Knowledge collaboration optimization method and system for cross-enterprise financial tax robot

Abstract

This application provides a knowledge collaborative optimization method and system for cross-enterprise financial and tax robots, relating to the field of knowledge graph technology. The method includes: collecting multi-source financial and tax datasets from multiple target enterprises, extracting and modeling knowledge to generate an enterprise financial and tax knowledge graph model, and loading it into the cross-enterprise financial and tax robot; constructing multi-target enterprise profiles and crawling updated financial and tax regulations and guidelines in real time; updating the enterprise financial and tax knowledge graph model through reinforcement learning based on the updated regulations, guidelines, and target enterprise profiles; acquiring cross-enterprise invoice information to be processed, and using the cross-enterprise financial and tax robot to collaboratively optimize the processing of this information based on the updated model of the enterprise financial and tax knowledge graph. This application solves the technical problems of low collaborative processing efficiency and poor accuracy caused by the dispersion of financial and tax knowledge and heterogeneous rules among enterprises in existing technologies. By constructing a cross-enterprise financial and tax knowledge graph and dynamically updating and optimizing it, the efficiency of knowledge collaborative processing is improved.

Description

Technical Field

[0001] This application relates to the field of knowledge graph technology, specifically to a knowledge collaboration optimization method and system for cross-enterprise financial and tax robots. Background Technology

[0002] In current cross-enterprise financial and tax collaborative processing scenarios, most enterprises still use their own independent financial and tax management systems. Due to the dispersed financial and tax knowledge and the use of different financial and tax rules by each enterprise, the financial and tax data across entities is highly fragmented and heterogeneous. The lack of unified financial and tax rules and knowledge bases results in low efficiency and poor accuracy in cross-enterprise collaborative processing. The differences in financial and tax rules used by different enterprises, coupled with the lack of unified standards, lead to insufficient information integration and deviations in the understanding and application of rules, easily causing accounting errors and compliance risks, thus affecting the overall effectiveness of financial and tax management. Current financial and tax management systems typically only operate within a single enterprise, with limited cross-enterprise collaborative processing capabilities. Existing methods rely on manual intervention and updates to address changes in financial and tax rules, which is time-consuming, error-prone, and unable to adapt to dynamic changes in financial and tax rules in a timely manner, resulting in low efficiency and poor accuracy when handling multi-enterprise collaborative tasks.

[0003] In summary, existing technologies suffer from low efficiency and poor accuracy in collaborative processing of cross-enterprise financial and tax tasks due to the dispersed financial and tax knowledge and heterogeneous rules among enterprises. Summary of the Invention

[0004] The purpose of this application is to provide a knowledge collaboration optimization method and system for cross-enterprise financial and tax robots, in order to solve the technical problems of low collaborative processing efficiency and poor accuracy when handling cross-enterprise financial and tax tasks due to the dispersion of financial and tax knowledge and heterogeneous rules among enterprises.

[0005] To achieve the above objectives, this application provides a knowledge collaborative optimization method and system for cross-enterprise financial and tax robots.

[0006] Firstly, this application provides a knowledge collaborative optimization method for cross-enterprise financial and tax robots. This method is implemented through a knowledge collaborative optimization system for cross-enterprise financial and tax robots. The method includes: collecting multi-source financial and tax datasets from multiple target enterprises; performing knowledge extraction and modeling on the multi-source datasets to generate an enterprise financial and tax knowledge graph model; and loading the enterprise financial and tax knowledge graph model into the cross-enterprise financial and tax robot; performing profile analysis based on the attribute data of the multiple target enterprises to construct multi-target enterprise profiles, while simultaneously crawling updated financial and tax regulations and guidelines in real time; updating the enterprise financial and tax knowledge graph model through reinforcement learning based on the updated regulations and guidelines and the target enterprise profiles to generate an updated enterprise financial and tax knowledge graph model; acquiring cross-enterprise invoice information to be processed; and using the cross-enterprise financial and tax robot to collaboratively optimize the cross-enterprise invoice information based on the updated enterprise financial and tax knowledge graph model.

