism is used to determine the final classification result of the input data.

[0015] Standardized operation stage: After completing data classification, the classified data is standardized in strict accordance with accounting standards and the financial system formulated within the enterprise.

[0016] Preferably, in the process of splitting nodes of the decision tree, a method based on the hierarchical analysis method combined with the experience of accounting experts is used to automatically assign weights to the splitting attributes of the decision tree nodes. The specific implementation logic is as follows:

[0017] Step A: Determine the attribute set and decision target: First, train the subset For expense reimbursement data, the attributes include "reimbursement item," "reimbursement amount," "reimbursement time," and "reimbursement department." For procurement data, the attributes include "purchased item category," "purchase quantity," "supplier," and "purchase date." For sales data, the attributes include "sales item," "sales amount," "sales time," and "sales department." Accurately classifying the above accounting attribute data into corresponding accounting subjects is the decision-making goal.

[0018] Step B, construct judgment matrix: Combine the experience of accounting experts and the sample attributes of accounting subjects for different business types to construct a judgment matrix, set the training subset of The attribute set is , for each attribute data in the accounting subject sample, the judgment matrix is ​​constructed: ,in Representation attributes Relative to attributes The importance of , When constructing the judgment matrix, the 1-9 scale method is used to compare the relative importance of each attribute in each accounting subject sample. Specifically, if the attribute Relative to attributes Equally important, ;like Compare Slightly more important, ; obviously important, ; Strongly important, ; Extremely important, If it is between the two, The values ​​are 2, 4, 6, and 8 respectively;

[0019] Step C, calculate the attribute weight vector: use the eigenvector method to calculate the maximum eigenvalue of the judgment matrix and its corresponding eigenvector , by solving the equation , and obtain the feature vector ;

[0020] Step D, consistency test: calculate consistency index , the formula is ,in is the number of attributes; then find the average random consistency index , which is based on the number of attributes Obtained from standard tables; calculated agreement ratio , the formula is ,when When , the consistency of the received judgment matrix is ​​good, that is, the expert's judgment is reasonable, and the obtained weight vector is valid; if , then readjust the judgment matrix until the consistency test passes;

[0021] Step E: Integrate expert weight results: For the training subset For each attribute, the weights given by the experts are combined and the simple average method is used. The number of experts is set to , No. Attributes given by experts The weight of , then the integrated attributes The final weight The calculation formula is: ;

[0022] Step F: Apply weights to decision tree node splitting: Apply the final attribute weights obtained to the decision tree node splitting process.

[0023] Preferably, the maximum eigenvalue of the judgment matrix is ​​calculated using the eigenvector method and its corresponding eigenvector The specific steps are as follows:

[0024] Normalize the judgment matrix by column: Assume that the constructed judgment matrix is , for each column of the matrix , calculate the sum of the elements in this column , then each element in the matrix Divide by the sum of the elements in the column , get the normalized matrix elements Through the above calculation, the column normalization of the entire judgment matrix is ​​completed, and the normalized matrix is ​​obtained ;

[0025] Sum the normalized matrix row by row: For the normalized matrix , calculate the sum of the elements in each row ;

[0026] Normalize the rows and vectors to get the eigenvectors : Rows and vectors Normalize, that is, calculate , and obtain the feature vector , the eigenvector Each component in The relative weight of each attribute in the corresponding accounting subject sample;

[0027] Calculate the maximum eigenvalue :calculate , and get the vector , then calculate ,in is a vector No. elements, the largest eigenvalue Used for subsequent consistency testing to determine the rationality of expert judgment.

[0028] Preferably, the specific contents of the automated process processing in step 3 are as follows:

[0029] Preset process rules: For reimbursement and sales processes, the company presets the approvers and approval order for different amount ranges, reimbursement and sales items. For the purchase and payment process, the company determines the approval levels based on factors such as purchase order amount and supplier type. These pre-set rules are then compiled into code logic to serve as the basis for subsequent automated process execution.

[0030] Reimbursement and sales process processing: according to the classification results of the collected data in step 2 of the invention, the expense reimbursement data and sales data are automatically screened out; according to the preset rules, the corresponding approver of each reimbursement and sales order is determined, and the reimbursement order is sent to the approver's work interface through an internal message push mechanism or integration with the enterprise office software, along with detailed reimbursement and sales information, including but not limited to the name of the reimbursement and sales person, department, reimbursement item, amount and bill; after receiving the reimbursement and sales order, the approver performs the approval operation and chooses to approve, reject or return for modification; if the approver approves, the data is automatically transferred to the financial accounting link; if rejected or returned for modification, the reimbursement or sales person is notified in a timely manner and the reason is explained; in the financial accounting link, based on the reimbursement and sales order data, in accordance with accounting standards and corporate financial systems, accounting vouchers are automatically generated to complete the accounting processing;

