Multi-terminal ecosystem fund allocation management system for e-commerce enterprise

Abstract

The present application provides a multi-terminal ecosystem fund allocation management system for an e-commerce enterprise, comprising: acquiring tax jurisdiction information of each business entity, comparing tax policies of different regions, and determining whether there is a difference between the tax policies; if there is a difference between the tax policies, acquiring from financial systems of each business entity fund distribution-related transaction data and tax information, and obtaining a unified-format tax information dataset by means of data cleaning and integration; using an association rule mining technology, finding an association mode between fund allocation and tax processing from the tax information dataset, so as to form a tax association rule library for fund allocation; on the basis of the tax association rule library for fund allocation, constructing a tax benefit evaluation model for fund allocation, and predicting a tax impact of a solution by means of inputting different fund allocation solution parameters.

Description

A multi-terminal ecosystem fund sharing management system for e-commerce enterprises Technical Field

[0001] This invention relates to the field of information technology, and in particular to a multi-terminal ecosystem fund sharing management system for e-commerce enterprises. Background Technology

[0002] E-commerce companies operate across multiple platforms and ecosystems, involving various business entities and transaction scenarios, making the allocation of funds among these entities and scenarios highly complex. Because each business entity is located in a different tax jurisdiction and is subject to different tax policies, the formulation of a fund allocation plan requires a comprehensive assessment of the tax implications. Technical issues

[0003] However, when e-commerce companies actually conduct cross-entity and cross-regional cost-sharing, they often find it difficult to comprehensively collect tax information from each entity and accurately assess the tax benefits of different cost-sharing schemes. At the same time, the rapid pace of business innovation and diverse transaction models on e-commerce platforms lead to constantly evolving tax issues, further increasing the difficulty of assessing the tax benefits of cost-sharing. Furthermore, some emerging businesses lack clear tax policies, and the lack of standardized guidelines for the tax treatment of cost-sharing also poses challenges to tax benefit assessment. Therefore, it is urgent to establish a sound tax assessment mechanism for cost-sharing, dynamically track changes in tax policies, deeply analyze the cash flow process in business scenarios, accurately quantify the tax impact of cost-sharing, provide strong support for cost-sharing decisions, and achieve a balance between tax benefits and business development. Technical solutions

[0004] This invention provides a multi-terminal ecosystem fund sharing management system for e-commerce enterprises, mainly comprising:

[0005] Obtain information on the tax jurisdiction of each business entity, compare the tax policies of the different regions, and determine whether there are differences in tax policies.

[0006] If the aforementioned tax policy differences exist, then the relevant transaction data and tax information on fund allocation are obtained from the financial systems of each business entity, and a unified format tax information dataset is obtained through data cleaning and integration.

[0007] By using association rule mining technology, the association patterns between fund allocation and tax treatment are discovered from the tax information dataset, forming a tax association rule library for fund allocation;

[0008] Based on the tax-related rule base of the aforementioned fund allocation, a fund allocation tax benefit assessment model is constructed. By inputting parameters of different fund allocation schemes, the tax impact of the schemes is predicted.

[0009] For emerging businesses, the emerging businesses are grouped according to their business attributes and capital flow characteristics. Emerging business groups similar to existing businesses are identified, and the tax treatment experience of similar business groups is referenced to determine the tax risks of the emerging businesses.

[0010] Establish a knowledge base for tax assessment of fund allocation, and store and manage tax policy changes, fund allocation cases, and tax benefit assessment results in a structured manner. When new business scenarios or tax policy changes occur, relevant information can be retrieved from the knowledge base to assist in the tax benefit assessment of fund allocation. Beneficial effects

[0011] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects:

[0012] This invention discloses a method for assessing the tax benefits of cross-regional capital allocation for business entities. The method first obtains information on the tax jurisdictions of each business entity, compares the differences in tax policies across regions, and retrieves relevant transaction data and tax information on capital allocation from the financial system. It then uses association rule mining technology to discover the correlation patterns between capital allocation and tax treatment, constructing a capital allocation tax benefit assessment model. For emerging businesses, they are grouped according to business attributes and capital flow characteristics, and tax risks are assessed by referring to the tax treatment experience of similar business groups. This invention also establishes a knowledge base for capital allocation tax assessment, storing tax policy changes, capital allocation cases, and assessment results, providing support for tax benefit assessment in new business scenarios or when policies change. This method can effectively assess the tax impact of cross-regional capital allocation schemes, optimize capital allocation strategies, reduce tax risks, and improve the overall tax benefits of enterprises. Attached Figure Description

[0013] Figure 1 is a flowchart of a multi-terminal ecosystem fund sharing management system for e-commerce enterprises according to the present invention.

[0014] Figure 2 is a schematic diagram of a multi-terminal ecosystem fund sharing management system for e-commerce enterprises according to the present invention.