[0007] Optionally, a data cleaning process is constructed, which includes removing duplicate data, filling missing values, handling outliers, and standardizing data formats. The multi-source financial and tax dataset is cleaned according to the data cleaning process to obtain a usable financial and tax dataset. The usable financial and tax dataset is then subjected to text recognition and extraction and data field parsing using OCR and NLP to obtain a key financial and tax dataset. Based on the key financial and tax dataset, knowledge extraction and modeling optimization are performed to generate an enterprise financial and tax knowledge graph model.

[0008] Optionally, based on the enterprise's financial and tax processing needs, knowledge entity types are defined, including enterprises, accounting subjects, tax items, regulations and standards, and transaction events; based on the knowledge entity types, entity relationship types are defined, including financial relationships, tax relationships, and business relationships; knowledge is extracted from the key financial and tax dataset according to the knowledge entity types and entity relationship types to obtain an enterprise financial and tax knowledge set; graph structure information is designed, and the enterprise financial and tax knowledge set is fused, modeled, and optimized based on the graph structure information to generate an enterprise financial and tax knowledge graph model.

[0009] Optionally, knowledge fusion is performed on the enterprise financial and tax knowledge set based on the graph structure information to construct an initial financial and tax knowledge graph model; conflict detection is performed on the initial financial and tax knowledge graph model to obtain financial and tax conflict knowledge, which includes data conflict and semantic conflict; reasoning enhancement and optimization are performed on the initial financial and tax knowledge graph model based on the financial and tax conflict knowledge to generate an enterprise financial and tax knowledge graph model.

[0010] Optionally, an enterprise profile tagging system is designed, which includes basic attribute tags, financial attribute tags, tax attribute tags, and business attribute tags; the attribute data of the multi-target enterprises are tagged according to the enterprise profile tagging system to obtain multi-enterprise profile tag data; based on the multi-enterprise profile tag data, profile modeling analysis and verification correction are performed to construct multi-target enterprise profiles.

[0011] Optionally, the tax and financial regulations and guidelines are standardized and semantically parsed to obtain tax and financial regulations and rules information; the target enterprise profile is matched with the tax and financial regulations and rules information to obtain an enterprise-related regulations and rules set; based on the enterprise-related regulations and rules set, the enterprise tax and financial knowledge graph model is updated by knowledge impact quantification and reinforcement learning to generate an enterprise tax and financial knowledge graph update model.

[0012] Optionally, the cross-enterprise financial and tax robot is decomposed into collaborative tasks to obtain a set of enterprise financial and tax collaborative tasks, which includes invoice processing, knowledge query, rule generation, and collaborative accounting. The cross-enterprise invoice information to be processed is then collaboratively optimized according to the enterprise financial and tax knowledge graph update model based on the enterprise financial and tax collaborative task set.

[0013] Optionally, the cross-enterprise invoice information to be processed is identified and verified according to the enterprise financial and tax collaborative task set to obtain the cross-enterprise invoice information; knowledge query and rule generation are performed on the cross-enterprise invoice information based on the enterprise financial and tax knowledge graph update model to determine the enterprise related accounting rule set; the cross-enterprise invoice information is processed for collaborative accounting optimization using the enterprise related accounting rule set to obtain the cross-enterprise financial and tax collaborative accounting result.

[0014] Optionally, based on the enterprise financial and tax knowledge graph update model, knowledge queries are performed on the cross-enterprise invoice information to obtain a cross-enterprise related financial and tax knowledge set; dynamic rule derivation is performed on the cross-enterprise related financial and tax knowledge set to determine the enterprise related accounting rule set.