[0031] Procurement payment process processing: According to the classification results of the collected data in step 2 of the invention, the procurement data is automatically screened out, and the purchase order data and the associated invoice data in the procurement data are monitored in real time; when the procurement data information matches and meets the preset payment conditions, a payment application is automatically generated, including but not limited to the purchase order number, supplier information, payment amount and payment method; after the payment application is generated, it is pushed to the corresponding approver in sequence according to the preset approval hierarchy rules, and the approver performs the approval operation. After the approval is passed, if all levels of approval are completed, an instruction is automatically sent to the company's capital payment system to trigger the payment operation and complete the payment;

[0032] Data flow and processing progress monitoring: Real-time monitoring of the input data flow path and processing progress in the reimbursement and sales processes, as well as the purchase and payment processes. This is achieved by setting markers and timestamps at each process node to record the time when data enters and leaves each link, as well as its current location.

[0033] Exception handling and reminders: Continuously check whether there are any exceptions in the data flow and processing process; if the approval timeout occurs, immediately activate the reminder mechanism and send a reminder email or text message to the approver, reimbursement person or relevant business manager to urge them to complete the approval operation; if it is found that the data does not comply with the preset rules, send a notification to inform the details of the exception and suspend the relevant process until the problem is resolved.

[0034] Preferably, an automatic process processing system for accounting data includes a data acquisition module, a data classification and standardization module, an automated process processing module, and a data storage and analysis module; the data acquisition module includes an adapter interface development component and an OCR scanning and data cleaning component, which can realize data connection with the internal business system of the enterprise and data collection and cleaning of paper bills; the data classification and standardization module integrates a machine learning algorithm module and a data standardization processing component, which can complete accurate classification and standardization operations of data; the automated process processing module has a process rule preset component and a monitoring and early warning component, which can automatically process data flow according to preset rules and provide timely early warning when anomalies occur; the data storage and analysis module includes a data encryption storage component, a report generation component and a data visualization component, which can realize secure storage of data, report generation and visual display.

[0035] Preferably, an electronic device includes: a processor and a memory, wherein the memory stores a computer program that can be called by the processor:

[0036] The processor executes an automatic process processing system for accounting data by calling the computer program stored in the memory.

[0037] Preferably, a computer program product stored on a computer-readable medium includes a computer-readable program, which, when executed on an electronic device, provides a user input interface to implement an automatic process processing system for accounting data.

[0038] Compared with existing technologies, this invention has the following advantages: It improves data processing efficiency: it automatically and regularly captures data from internal business systems such as the enterprise's OA, procurement, and sales systems through an adapter interface, scans paper receipts using optical character recognition technology, and automatically cleans the data, replacing traditional manual data entry. This significantly reduces the time and labor costs of data collection.

[0039] Improved data processing accuracy: We use improved machine learning algorithms, particularly the improved random forest algorithm, to build a classification model. By combining the experience of accounting experts to assign weights to different attributes, we can accurately categorize collected data into corresponding accounting categories. This significantly reduces the classification error rate compared to traditional manual classification or simple machine learning methods.

[0040] Enhanced data security and confidentiality: Accounting data is encrypted and stored using industry-proven, highly secure encryption algorithms, such as the AES encryption algorithm, to effectively prevent data theft or tampering during storage. A regular backup mechanism is also in place to prevent data loss due to unexpected events such as hardware failure and virus attacks, ensuring data durability and availability. Access rights are assigned to different data areas based on individual responsibilities, ensuring that only authorized personnel have access, further ensuring data security.

[0041] Optimize enterprise financial management decision support: Utilize data analysis tools and algorithms to automatically generate various financial statements such as balance sheets, income statements, and cash flow statements based on stored data, and present financial data in intuitive charts through data visualization technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 Schematic diagram of the method flow of the present invention;

[0043] Figure 2 This is a schematic diagram of the implementation process of the improved random forest algorithm of the present invention;

[0044] Figure 3 Schematic diagram of the system structure of the present invention;

[0045] Figure 4 The figure is a structural diagram of an electronic device. DETAILED DESCRIPTION

[0046] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0047] Example 1

[0048] See also Figure 1-2 The present invention provides a technical solution: a method for automatically processing accounting data, comprising the following steps:

[0049] Step 1: Data collection: Use the adapter interface to connect to the company's OA, procurement, and sales internal business systems. Daily, electronic expense reimbursement forms submitted by employees are captured from the OA system. Purchase orders and corresponding invoice data are obtained from the procurement system. Then, optical character recognition technology is used to scan paper invoices and bank statements. After scanning, incorrect characters recognized by OCR are automatically verified and corrected based on the accounting professional vocabulary and grammatical rules. At the same time, the data format is unified, duplicate data is removed, and missing values ​​are filled. This completes data cleaning and obtains the collected data.