[0015] Figure 3 is another schematic diagram of a multi-terminal ecosystem fund sharing management system for e-commerce enterprises according to the present invention. Embodiments of the present invention

[0016] The technical solutions of the embodiments of the present invention will be clearly and thoroughly described below with reference to the accompanying drawings. The described embodiments are merely some embodiments of the present invention.

[0017] As shown in Figures 1-3, the multi-terminal ecosystem fund sharing management system for e-commerce enterprises in this embodiment may specifically include:

[0018] Step S101: Obtain the tax jurisdiction information of each business entity, compare the tax policies of different regions, and determine whether there are differences in tax policies.

[0019] The process involves: acquiring the registered location information of at least one business entity and determining its tax jurisdiction; acquiring the corresponding tax policy documents for each jurisdiction; performing text analysis on the tax policy documents using natural language processing (NLP) to extract key tax elements such as tax rates, tax bases, and tax payment periods; storing these key tax elements in a pre-defined structured database; performing cross-regional comparative analysis on the key tax elements in the structured database using a data comparison tool to identify tax policy differences between different regions; quantifying and evaluating the identified tax policy differences to calculate a tax policy difference index; constructing a tax risk scoring function using a logistic regression model based on the tax policy difference index and the business entity's financial data; inputting the relevant data of the business entity into the tax risk scoring function to calculate its tax risk score; determining whether the tax risk score exceeds a preset threshold; classifying the business entity as high-risk if it does, and as medium-to-low-risk if it does not; developing corresponding tax compliance management measures and risk control recommendations for business entities at different risk levels; and generating a tax risk assessment report containing the risk rating results and control recommendations.

[0020] Specifically, the registered location information of a multinational company, Company A, was obtained through the Enterprise Credit Information Disclosure System, confirming that it falls under the tax jurisdiction of Pudong New Area, Shanghai. For this region, the latest corporate income tax policy documents were obtained from the Shanghai Electronic Tax Bureau website. A BERT model was used to perform semantic analysis on the policy documents, extracting key tax elements such as a 25% corporate income tax rate, tax base being taxable income, and a tax payment deadline of May 31st of the following year, which were then stored in a MongoDB database in JSON format. A Python script was written to compare the tax elements of Pudong New Area, Shanghai, with data from Haidian District in Beijing and Nanshan District in Shenzhen, using a cosine similarity algorithm to calculate the difference. The results showed a 12% difference in corporate income tax policy between Pudong New Area and other regions. Based on this difference index, and combined with Company A's financial data for the past three years, a logistic regression model was constructed using the sklearn library, yielding the tax risk scoring function: Score = 1 / (1 + e^(-8X1 - 6X2 + 4X3)), where X1 is total profit, X2 is income tax expense, and X3 is deferred tax assets. By inputting Company A's relevant financial indicators into a function, its tax risk score was calculated to be 62 points, exceeding the preset threshold of 50 points. Therefore, Company A is classified as a high-risk company. For high-risk enterprises, the tax authorities should strengthen tax audits. It is recommended that Company A engage a third-party tax firm to conduct a compliance self-assessment and provide corresponding policy consultation and risk warning services regarding potential tax planning activities, and submit a tax risk assessment report to management for reference.

[0021] Step S102: If the tax policy differences exist, obtain the relevant transaction data and tax information on fund allocation from the financial systems of each business entity, and obtain a unified format tax information dataset through data cleaning and integration.

[0022] If differences in tax policies exist among various business entities, transaction data and tax information related to fund allocation are obtained from the financial systems of each entity. The obtained transaction data and tax information are preprocessed, using Excel's data filtering function to remove null and outlier values, resulting in a pre-cleaned dataset. Based on pre-defined data integration rules, SQL statements are used to integrate the cleaned data, converting data from different sources and formats into a unified data table. Excel's pivot table function is used to extract key fields related to tax information, including transaction amount, tax rate, and tax type, constructing a tax information feature table. The feature table is imported into the R language environment, and a tax information classification model is built using the random forest algorithm. The model performance is evaluated using cross-validation to ensure that the classification accuracy reaches the target. Set a threshold; use a trained random forest model to classify the integrated data and obtain the tax attribute of each data point; based on the classification results, use the SQL GROUPBY statement to divide data with the same tax attribute into different data tables, forming subsets with different tax attributes; perform sampling checks on each subset to ensure the accuracy of the classification results; if errors are found, use the SQL UPDATE statement to manually correct them; finally, use the SQL UNIONALL statement to merge the various subsets to obtain a master data table with a unified format and complete tax information; perform data quality checks on the master data table, including verification of completeness, consistency, and accuracy; if problems are found, return to the corresponding steps for correction; after verification, export the master data table as CSV format as input data for subsequent tax processing.