[0015] Secondly, this application also provides a knowledge collaborative optimization system for cross-enterprise financial and tax robots, used to execute the knowledge collaborative optimization method for cross-enterprise financial and tax robots as described in the first aspect. The knowledge collaborative optimization system for cross-enterprise financial and tax robots includes: a knowledge extraction and modeling module, used to collect multi-source financial and tax datasets from multiple target enterprises, perform knowledge extraction and modeling on the multi-source financial and tax datasets, generate an enterprise financial and tax knowledge graph model, and load the enterprise financial and tax knowledge graph model into the cross-enterprise financial and tax robot; a profile analysis module, used to perform profile analysis based on the attribute data of the multiple target enterprises, construct multi-target enterprise profiles, and simultaneously crawl updated financial and tax regulations and guidelines in real time; a reinforcement learning update module, used to perform reinforcement learning updates on the enterprise financial and tax knowledge graph model based on the updated financial and tax regulations and guidelines and the target enterprise profiles, generating an updated enterprise financial and tax knowledge graph model; and a collaborative optimization processing module, used to acquire cross-enterprise invoice information to be processed, and to perform collaborative optimization processing on the cross-enterprise invoice information to be processed by the cross-enterprise financial and tax robot based on the updated enterprise financial and tax knowledge graph model.

[0016] One or more technical solutions provided in this application have at least the following technical effects or advantages: By collecting multi-source financial and tax datasets from multiple target enterprises, knowledge extraction and modeling are performed on the multi-source financial and tax datasets to generate an enterprise financial and tax knowledge graph model, which is then loaded into a cross-enterprise financial and tax robot. Based on the attribute data of the multiple target enterprises, profiling analysis is performed to construct multi-target enterprise profiles, while simultaneously crawling updated financial and tax regulations and guidelines in real time. Based on the updated financial and tax regulations and guidelines and the target enterprise profiles, reinforcement learning is used to update the enterprise financial and tax knowledge graph model, generating an updated enterprise financial and tax knowledge graph model. Cross-enterprise invoice information to be processed is acquired, and the cross-enterprise financial and tax robot performs collaborative optimization processing on the cross-enterprise invoice information based on the updated enterprise financial and tax knowledge graph model. In other words, by collecting multi-source financial and tax data, a corporate financial and tax knowledge graph model is constructed. Based on corporate attribute data, a tag system is designed, tag values ​​are calculated, and model verification is performed to form a multi-objective corporate profile. Updated financial and tax regulations and guidelines are crawled, and the updated regulations are associated and matched with the corporate profile to quantify the impact of knowledge. Reinforcement learning is used to incrementally update the knowledge graph. After receiving the invoices to be processed, the task is broken down, and knowledge query and rule generation are performed based on the updated knowledge graph. Finally, collaborative optimization processing is completed to improve the efficiency and accuracy of cross-enterprise financial and tax collaborative processing.

[0017] The above description is merely an overview of the technical solution of this application. To better understand the technical means of this application and to facilitate its implementation according to the description, and to make the above and other objects, features, and advantages of this application more apparent, specific embodiments of this application are described below. It should be understood that the content described in this section is not intended to identify key or important features of the embodiments of this application, nor is it intended to limit the scope of this application. Other features of this application will become readily apparent through the following description. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely exemplary. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0019] Figure 1 This is a flowchart illustrating the knowledge collaboration optimization method for cross-enterprise financial and tax robots proposed in this application.

[0020] Figure 2 This is a schematic diagram of the knowledge collaborative optimization system for cross-enterprise financial and tax robots, as described in this application.

[0021] Figure labeling: Knowledge extraction and modeling module 11, profile analysis module 12, reinforcement learning update module 13, collaborative optimization processing module 14. Detailed Implementation

[0022] This application provides a knowledge collaboration optimization method and system for cross-enterprise financial and tax robots, solving the technical problems of low efficiency and poor accuracy in collaborative processing of cross-enterprise financial and tax tasks caused by the dispersion of financial and tax knowledge and heterogeneous rules among enterprises. By collecting multi-source financial and tax data, a corporate financial and tax knowledge graph model is constructed. A tag system is designed based on corporate attribute data, tag values ​​are calculated and modeled for verification, forming a multi-objective corporate profile. Updated financial and tax regulations and guidelines are crawled, and the updated regulations are associated and matched with the corporate profiles to quantify the impact of knowledge. Reinforcement learning is used to incrementally update the knowledge graph. Upon receiving pending invoices, the task is broken down, and knowledge queries and rule generation are performed based on the updated knowledge graph, ultimately completing the collaborative optimization process and improving the efficiency and accuracy of cross-enterprise financial and tax collaborative processing.