[0050] Step 2: Data classification and standardization: Using an improved machine learning algorithm, build a classification model based on the company's accounting system. Input the cleaned data collected in step 1 into the classification model and accurately classify it into the corresponding accounting categories.

[0051] Step 3: Automated Processing: The reimbursement data, purchase data, and sales data categorized in Step 2 are processed according to the company's financial approval and accounting processes. The processing flow is based on pre-defined rules in the system. The specific processing flow is as follows:

[0052] Preset process rules: For reimbursement and sales processes, the company presets the approvers and approval order for different amount ranges, reimbursement and sales items. For the purchase and payment process, the company determines the approval levels based on factors such as purchase order amount and supplier type. These pre-set rules are then compiled into code logic to serve as the basis for subsequent automated process execution.

[0053] Reimbursement and sales process processing: according to the classification results of the collected data in step 2 of the invention, the expense reimbursement data and sales data are automatically screened out; according to the preset rules, the corresponding approver of each reimbursement and sales order is determined, and the reimbursement order is sent to the approver's work interface through an internal message push mechanism or integration with the enterprise office software, along with detailed reimbursement and sales information, including but not limited to the name of the reimbursement and sales person, department, reimbursement item, amount and bill; after receiving the reimbursement and sales order, the approver performs the approval operation and chooses to approve, reject or return for modification; if the approver approves, the data is automatically transferred to the financial accounting link; if rejected or returned for modification, the reimbursement or sales person is notified in a timely manner and the reason is explained; in the financial accounting link, based on the reimbursement and sales order data, in accordance with accounting standards and corporate financial systems, accounting vouchers are automatically generated to complete the accounting processing;

[0054] Procurement payment process processing: According to the classification results of the collected data in step 2 of the invention, the procurement data is automatically screened out, and the purchase order data and the associated invoice data in the procurement data are monitored in real time; when the procurement data information matches and meets the preset payment conditions, a payment application is automatically generated, including but not limited to the purchase order number, supplier information, payment amount and payment method; after the payment application is generated, it is pushed to the corresponding approver in sequence according to the preset approval hierarchy rules, and the approver performs the approval operation. After the approval is passed, if all levels of approval are completed, an instruction is automatically sent to the company's capital payment system to trigger the payment operation and complete the payment;

[0055] Data flow and processing progress monitoring: Real-time monitoring of the input data flow path and processing progress in the reimbursement and sales processes, as well as the purchase and payment processes. This is achieved by setting markers and timestamps at each process node to record the time when data enters and leaves each link, as well as its current location.

[0056] Exception handling and reminders: Continuously check whether there are any exceptions in the data flow and processing process; if the approval timeout occurs, immediately activate the reminder mechanism and send a reminder email or text message to the approver, reimbursement person or relevant business manager to urge them to complete the approval operation; if it is found that the data does not comply with the preset rules, send a notification to inform the details of the exception and suspend the relevant process until the problem is resolved.

[0057] Step 4: Data Storage and Analysis: The accounting data processed in Step 3 is encrypted and stored in the enterprise database. A regular backup mechanism is established to prevent data loss. Access permissions are set for different data based on personnel responsibilities to ensure data security. At the same time, data analysis tools and algorithms are used to automatically generate financial statement data, including but not limited to balance sheets, income statements, and cash flow statements, based on the stored data. Financial data is presented in intuitive charts using data visualization technology.

[0058] The RSA asymmetric encryption algorithm is used to encrypt and store processed accounting data in the enterprise database. This encryption algorithm is a mature technology and is briefly described here. The RSA asymmetric encryption algorithm uses a pair of keys: a public key and a private key. The public key is used to encrypt data, and the private key is used to decrypt it. In enterprise accounting data storage scenarios, the public key can be distributed to relevant personnel who need to upload or access data, such as financial personnel and auditors. When these personnel upload accounting data to the database, the public key is used to encrypt the data. During transmission and storage, even if the encrypted data is obtained by a third party, it cannot be decrypted without the corresponding private key. The enterprise data management department, which holds the private key, can decrypt the data when needed, ensuring data security and confidentiality.

[0059] In step 2, an improved machine learning algorithm is used, specifically an improved random forest algorithm that assigns different weights to different attributes based on accounting expertise. A classification model is constructed using the improved random forest algorithm, and the collected data is automatically and accurately classified using the classification model.