[0023] Specifically, detailed financial transaction data for 2022 was retrieved from the SAP systems of Companies A, B, and C, totaling 50,000 records. Using Excel's data filtering function, records with transaction amounts of 0 or empty tax rates were removed, resulting in 80,000 valid records. Based on pre-defined data integration rules, SQL statements were used to standardize fields such as transaction time and transaction object in Company A's data table, converting the amount field in Company B's data table from USD to RMB, and converting the tax type field in Company C's data table from code to text description, ultimately resulting in a unified data table format. An Excel pivot table was then used to construct a tax information feature table containing 20,000 records, based on transaction amount, tax rate, and tax type. This feature table was imported into the R language environment, and a tax information classification model was built using the random forest algorithm. Through 5-fold cross-validation, the model achieved a classification accuracy of 92%. This model was then used to classify the 80,000 integrated data records by tax attribute, resulting in 50,000 records of value-added tax, 8,000 records of corporate income tax, and 5,000 records of individual income tax. Based on the classification results, three subsets were generated using the SQL GROUPBY statement. Sampling checks revealed 200 incorrect tax rates in the corporate income tax data, which were corrected using the UPDATE statement. Finally, the three subsets were merged using the UNION ALL statement to obtain a master data table with a unified format and complete tax information. The master data table underwent a data quality check, and after verification of completeness, consistency, and accuracy, the data quality was deemed satisfactory. The master data table was exported as a CSV file to serve as input data for subsequent tax processing.

[0024] Step S103: Using association rule mining technology, discover the association patterns between fund allocation and tax processing from the tax information dataset to form a tax association rule library for fund allocation.

[0025] The process involves: acquiring a tax information dataset, including taxpayer information, transaction records, and invoice data; performing data cleaning using SQL to remove duplicate records, outliers, and missing data; integrating the data using ETL tools to unify the format and encoding of data from different sources; transforming the data using Python, employing either standardized numerical features or one-hot encoded categorical features; conducting exploratory data analysis on the preprocessed dataset, using descriptive statistics and visualization tools to identify data distribution and potential patterns; selecting appropriate feature subsets for subsequent modeling based on the analysis results; using the Apriori algorithm for association rule mining, setting a minimum support of 0.5 and a minimum confidence of 7 to obtain an initial set of association rules between fund allocation and tax treatment; and filtering the rules using a lift metric. Rules with an improvement of less than 1 are removed; the remaining rules undergo manual review, and unreasonable rules are removed using domain knowledge; the filtered rules are categorized according to tax business type and taxpayer category to construct a multi-layered rule base structure; a Java-based rule engine is developed and the rule base is imported into it; for new tax data, the rule engine is used for matching and inference, outputting compliance assessment results for fund allocation behavior and corresponding tax treatment suggestions; after review by tax officials, the processing results are fed back to the system; the dataset is updated regularly and the association rule mining process is re-executed; the differences between the old and new rule sets are compared to identify newly emerging patterns; the rule base is dynamically updated and optimized based on actual application effects; throughout the process, data anonymization technology is used to protect sensitive information, and access control and audit log mechanisms are used to ensure data security. A web application is developed to provide rule query, editing, and statistical analysis functions. A visualization component is integrated to display the relationships between rules and application effects in chart form. The application interface and functions are continuously optimized based on user feedback and system usage data to improve the work efficiency of tax officials.

[0026] Specifically, 5 million tax data records for 2022 were obtained from enterprise ERP systems, tax filing systems, and third-party e-invoice platforms, including taxpayer registration information, sales and input invoices, and fund transfer records. SQL Server's data cleaning function was used to remove duplicate records through uniqueness constraints, set value ranges based on business rules, eliminate abnormal transaction amounts and tax rates, and fill missing value fields with the mean or mode, resulting in a cleaned dataset of 4.5 million records. Informatica PowerCenter was used to perform schema matching on the data tables from the three data sources, unifying field names and data types, and converting them to a unified data format and encoding standard. Python's Pandas and Scikit-learn libraries were used to perform min-max normalization on numerical features such as transaction amounts and tax rates, and one-hot encoding transformation on categorical features such as taxpayer type and tax type. The preprocessed dataset was visualized and analyzed using Tableau. Histograms and box plots were used to identify the distribution of each field, and scatter plots and radar charts were used to discover the correlation patterns between tax elements. Twenty feature subsets closely related to fund allocation, such as transaction amount, taxpayer type, and tax type, were selected. The FP-Growth algorithm was used for association rule mining, with a minimum support of 0.5 and a minimum confidence of 7, resulting in 5000 initial rules. The lift of each rule was calculated, and 2000 rules with a lift less than 1 were removed. Tax experts reviewed the remaining 3000 rules, removing 500 unreasonable rules, such as "transaction amount is not 0 when tax rate is 0". The remaining 2500 rules were categorized according to tax type (e.g., value-added tax, corporate income tax) and taxpayer type (e.g., large enterprises, SMEs), constructing a three-layer rule base structure and storing it in a MySQL database. Based on the Drools rule engine, a tax processing inference program was developed using Java. This program converts rules from the rule base into Drools rule file format, performing rule matching, conflict resolution, and result output during the inference process. Each time new tax return data is processed, the system automatically invokes the inference program to assess the compliance of the taxpayer's cost-sharing behavior, generating a tax processing recommendation report which is then sent to tax officials for review. Tax officials confirm or modify the system's recommendations, feeding the results back into the system for iterative rule base updates. A rule mining process is automatically triggered every quarter, adding new tax data to the training set, mining new association rules, comparing them with the existing rule set, identifying newly emerging cost-sharing patterns, and dynamically updating the rule base. During system application, the MD5 algorithm is used to encrypt and anonymize sensitive fields such as taxpayer identification numbers and invoicing information. Role-based access control mechanisms and operational auditing functions are used to ensure the confidentiality, integrity, and traceability of tax data.Finally, a tax processing web application was developed using the Spring framework, embedding the Drools rule engine to provide functions such as rule base management, tax data processing, and statistical analysis. Visualization libraries such as Echarts were used to present analysis results such as rule relationship network diagrams and tax violation risk distribution maps. After the system went live, feedback from tax personnel was continuously collected to optimize business processes and interface layout, improving the system's usability and work efficiency.