[0023] The technical solutions of this application will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. It should be understood that this application is not limited to the exemplary embodiments described herein. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application. It should also be noted that, for ease of description, only the parts related to this application are shown in the accompanying drawings, not all of them.

[0024] Example 1, please refer to the appendix. Figure 1 This application provides a knowledge collaborative optimization method for cross-enterprise financial and tax robots. The method is applied to a knowledge collaborative optimization system for cross-enterprise financial and tax robots, and specifically includes the following steps: Collect multi-source financial and tax datasets from multiple target enterprises, perform knowledge extraction and modeling on the multi-source financial and tax datasets to generate an enterprise financial and tax knowledge graph model, and load the enterprise financial and tax knowledge graph model into a cross-enterprise financial and tax robot.

[0025] Furthermore, this application also includes the following steps: constructing a data cleaning process, which includes removing duplicate data, filling missing values, handling outliers, and unifying data formats; performing data cleaning processing on the multi-source financial and tax dataset according to the data cleaning process to obtain a usable financial and tax dataset; performing text recognition extraction and data field parsing on the usable financial and tax dataset using OCR and NLP to obtain a key financial and tax dataset; and performing knowledge extraction and modeling optimization based on the key financial and tax dataset to generate an enterprise financial and tax knowledge graph model.

[0026] Furthermore, this application also includes the following steps: defining knowledge entity types according to the enterprise's financial and tax processing needs, wherein the knowledge entity types include enterprises, accounting subjects, tax items, regulations and standards, and transaction events; defining entity relationship types based on the knowledge entity types, wherein the entity relationship types include financial relationships, tax relationships, and business relationships; extracting knowledge from the key financial and tax dataset according to the knowledge entity types and the entity relationship types to obtain an enterprise financial and tax knowledge set; designing graph structure information, and performing fusion modeling and optimization on the enterprise financial and tax knowledge set based on the graph structure information to generate an enterprise financial and tax knowledge graph model.

[0027] Furthermore, this application also includes the following steps: performing knowledge fusion on the enterprise financial and tax knowledge set based on the graph structure information to construct an initial financial and tax knowledge graph model; performing conflict detection on the initial financial and tax knowledge graph model to obtain financial and tax conflict knowledge, wherein the financial and tax conflict knowledge includes data conflict and semantic conflict; and performing reasoning enhancement optimization on the initial financial and tax knowledge graph model based on the financial and tax conflict knowledge to generate an enterprise financial and tax knowledge graph model.

[0028] Specifically, raw financial and tax data is collected from various heterogeneous data sources across multiple enterprises. Interactions are established with each target enterprise to clarify the form, storage location, and access permissions of their financial and tax data. For enterprises using accounting software, structured data such as general ledgers, vouchers, and charts of accounts are directly read via database interfaces or API calls. For data stored in spreadsheet format, files of specified formats are imported in batches. For paper invoices or PDF electronic invoices, original files or scanned copies are collected. A complete data catalog must be established during the collection process, recording the source enterprise, data table name, field meaning, and other information for each data entry, forming the original multi-source financial and tax dataset. The multi-source financial and tax dataset is a collection of raw financial and tax-related data collected from different enterprises and systems. Its sources include enterprise accounting software, tax filing systems, general ledger modules and purchasing and sales modules in ERP systems, as well as various original invoices. These data may differ in format, structure, and storage method, including structured data, semi-structured data, and unstructured data.

[0029] A set of data cleaning rules is pre-designed, including deduplication, missing value imputation, outlier handling, and data format standardization, with the aim of improving data quality. For deduplication, duplicate records are identified using key fields such as invoice code + invoice number + amount combination, retaining only the first occurrence or most recent record. For missing value imputation, for missing company names, the missing name is filled in from the company's basic information table based on the taxpayer identification number; for missing amount fields, if it cannot be derived from other fields, it is marked as pending verification and an alert is generated. For outlier handling, rules such as the amount must be greater than zero and the date must be within a reasonable business period can be set; records violating these rules are intercepted and transferred to manual review. Regarding data format standardization, standard format specifications need to be established, such as converting all dates to YYYY-MM-DD format, standardizing all amount fields to two decimal places in RMB, and standardizing all account codes to the length of the Ministry of Finance's standard coding system.