[0060] The specific implementation steps of the improved random forest algorithm are as follows:

[0061] Data preparation stage: From the historical accounting data of the enterprise, a sample data set of accounting subjects is constructed by using stratified sampling combined with dynamic proportion adjustment. Suppose the historical accounting data set is , data set Contains various accounting data samples in different proportions, including but not limited to expense reimbursement data , procurement data , sales data The corresponding accounting subject samples; in stratified sampling, first Divided by business type non-overlapping layers, i.e. , for each layer , the data volume is recorded as and the importance is recorded as , importance It is determined by accounting experts based on a comprehensive assessment of the impact of different business type data on accounting subject classification, with a value range of ,and , dynamically adjust the extraction ratio according to the amount and importance of each layer of data After calculating the extraction ratio of each layer according to the above formula, randomly extract data from each layer according to the corresponding ratio to construct a diverse training subset ;

[0062] Based on the above description, here is an example: For example, assuming , ; , ; , ,but , , , each subset contains rich and proportionate samples of various types of business data.

[0063] Advantages of the data preparation stage: Compared with traditional algorithms, the training subset constructed in this way It can more comprehensively cover the diversity of accounting data, enable the model to learn the complex relationship between data features and accounting subjects in different business scenarios, improve the model's ability to classify complex business data, and reduce classification bias caused by uneven data distribution.

[0064] Decision tree training phase: for each constructed training subset , respectively carry out the decision tree training work; in the process of node splitting of the decision tree, based on the hierarchical analysis method combined with the experience of accounting experts, automatically assign weights to the decision tree node splitting attributes, and according to the calculated weight distribution method, for each training subset Perform recursive partitioning and construct multiple decision trees;

[0065] This approach integrates accounting business logic into the decision tree construction process. Compared with traditional algorithms, decision trees can more accurately grasp key attributes during classification and improve classification accuracy. This is especially true when processing accounting data, which has clear business rules. It can effectively avoid misclassification caused by relying solely on general indicators and ignoring the essence of the business.

[0066] Random forest integration stage: the above-trained multiple decision trees are combined into a random forest classification model, i.e., a classification model;

[0067] Compared with traditional algorithms, the improved random forest can better capture the potential connections between accounting data. When faced with data classification involving multiple business links, it can comprehensively consider multiple factors, further improve classification accuracy, and adapt to the company's complex financial data processing needs.

[0068] Data classification execution phase: The collected data cleaned in step 1 is input into the constructed classification model. Each decision tree in the classification model independently classifies the input data. Each decision tree analyzes the various features of the input data based on its own training rules and attribute weights, and then generates a classification result. After all decision trees have completed the classification, a voting mechanism is used to determine the final classification result of the input data.

[0069] Here is an example of how to use a voting mechanism to determine the final classification result of input data:

[0070] Initialize vote counts: Create corresponding vote count records for all possible accounting account categories, with the initial vote count set to 0. For example, if the accounting account categories include "Administrative Expenses - Travel Expenses," "Administrative Expenses - Office Expenses," and "Sales Expenses - Business Promotion Expenses," prepare a "counter" for each category to record the vote count, and initially record 0 votes for each.

[0071] Counting the voting results of the decision trees: After each decision tree completes its classification of the input data, the number of votes for the corresponding category is increased by 1. For example, if the first decision tree determines that a certain expense reimbursement data belongs to "Administrative Expenses - Travel Expenses", the number of votes for "Administrative Expenses - Travel Expenses" is increased by 1; if the second decision tree determines that it belongs to "Sales Expenses - Business Promotion Expenses", the number of votes for "Sales Expenses - Business Promotion Expenses" is increased by 1, and so on, until the votes of all decision trees are counted;

[0072] Find the category with the most votes: After counting the votes from all decision trees, compare the votes for each accounting category. Find the category with the most votes. If "Administrative Expenses - Travel Expenses" has the most votes of all categories, it is the "leader" for this classification.

[0073] Handling Ties: If multiple categories have the same number of votes, further processing is required. This can be done by randomly selecting one of the tied categories as the final classification. Alternatively, a secondary vote can be performed, allowing the decision trees associated with these tied categories to vote again. Alternatively, based on the actual accounting practices and prior knowledge, the categories that are more common in the company's accounting operations and more consistent with business logic can be prioritized.

[0074] Final classification results: After the above steps, the categories determined are the final classification results of the input data. This result is used for subsequent accounting data processing, such as bookkeeping and generating financial statements.

[0075] Compared with traditional simple majority voting, this weighted voting method can more reasonably integrate the classification results of the decision tree, avoid the final results affected by the incorrect classification of individual low-quality decision trees, and improve the reliability and stability of data classification. Especially when processing large-scale accounting data, it can effectively improve the classification quality.