[0027] Step S104: Based on the tax-related rule library of the fund allocation, construct a fund allocation tax benefit assessment model, and predict the tax impact of different fund allocation schemes by inputting different fund allocation scheme parameters.

[0028] Data on cost-sharing schemes, including the amount, recipient, and proportion, is obtained from the enterprise's financial system. Tax rule data, including tax types, rates, and tax base information, is also obtained from the tax system. The cost-sharing scheme data and tax rule data are mapped together to establish a cost-sharing tax-related rule library. Historical cost-sharing scheme data and its actual tax impact are preprocessed, including data cleaning and normalization. The preprocessed data is used to train decision tree and logistic regression models as the foundation for cost-sharing tax benefit assessment. The prediction results from the trained decision tree and logistic regression models are used as input to construct a simple feedforward neural network. This neural network is trained using historical data to obtain a fused cost-sharing tax benefit assessment model. New cost-sharing scheme parameters are input, and the relevant tax rules are determined from the cost-sharing tax-related rule library. The cost-sharing scheme parameters and the relevant tax rules are then input into the fused model to obtain predictions. The system assesses the tax impact of each funding allocation scheme. Based on the predicted tax impact, a comprehensive tax impact score is calculated. The score is calculated using a weighted average method, with weights pre-set according to the importance of each tax impact. The comprehensive tax impact score is compared with a preset tax benefit threshold to determine the tax benefit of the allocation scheme. This assessment process is repeated for multiple funding allocation schemes, ranking them based on their comprehensive tax impact scores. The top three schemes with the highest scores are selected as recommended optimization schemes for funding allocation. The model is regularly updated using new historical data, including retraining the base model and the fusion model, to adapt to the constantly changing tax environment. The model's predicted results are compared with the actual tax impact to continuously optimize model performance. During the assessment process, if any anomalies are encountered, such as missing data or data exceeding expectations, the system will record the anomaly and prompt for manual intervention. For the final recommended optimization scheme for funding allocation, a detailed tax impact analysis report is generated to provide a reference for corporate decision-making.

[0029] Specifically, 1000 records of 2022 fund allocation scheme data, including allocation amount, allocation department, and allocation ratio, were extracted from the enterprise ERP system. Tax rules, including tax rates and calculation methods for value-added tax and corporate income tax, were obtained from the Golden Tax System Phase IV. Using Python's Pandas library, a left join operation was performed between the fund allocation scheme data and the tax rule data using a unique ID field to match the tax rules associated with each allocation record, constructing a fund allocation tax association rule library. The 1000 historical fund allocation scheme records were cleaned, removing records with amounts of 0 or empty, and the allocation ratios were normalized, scaling the numerical range to between 0 and 1. Using decision tree and logistic regression algorithms, a tax benefit assessment model was trained with allocation amount, allocation ratio, and tax rate as features and actual tax burden impact as labels. 10-fold cross-validation was used, achieving average accuracies of 85% and 82%, respectively. The prediction results of decision trees and logistic regression are input into a feedforward neural network with three hidden layers and ten nodes per layer. The model is trained using backpropagation with mean squared error as the loss function. After 500 iterations, a fused tax benefit assessment model is obtained, achieving an accuracy of 88%. For a new cost-sharing scheme, such as allocating 1 million yuan to departments A, B, and C in a 3:5:2 ratio, relevant tax rules, including a 13% VAT rate and a 25% corporate income tax rate, are matched using an association rule base. The allocation parameters and tax rules are input into the fused model, predicting after-tax cost savings of 300,000 yuan and a 2% reduction in overall tax burden. Based on the weights of six indicators (cost savings, tax reduction, etc.) pre-set by the company, a weighted average method is used to calculate the overall tax impact score of the scheme, which is 85 points. The other nine alternative cost-sharing schemes are evaluated again. Based on the overall score, the top three schemes are selected: Scheme 2 (91 points), Scheme 5 (88 points), and Scheme 7 (86 points). Each quarter, the model is retrained using the latest 1,000 historical data points. A sliding window method is then used to compare the predicted results with the actual tax impact, calculate the prediction bias, and continuously optimize the model. For the three recommended cost-sharing optimization schemes, the system automatically generates a tax impact analysis report including dimensions such as tax type, tax base, tax rate, tax payment time, and tax-related risks, for reference by finance personnel and management.