[0076] Standardization operation stage: After completing the data classification, the classified data is standardized in strict accordance with the accounting standards and the financial system formulated within the enterprise; for example, in terms of the unification of the amount precision and the conversion of measurement units, the traditional method may lack the deep integration of accounting standards and the enterprise's personalized financial system. In the standardization process, the present invention not only follows the general accounting principles, but also combines the specific business requirements and financial accounting habits of the enterprise. For example, for some special industry enterprises, there may be higher requirements for the amount precision. On the basis of retaining two decimal places, the third decimal place needs to be specially set with rounding rules; in the conversion of measurement units, full consideration is given to the customary usage in the internal business processes of the enterprise. For example, in manufacturing enterprises, for raw material procurement data, some special specifications of measurement units are accurately converted according to the conversion table used by the enterprise for a long time.

[0077] Here we take the unification of amount precision as an example, assuming the original amount data is , the data after unified precision is For enterprises in special industries, if there are special rounding rules for the amount precision, it is assumed that the rule is that when the third decimal place is greater than or equal to Carry when less than Time to discard ( According to the enterprise requirements), the calculation formula is:

[0078]

[0079] In terms of measurement unit conversion, suppose there is a specific conversion table within the enterprise. For raw material procurement data, the original measurement unit data is , the converted unified measurement unit data is , the conversion relationship is , is the conversion function determined according to the conversion table.

[0080] Compared with traditional standardization methods, the improved method can better meet the actual financial management needs of enterprises, ensure that the processed data is seamlessly integrated into the internal financial processes of the enterprise, and improve the availability and accuracy of data in subsequent financial analysis, report generation and other links.

[0081] During the node splitting process of the decision tree, the weights are automatically assigned to the node splitting attributes of the decision tree based on the analytic hierarchy process combined with the experience of accounting experts. The specific implementation logic is as follows:

[0082] Step A: Determine the attribute set and decision target: First, train the subset For expense reimbursement data, the attributes include "reimbursement item," "reimbursement amount," "reimbursement time," and "reimbursement department." For procurement data, the attributes include "purchased item category," "purchase quantity," "supplier," and "purchase date." For sales data, the attributes include "sales item," "sales amount," "sales time," and "sales department." Accurately classifying the above accounting attribute data into corresponding accounting subjects is the decision-making goal.

[0083] Step B, construct judgment matrix: Combine the experience of accounting experts and the sample attributes of accounting subjects for different business types to construct a judgment matrix, set the training subset of The attribute set is , for each attribute data in the accounting subject sample, the judgment matrix is ​​constructed: ,in Representation attributes Relative to attributes The importance of , When constructing the judgment matrix, the 1-9 scale method is used to compare the relative importance of each attribute in each accounting subject sample. Specifically, if the attribute Relative to attributes Equally important, ;like Compare Slightly more important, ; obviously important, ; Strongly important, ; Extremely important, If it is between the two, The values ​​are 2, 4, 6, and 8 respectively;

[0084] For example, experts generally believe that "reimbursement items" directly determine the accounting items to which expenses are attributed, and are more important than "reimbursement time" in determining accounting items. Therefore, in the judgment matrix, the position of the "reimbursement items" row and the "reimbursement time" column is assigned a value of 9; according to The position of the "Reimbursement Time" row and the "Reimbursement Item" column is assigned If the experts judge that "Reimbursement Amount" and "Reimbursement Person's Department" are equally important for determining accounting items, the "Reimbursement Amount" row and "Reimbursement Person's Department" column, as well as the "Reimbursement Person's Department" row and "Reimbursement Amount" column in the matrix, are assigned a value of 1. This completes the pairwise comparison and assignment of all attributes, constructing a matrix suitable for determining attribute importance in expense reimbursement data.

[0085] Step C, calculate the attribute weight vector: use the eigenvector method to calculate the maximum eigenvalue of the judgment matrix and its corresponding eigenvector , by solving the equation , and obtain the feature vector ;

[0086] Step D, consistency test: calculate consistency index , the formula is ,in is the number of attributes; then find the average random consistency index , which is based on the number of attributes Obtained from standard tables; calculated agreement ratio , the formula is ,when When , the consistency of the received judgment matrix is ​​good, that is, the expert's judgment is reasonable, and the obtained weight vector is valid; if , then readjust the judgment matrix until the consistency test passes;

[0087] Step E: Integrate expert weight results: For the training subset For each attribute, the weights given by the experts are combined and the simple average method is used. The number of experts is set to , No. Attributes given by experts The weight of , then the integrated attributes The final weight The calculation formula is: ;

[0088] Step F, apply weights to decision tree node splitting: apply the final attribute weights obtained to the decision tree node splitting process;

[0089] The eigenvector method is used to calculate the maximum eigenvalue of the judgment matrix and its corresponding eigenvector The specific steps are as follows:

[0090] Normalize the judgment matrix by column: Assume that the constructed judgment matrix is , for each column of the matrix , calculate the sum of the elements in this column , then each element in the matrix Divide by the sum of the elements in the column , get the normalized matrix elements Through the above calculation, the entire judgment matrix is ​​normalized by column, and the normalized matrix is ​​obtained ;

[0091] Sum the normalized matrix row by row: For the normalized matrix , calculate the sum of the elements in each row ;

[0092] Normalize the rows and vectors to get the eigenvectors : Rows and vectors Normalize, that is, calculate , and obtain the feature vector , the eigenvector Each component in The relative weight of each attribute in the corresponding accounting subject sample;

[0093] Calculate the maximum eigenvalue :calculate , and get the vector , then calculate ,in is a vector No. elements, the largest eigenvalue Used for subsequent consistency testing to determine the rationality of expert judgment.

[0094] Example 2

[0095] See also Figure 3, an automatic process processing system for accounting data, including a data acquisition module, a data classification and standardization module, an automated process processing module, and a data storage and analysis module; the data acquisition module includes an adapter interface development component and an OCR scanning and data cleaning component, which can realize data connection with the internal business system of the enterprise and data collection and cleaning of paper bills; the data classification and standardization module integrates a machine learning algorithm module and a data standardization processing component, which can complete the accurate classification and standardization operations of data; the automated process processing module has a process rule preset component and a monitoring and early warning component, which can automatically process data flow according to preset rules and issue timely early warnings when abnormalities occur; the data storage and analysis module includes a data encryption storage component, a report generation component and a data visualization component to realize secure storage of data, report generation and visual display.

[0096] Example 3

[0097] According to an exemplary embodiment, an electronic device includes: a processor and a memory, wherein the memory stores a computer program that can be called by the processor;

[0098] The processor executes the above-mentioned automatic process processing system for accounting data by calling the computer program stored in the memory.

[0099] Figure 4 This is a structural diagram of an electronic device provided in an embodiment of the present application. The electronic device may have relatively large differences due to different configurations or performances, and can include one or more processors (Central Processing Units, CPU) and one or more memories, wherein at least one computer program is stored in the memory, and the at least one computer program is loaded and executed by the processor to implement an automatic process processing system for accounting data provided by the above-mentioned various method embodiments.

[0100] The electronic device may also include other components for realizing the functions of the device, for example, the electronic device may also include components such as a wired or wireless network interface and an input / output interface for input and output.

[0101] This embodiment also provides a computer program product stored on a computer-readable medium, including a computer-readable program, which, when executed on an electronic device, provides a user input interface to implement the aforementioned automatic process processing system for accounting data.

[0102] It should be understood that in the various embodiments of the present application, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0103] It should be understood that determining B based on A does not mean determining B based solely on A. B can also be determined based on A and / or other information.

[0104] The present invention discloses a method and system for automatic process processing of accounting data. The method first uses an adapter interface to collect data from the company's internal business systems such as OA, procurement, and sales, and uses optical character recognition technology to process paper bills, while cleaning the data based on the accounting professional vocabulary library and grammatical rules. Then, an improved random forest algorithm is used to build a classification model, and the data is accurately classified and standardized with reference to the company's accounting subject system. Then, according to the preset corporate financial approval and account processing process rules, business processes such as expense reimbursement and procurement payment are automatically processed, and the data flow and processing progress are monitored in real time, and timely warnings are issued for anomalies. Finally, the processed data is encrypted and stored in the company database, a backup mechanism is established, access rights are set, and financial statements are generated and visualized using data analysis tools. The invention realizes the automation and intelligence of accounting data processing, effectively improves data processing efficiency and accuracy, enhances data security, provides strong decision-making support for corporate financial management, and helps companies improve their competitiveness.

[0105] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A method for automatic process processing of accounting data, characterized in that: The following steps are involved: Step 1: Data collection: Use the adapter interface to connect to the company's OA, procurement, and sales internal business systems. Daily, electronic expense reimbursement forms submitted by employees are captured from the OA system. Purchase orders and corresponding invoice data are obtained from the procurement system. Then, optical character recognition technology is used to scan paper invoices and bank statements. After scanning, incorrect characters recognized by OCR are automatically verified and corrected based on the accounting professional vocabulary and grammatical rules. At the same time, the data format is unified, duplicate data is removed, and missing values ​​are filled. This completes data cleaning and obtains the collected data. Step 2: Data classification and standardization: Using an improved machine learning algorithm, build a classification model based on the company's accounting system. Input the cleaned data collected in step 1 into the classification model and accurately classify it into the corresponding accounting categories. Step 3: Automated Processing: The reimbursement data, purchase data, and sales data categorized in Step 2 are processed according to the company's financial approval and accounting processes, where the processing process is based on pre-set process rules. Step 4: Data Storage and Analysis: The accounting data processed in Step 3 is encrypted and stored in the enterprise database. A regular backup mechanism is established to prevent data loss. Access permissions are set for different data based on personnel responsibilities to ensure data security. At the same time, data analysis tools and algorithms are used to automatically generate financial statement data, including but not limited to balance sheets, income statements, and cash flow statements, based on the stored data. Financial data is presented in intuitive charts using data visualization technology. In step 2, an improved machine learning algorithm is used, specifically an improved random forest algorithm that assigns different weights to different attributes based on accounting expertise. A classification model is constructed using the improved random forest algorithm, and the collected data is automatically and accurately classified using the classification model. The specific implementation steps of the improved random forest algorithm are as follows: Data preparation stage: From the historical accounting data of the enterprise, a sample data set of accounting subjects is constructed by using stratified sampling combined with dynamic proportion adjustment. Suppose the historical accounting data set is , data set Contains various accounting data samples in different proportions, including but not limited to expense reimbursement data , procurement data , sales data The corresponding accounting subject samples; in stratified sampling, first Divided by business type non-overlapping layers, i.e. , for each layer , the data volume is recorded as and the importance is recorded as , importance It is determined by accounting experts based on a comprehensive assessment of the impact of different business type data on accounting subject classification, with a value range of ,and , dynamically adjust the extraction ratio according to the amount and importance of each layer of data After calculating the extraction ratio of each layer according to the above formula, randomly extract data from each layer according to the corresponding ratio to construct a diverse training subset ; Decision tree training phase: for each constructed training subset , respectively carry out the decision tree training work; in the process of node splitting of the decision tree, based on the hierarchical analysis method combined with the experience of accounting experts, automatically assign weights to the decision tree node splitting attributes, and according to the calculated weight distribution method, for each training subset Perform recursive partitioning and construct multiple decision trees; Random forest integration stage: the above-trained multiple decision trees are combined into a random forest classification model, i.e., a classification model; Data classification execution phase: The collected data cleaned in step 1 is input into the constructed classification model. Each decision tree in the classification model independently classifies the input data. Each decision tree analyzes the various features of the input data based on its own training rules and attribute weights, and then generates a classification result. After all decision trees have completed the classification, a voting mechanism is used to determine the final classification result of the input data. Standardization operation stage: After completing data classification, the classified data is standardized in strict accordance with accounting standards and the financial system established within the enterprise; During the node splitting process of the decision tree, the weights are automatically assigned to the node splitting attributes of the decision tree based on the analytic hierarchy process combined with the experience of accounting experts. The specific implementation logic is as follows: Step A: Determine the attribute set and decision target: First, train the subset Attributes related to the data. Specifically, for expense reimbursement data, these attributes include "Reimbursement Item," "Reimbursement Amount," "Reimbursement Time," and "Reimbursement Person's Department." For procurement data, these attributes include "Purchased Item Category," "Purchased Quantity," "Supplier," and "Purchasing Date." For sales data, these attributes include "Sales Item," "Sales Amount," "Sales Time," and "Sales Person's Department." Accurately classifying these accounting attribute data into corresponding accounting subjects serves as the decision-making goal. Step B, construct judgment matrix: Combine the experience of accounting experts and the sample attributes of accounting subjects for different business types to construct a judgment matrix and set the training subset of The attribute set is , for each attribute data in the accounting subject sample, the judgment matrix is ​​constructed: ,in Representation attributes Relative to attributes The importance of , When constructing the judgment matrix, the 1-9 scale method is used to compare the relative importance of each attribute in each accounting subject sample. Specifically, if the attribute Relative to attributes Equally important, ;like Compare Slightly more important, ; obviously important, ; Strongly important, ; Extremely important, If it is between the two, The values ​​are 2, 4, 6, and 8 respectively; Step C, calculate the attribute weight vector: use the eigenvector method to calculate the maximum eigenvalue of the judgment matrix and its corresponding eigenvector , by solving the equation , and obtain the feature vector ; Step D, consistency test: calculate consistency index , the formula is ,in is the number of attributes; then find the average random consistency index , which is based on the number of attributes Obtained from standard tables; calculated agreement ratio , the formula is ,when When , the consistency of the received judgment matrix is ​​good, that is, the expert's judgment is reasonable, and the obtained weight vector is valid; if , then readjust the judgment matrix until the consistency test passes; Step E: Integrate expert weight results: For the training subset For each attribute, the weights given by the experts are combined and the simple average method is used. The number of experts is set to , No. Attributes given by experts The weight of , then the integrated attributes The final weight The calculation formula is: ; Step F, apply weights to decision tree node splitting: apply the final attribute weights obtained to the decision tree node splitting process; in Calculate the maximum eigenvalue of the judgment matrix using the eigenvector method and its corresponding eigenvector The specific steps are as follows: Normalize the judgment matrix by column: Assume that the constructed judgment matrix is , for each column of the matrix , calculate the sum of the elements in this column , then each element in the matrix Divide by the sum of the elements in the column , get the normalized matrix elements Through the above calculation, the entire judgment matrix is ​​normalized by column, and the normalized matrix is ​​obtained ; Sum the normalized matrix row by row: For the normalized matrix , calculate the sum of the elements in each row ; Normalize the rows and vectors to get the eigenvectors : Rows and vectors Normalize, that is, calculate , and obtain the feature vector , the eigenvector Each component in The relative weight of each attribute in the corresponding accounting subject sample; Calculate the maximum eigenvalue :calculate , and get the vector , then calculate ,in is a vector No. elements, the largest eigenvalue Used for subsequent consistency testing to determine the rationality of expert judgment.