[0030] Step S105: For emerging businesses, group the emerging businesses according to their business attributes and capital flow characteristics, identify emerging business groups similar to existing businesses, and judge the tax risks of the emerging businesses by referring to the tax treatment experience of similar business groups.

[0031] Based on the business attributes and cash flow characteristics of emerging businesses, K-means clustering algorithm is used to group emerging businesses into multiple emerging business groups. For each emerging business group, cosine similarity calculation is used to identify the most similar existing business group. Tax processing experience data, including tax type identification, applicable tax rate, and input tax deduction information, is obtained from similar existing business groups. Based on this experience data, a random forest algorithm is used to construct a tax risk assessment model. The model inputs business attributes and cash flow data, and the output is a tax risk score. The business attributes and cash flow data of the emerging business group are then input into the tax risk assessment model to calculate the tax risk score for that emerging business group. Tax risk scores are set; a tax risk score threshold is set, such as 7. If the calculated score is greater than or equal to 7, it is judged as high risk; if it is less than 7 but greater than or equal to 4, it is medium risk; and if it is less than 4, it is low risk. For high-risk business groups, the system automatically sends risk warning notifications to business personnel, prompting them to take tax risk prevention and control measures. The system extracts the latest operational data of emerging business groups from the business database every day, including transaction amount, transaction frequency and customer type, inputs it into the tax risk assessment model, updates the risk score and level, and realizes dynamic monitoring of tax risks. If the risk level changes, the system will generate a risk change report and push it to relevant business personnel and tax specialists.

[0032] Specifically, emerging business data from 2022 was extracted from the enterprise's ERP system, including dimensions such as business type, transaction amount, transaction frequency, and customer type, totaling 5,000 records. Using the K-means clustering algorithm with K set to 5, these 5,000 emerging business data records were clustered into five emerging business groups: A, B, C, D, and E. For each emerging business group, cosine similarity was calculated to identify the three business groups with the highest similarity in the existing business data. For example, for emerging business group A, the most similar existing business groups were A1, A2, and A3, with similarities of 8, 7, and 6, respectively. From the historical data of the three existing business groups A1, A2, and A3, tax processing experience data involving taxes such as value-added tax, corporate income tax, and stamp duty were extracted, including tax rates, tax calculation methods, and tax payment times, totaling 120 records. The business attributes and cash flow data of emerging business group A were merged with the 120 tax experience data records, and a tax risk assessment model was trained using the random forest algorithm. The model parameters were then optimized using grid search, resulting in a model with an accuracy of 90%. Inputting five business attributes and eight cash flow indicators of emerging business group A into the model, the predicted tax risk score is 2 points, higher than the preset threshold of 7 points, thus classifying it as a high-risk business group. The system automatically generates a tax risk warning notification for business group A and pushes it to Zhang San, the business department head, prompting him to develop a tax risk response plan. The system automatically extracts the latest operational data of business group A from the business database daily, inputs it into the tax risk assessment model, and recalculates the risk score. For example, if the risk score updated on December 28th is 5 points, lower than the threshold of 7 points, the risk level is downgraded from high risk to medium risk. The system generates a risk change report and pushes it to Zhang San and tax manager Li Si, indicating that the risk level has changed and the risk response plan needs to be adjusted.

[0033] Step S106: Establish a knowledge base for tax assessment of fund allocation, and store and manage tax policy changes, fund allocation cases, and tax benefit assessment results in a structured manner. When new business scenarios or tax policy changes occur, relevant information is retrieved from the knowledge base to assist in the tax benefit assessment of fund allocation.