2. The method for automatic process processing of accounting data according to claim 1, characterized in that: The specific contents of the automated process in step 3 are as follows: Preset process rules: For reimbursement and sales processes, the company presets the approvers and approval order for different amount ranges, reimbursement and sales items. For the purchase and payment process, the company determines the approval levels based on factors such as purchase order amount and supplier type. These pre-set rules are then compiled into code logic to serve as the basis for subsequent automated process execution. Reimbursement and sales process processing: according to the classification results of the collected data in step 2 of the invention, the expense reimbursement data and sales data are automatically screened out; according to the preset rules, the corresponding approver of each reimbursement and sales order is determined, and the reimbursement order is sent to the approver's work interface through an internal message push mechanism or integration with the enterprise office software, along with detailed reimbursement and sales information, including but not limited to the name of the reimbursement and sales person, department, reimbursement item, amount and bill; after receiving the reimbursement and sales order, the approver performs the approval operation and chooses to approve, reject or return for modification; if the approver approves, the data is automatically transferred to the financial accounting link; if rejected or returned for modification, the reimbursement or sales person is notified in a timely manner and the reason is explained; in the financial accounting link, based on the reimbursement and sales order data, in accordance with accounting standards and corporate financial systems, accounting vouchers are automatically generated to complete the accounting processing; Procurement payment process processing: According to the classification results of the collected data in step 2 of the invention, the procurement data is automatically screened out, and the purchase order data and the associated invoice data in the procurement data are monitored in real time; When the purchase data information matches and meets the preset payment conditions, a payment application is automatically generated, including but not limited to the purchase order number, supplier information, payment amount, and payment method. After the payment application is generated, it is pushed to the corresponding approvers in sequence according to the preset approval hierarchy rules. The approvers perform the approval operation. Once the approval is passed, if all levels of approval are completed, an instruction is automatically sent to the company's funds payment system, triggering the payment operation and completing the payment. Data flow and processing progress monitoring: Real-time monitoring of the input data flow path and processing progress in the reimbursement and sales processes, as well as the purchase and payment processes. This is achieved by setting markers and timestamps at each process node to record the time when data enters and leaves each link, as well as its current location. Exception handling and reminders: Continuously check whether there are any exceptions in the data flow and processing process; if the approval timeout occurs, immediately activate the reminder mechanism and send a reminder email or text message to the approver, reimbursement person or relevant business manager to urge them to complete the approval operation; if it is found that the data does not comply with the preset rules, send a notification to inform the details of the exception and suspend the relevant process until the problem is resolved.

3. An automatic process-based accounting data processing system, characterized by: The method for automatic process-based processing of accounting data according to any one of claims 1 to 2 comprises a data acquisition module, a data classification and standardization module, an automated process processing module, and a data storage and analysis module; the data acquisition module comprises an adapter interface development component and an OCR scanning and data cleaning component, which realize data connection with the enterprise's internal business system and data collection and cleaning of paper bills; the data classification and standardization module integrates a machine learning algorithm module and a data standardization processing component to complete accurate classification and standardization operations of data; the automated process processing module comprises a process rule preset component and a monitoring and early warning component, which automatically processes data processes according to preset rules and promptly issues early warnings when anomalies occur; The data storage and analysis module includes a data encryption storage component, a report generation component and a data visualization component to achieve secure data storage, report generation and visual display.

4. An electronic device, characterized in that: include: A processor and a memory, wherein the memory stores a computer program that can be called by the processor: The processor calls the computer program stored in the memory to execute and implement the automatic process processing system for accounting data as described in claim 3.

5. A computer program product stored on a computer-readable medium, characterized in that: It includes a computer-readable program that, when executed on an electronic device, provides a user input interface to implement an automatic flow processing system for accounting data as described in claim 3.

Images