[0034] The system acquires the latest tax policy changes and compares them with policy data in the existing tax assessment knowledge base for fund allocation. If new policies or changes are found, the corresponding entries in the knowledge base are updated. Simultaneously, new fund allocation cases are collected, and key information including allocation methods, amounts involved, and relevant parties is extracted. This information is entered into the knowledge base, and preliminary assessments are conducted on the cases based on preset tax benefit indicators, including tax burden rate and cash flow impact. A relational database is used to store the updated tax policies, fund allocation cases, and tax benefit assessment results. An index is created to optimize query efficiency. When new business scenarios emerge, relevant tax policies and historical allocation cases are retrieved from the database based on scenario characteristics, including transaction type and amount involved. Using the retrieved information, the K-nearest neighbor algorithm is employed to analyze the new business scenarios. The tax benefit assessment of the cost-sharing scheme involves several steps: calculating the similarity between the new scenario and historical cases, selecting the K most similar cases, and predicting the potential benefits of the new scheme based on the tax benefit results of these cases; then using a random forest algorithm to assess the tax risk of the cost-sharing scheme; first, training the model based on historical cases, with input features including the cost-sharing method, the amount involved, and the type of entity, and outputting the risk level; then inputting the new scheme into the model to obtain the risk prediction result; if the prediction result exceeds a preset threshold, the system automatically issues a risk warning; finally, the cost-sharing scheme, tax benefit assessment results, and risk assessment results under the new business scenario are entered into the database; new data is regularly analyzed to update the parameters of the assessment model and the risk assessment model to adapt to policy changes and new business scenarios; in this way, a continuously optimized closed-loop system is formed.

[0035] Specifically, the system automatically retrieves the latest tax policy announcements from the tax bureau's website daily, extracting key information such as policy name, release date, effective date, applicable entities, and preferential content, and compares it with the existing 2,000 policy data entries in the knowledge base. On December 28, 2022, the system discovered a new corporate income tax preferential policy that reduced the tax rate from 25% to 20%, applicable to small enterprises with annual taxable income below 1 million yuan. The system automatically added this policy to the knowledge base and updated the 100 affected tax assessment rules. Simultaneously, the system collected 50 new fund allocation cases from the internal ERP system and external case databases, extracting 20 key characteristics of each case, such as allocation amount, number of participating entities, entity type, and business type, and entered them into the knowledge base. Based on five preset tax benefit indicators, the system uses a weighted average method to score the new cases, obtaining a comprehensive tax benefit score for each case. When the marketing department proposes a new business scenario involving five subsidiaries jointly investing 10 million yuan to establish a new joint venture, the system, based on the scenario's characteristics, uses an inverted index to retrieve the 10 most relevant tax policies and 20 similar historical allocation cases from the knowledge base. The system calculates the Euclidean distance between the new scenario and each historical case, selects the five closest cases, and predicts the benefit scores of three possible allocation schemes under the new scenario as 7, 8, and 9 points, respectively, based on their tax benefit scores. Next, the system uses Baidu's open-source EasyDL tool to train a random forest tax risk prediction model based on 5000 extracted historical cases, achieving an accuracy of 95%. The new allocation scheme is input into the model for prediction, and the tax risk levels of the three schemes are low, medium, and high, respectively. The system automatically selects the scheme with the lowest risk, generates a tax risk warning, and sends it to the finance manager for review. After the finance manager confirms, the system enters the new scenario's allocation scheme, benefit assessment results, and risk level into the database. The system automatically analyzes 1,000 new cases each month, updates the tax benefit assessment model using the gradient boosting tree algorithm, and optimizes the random forest tax risk prediction model using grid search, continuously improving the system's assessment and prediction accuracy.

[0036] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A multi-terminal ecosystem fund sharing management system for e-commerce enterprises, characterized in that, The multi-terminal ecological fund sharing management system includes: Obtain information on the tax jurisdiction of each business entity, compare the tax policies of the different regions, and determine whether there are differences in tax policies. If the aforementioned tax policy differences exist, then the relevant transaction data and tax information on fund allocation are obtained from the financial systems of each business entity, and a unified format tax information dataset is obtained through data cleaning and integration. By using association rule mining technology, the association patterns between fund allocation and tax treatment are discovered from the tax information dataset, forming a tax association rule library for fund allocation; Based on the tax-related rule base of the aforementioned fund allocation, a fund allocation tax benefit assessment model is constructed. By inputting parameters of different fund allocation schemes, the tax impact of the schemes is predicted. For emerging businesses, the emerging businesses are grouped according to their business attributes and capital flow characteristics. Emerging business groups similar to existing businesses are identified, and the tax treatment experience of similar business groups is referenced to determine the tax risks of the emerging businesses. Establish a knowledge base for tax assessment of fund allocation, and store and manage tax policy changes, fund allocation cases, and tax benefit assessment results in a structured manner. When new business scenarios or tax policy changes occur, relevant information can be retrieved from the knowledge base to assist in the tax benefit assessment of fund allocation.

2. The multi-terminal ecological fund sharing management system according to claim 1, characterized in that, The step of obtaining information on the tax jurisdiction of each business entity, comparing the tax policies of different regions, and determining whether there are differences in tax policies includes: Obtain the registration location information of at least one business entity and determine the tax jurisdiction to which the business entity belongs; For the aforementioned tax jurisdiction, obtain the corresponding tax policy documents; Natural language processing technology is used to analyze the text of the tax policy document to extract key tax elements such as tax rate, tax base and tax payment period. The extracted key tax elements are stored in a pre-defined structured database; For the key tax elements in the structured database, cross-regional comparative analysis is performed using data comparison tools to obtain the differences in tax policies between different regions; The identified tax policy differences are quantitatively assessed, and a tax policy difference index is calculated. Based on the tax policy difference index and the financial data of the business entity, a tax risk scoring function is constructed using a logistic regression model. The relevant data of the business entity is input into the tax risk scoring function to calculate the tax risk score of the business entity; Determine whether the tax risk score exceeds a preset threshold. If so, classify the business entity as high-risk; otherwise, classify the business entity as medium-low risk. Develop corresponding tax compliance management measures and risk control recommendations for business entities with different risk levels; Generate a tax risk assessment report that includes risk rating results and management recommendations.

3. The multi-terminal ecological fund sharing management system according to claim 1, characterized in that, If the aforementioned tax policy differences exist, then transaction data and tax information related to fund allocation are obtained from the financial systems of each business entity. Through data cleaning and integration, a unified format tax information dataset is obtained, including: If it is determined that there are differences in the tax policies of the various business entities, then obtain the transaction data and tax information related to fund allocation from the financial systems of each business entity. The acquired transaction data and tax information are preprocessed, and Excel's data filtering function is used to remove null and outlier values ​​to obtain a pre-cleaned data set. Based on the preset data integration rules, SQL statements are used to integrate the cleaned data, converting data from different sources and in different formats into data tables in a unified format; Use Excel's pivot table function to extract key fields related to tax information from the integrated data table, including transaction amount, tax rate and tax type, and build a tax information feature table; Import the feature table into the R language environment and use the random forest algorithm to build a tax information classification model. The model performance was evaluated using cross-validation to ensure that the classification accuracy reached the preset threshold. The trained random forest model is used to classify the integrated data to obtain the tax attributes of each data point; Based on the classification results, the SQL GROUPBY statement is used to divide data with the same tax attributes into different data tables, forming subsets with different tax attributes. Each subset of the dataset is sampled and checked to ensure the accuracy of the classification results; If an error is found, use the SQL UPDATE statement to manually correct it; Finally, the SQL UNION ALL statement is used to merge the various subsets of data to obtain a master data table with a unified format and complete tax information. Perform data quality checks on the master data table, including verification of completeness, consistency, and accuracy; If any problems are found, return to the corresponding steps to correct them; After verification, the master data table is exported as a CSV file to be used as input data for subsequent tax processing.

4. The multi-terminal ecological fund sharing management system according to claim 1, characterized in that, The method of using association rule mining technology to discover the association patterns between fund allocation and tax treatment from the tax information dataset forms a tax association rule library for fund allocation, including: Obtain tax information datasets, including taxpayer information, transaction records, and invoice data; Use SQL to clean the data, removing duplicate records, outliers, and missing data. ETL tools are used for data integration to unify the format and encoding of data from different sources. Use Python to transform the data, employing either standardized numerical features or one-hot encoded categorical features. Exploratory data analysis is performed on the preprocessed dataset, using descriptive statistics and visualization tools to identify data distributions and potential patterns; Based on the analysis results, select an appropriate feature subset for subsequent modeling; The Apriori algorithm was used to mine association rules, with a minimum support of 0.5 and a minimum confidence of 7, to obtain an initial set of association rules between fund allocation and tax treatment. Rules are filtered using the lift metric, and rules with a lift less than 1 are removed. The remaining rules are manually reviewed, and unreasonable rules are removed by incorporating domain knowledge. The filtered rules are categorized according to tax business type and taxpayer category to construct a multi-level rule base structure; Develop a Java-based rule engine and import the rule library into it; For new tax data, a rules engine is used for matching and reasoning to output compliance assessment results and corresponding tax treatment suggestions for fund allocation behavior; After the tax officials review the application, they will send the processing results back to the system. Regularly update the dataset and re-execute the association rule mining process; Compare the differences between the old and new rule sets to identify newly emerging patterns; The rule base is dynamically updated and optimized based on practical application results; Throughout the process, data anonymization techniques are used to protect sensitive information, and access control and audit log mechanisms are employed to ensure data security.

5. The multi-terminal ecological fund sharing management system according to claim 1, characterized in that, The step involves constructing a tax benefit assessment model for fund allocation based on the tax-related rule base of the fund allocation scheme. By inputting parameters for different fund allocation schemes, the model predicts the tax impact of the schemes, including: Obtain data on the fund allocation plan from the company's financial system, including the allocation amount, allocation recipients, and allocation ratio parameters; Obtain tax rule data from the tax system, including information on tax types, tax rates, and tax base. Link and map the data of the fund allocation scheme with the data of tax rules to establish a fund allocation tax-related rule database; Preprocessing of historical funding allocation schemes and their actual tax impact data, including data cleaning and normalization; The preprocessed data was used to train decision tree and logistic regression models, which served as the basis for evaluating the tax benefits of capital allocation. Using the prediction results of the trained decision tree model and logistic regression model as input, a simple feedforward neural network is constructed. The neural network was trained using historical data to obtain a fused tax benefit assessment model for capital allocation. Input the parameters of the new cost-sharing scheme, and determine the tax rules related to the scheme through the cost-sharing tax association rule library; By inputting the parameters of the cost-sharing scheme and the associated tax rules into the fusion model, the predicted tax impact results are obtained. Calculate the overall tax impact score based on the predicted tax impact results; The scoring is calculated using a weighted average method, with the weights pre-set based on the importance of each tax impact. The obtained comprehensive tax impact score is compared with the preset tax benefit threshold to determine the tax benefit level of the fund allocation plan. Repeat the above evaluation process for multiple cost-sharing schemes, and rank the schemes according to the comprehensive tax impact score; The top three highest-rated options are selected as the recommended cost-sharing optimization schemes. Regularly update the model with new historical data, including retraining the base model and the fusion model, to adapt to the ever-changing tax environment; Compare and analyze the model's predictions with the actual tax impact to continuously optimize model performance; During the evaluation process, if any abnormal situation is encountered, such as missing data or data exceeding the expected range, the system will record the abnormality and prompt manual intervention. For the final recommended cost-sharing optimization plan, a detailed tax impact analysis report will be generated to provide a reference for corporate decision-making.

6. The multi-terminal ecological fund sharing management system according to claim 1, characterized in that, For emerging businesses, the process involves grouping them according to their business attributes and cash flow characteristics, identifying emerging business groups similar to existing businesses, and assessing the tax risks of these emerging businesses by referring to the tax treatment experience of similar business groups. Based on the business attributes and capital flow characteristics of emerging businesses, the K-means clustering algorithm is used to group emerging businesses into multiple emerging business groups. For each emerging business group; By calculating cosine similarity, the existing business group that is most similar to it is identified. Obtain tax processing experience data from similar existing business groups, including tax type identification, applicable tax rates, and input tax deduction information; Based on this empirical data, a tax risk assessment model is constructed using the random forest algorithm. The model input is business attributes and cash flow data, and the output is a tax risk score. The business attributes and cash flow data of the emerging business group are input into the tax risk assessment model to calculate the tax risk score of the emerging business group. Set a tax risk score threshold, such as 7. If the calculated score is greater than or equal to 7, it is considered high risk; if it is less than 7 but greater than or equal to 4, it is considered medium risk; and if it is less than 4, it is considered low risk. For high-risk business groups, the system automatically sends risk warning notifications to business personnel, prompting them to take tax risk prevention and control measures. The system extracts the latest operational data of emerging business groups from the business database daily, including transaction amount, transaction frequency and customer type, inputs it into the tax risk assessment model, updates the risk score and level, and realizes dynamic monitoring of tax risks. If the risk level changes, the system will generate a risk change report and push it to relevant business personnel and tax specialists.

7. The multi-terminal ecological fund sharing management system according to claim 1, characterized in that, The establishment of a knowledge base for tax assessment of fund allocation involves the structured storage and management of tax policy changes, fund allocation cases, and tax benefit assessment results. When new business scenarios or tax policy changes occur, relevant information is retrieved from this knowledge base to assist in the tax benefit assessment of fund allocation, including: Obtain the latest information on tax policy changes and compare it with policy data in the existing knowledge base for tax assessment of fund allocation; If new policies or changes are discovered, update the corresponding entries in the knowledge base; At the same time, we collect new cases of fund sharing and extract key information from these cases, including the sharing method, the amount involved, and the relevant parties. This information is entered into the knowledge base, and the case is initially evaluated based on preset tax benefit indicators, including tax burden rate and cash flow impact. Use a relational database to store updated tax policies, cost-sharing cases, and tax benefit assessment results; Create indexes to optimize query efficiency; When a new business scenario emerges, relevant tax policies and historical allocation cases are retrieved from the database based on the scenario characteristics; the scenario characteristics include the transaction type and the amount involved. Using the retrieved information, the K-nearest neighbor algorithm is employed to evaluate the tax benefits of the capital allocation scheme in the new business scenario; The specific steps include: calculating the similarity between the new scenario and historical cases, selecting the K most similar cases, and predicting the potential benefits of the new solution based on the tax benefit results of these cases; Next, the random forest algorithm is used to determine the tax risks of the cost-sharing scheme; First, the model is trained based on historical cases. The input features include the sharing method, the amount involved, and the type of entity. The output is the risk level. The new plan is then input into the model to obtain the risk prediction results; If the prediction result exceeds the preset threshold, the system will automatically issue a risk warning; Finally, the capital sharing plan, tax benefit assessment results, and risk assessment results under the new business scenario will be entered into the database; Regularly analyze new data and update the parameters of assessment and risk judgment models to adapt to policy changes and new business scenarios; In this way, a continuously optimizing closed-loop system is formed.

Images