[0030] Following the data cleaning process, data cleaning tools were used to automate the processing of the collected raw multi-source financial and tax datasets. During processing, a cleaning log was recorded for each step, including the number of duplicates removed, missing value imputation, outlier handling methods, and format conversion details. After cleaning, the resulting data underwent quality verification to ensure all rules were correctly applied, ultimately yielding a high-quality, standardized, and usable financial and tax dataset. The data records in this dataset possessed a consistent format and high integrity, but the unstructured text portions had not yet been parsed into structured fields.

[0031] For image files in the available financial and tax dataset, such as scanned invoices, the first step is to use an OCR engine for text recognition, converting the printed text in the image into editable text strings. For the recognized text and existing text fields in the dataset, such as summaries and notes, NLP techniques are further applied for information extraction. Specifically, a named entity recognition model based on the financial and tax domain can be built, capable of automatically identifying and extracting preset key fields such as invoice code, invoice number, invoice date, buyer's name, seller's name, amount in lowercase, and tax amount from the text. Simultaneously, existing fields in the structured data, such as accounting subject codes and debit amounts, can be directly mapped to the key dataset. Through this step, financial and tax information originally scattered across images, text, and structured records is uniformly extracted and integrated into a highly structured financial and tax key dataset. Each record in this dataset contains clearly defined fields and corresponding specific values, which can be directly used for subsequent knowledge graph construction. OCR, or Optical Character Recognition, is a technology used to convert text content in images or scanned documents into editable text data. It is mainly used to process paper or image-format financial and tax documents such as invoices and contracts. NLP, or Natural Language Processing, is a technology used to understand and process human language text. It performs semantic understanding, key information extraction (such as invoice number, amount, tax, and buyer / seller names), and field parsing on text recognized by OCR or existing text data.

[0032] The key financial and tax dataset is a collection of data extracted from available financial and tax datasets using OCR and NLP technologies. It consists of core fields directly related to financial and tax processing, including company name, taxpayer identification number, invoice code, invoice number, invoice date, amount, tax amount, tax rate, accounting subject, and counterparty name.

[0033] Based on the enterprise's financial and tax processing needs, the types of knowledge entities that need to be represented in the knowledge graph are clearly defined, namely, the basic building blocks of the knowledge graph, including enterprises, accounting subjects, tax items, regulations and standards, and transaction events. The enterprise's financial and tax processing needs refer to the information processing requirements of its daily operations, including financial bookkeeping, tax declaration, cost accounting, and compliance checks. For example, it needs to accurately record the accounting subject, applicable tax item, and applicable rules and regulations for each transaction.

[0034] Based on knowledge entity types, entity relationship types are defined to describe the semantic connections between different entities, including financial relationships, tax relationships, and business relationships. Financial relationships involve accounting relationships, such as enterprise-setting-accounting subject, transaction event-recording-accounting subject; tax relationships involve tax processing relationships, such as transaction event-applicable-tax item, enterprise-enjoying-tax incentives; business relationships involve business transactions between enterprises, such as enterprise-purchasing-enterprise, transaction event-included-product.

[0035] Based on knowledge entity types and entity relationship types, key financial and tax datasets are processed to automatically identify instances that conform to preset entity and relationship types, and transform them into structured knowledge representations to obtain the enterprise financial and tax knowledge set. For example, from a voucher record of Company A, "Borrowing 850.00 for Management Expenses - Travel Expenses" and "Crediting 850.00 for Bank Deposits," with the summary "Payment for User A's Business Trip to Beijing," the transaction event entity "Payment for Li Hua's Business Trip to Beijing" is extracted, along with its financial relationship with the "Management Expenses - Travel Expenses" entity and the "Bank Deposits" entity. From an invoice record of Company B, the transaction event entity "Purchase of Raw Materials" is extracted, along with two enterprise entities: the buyer, Company B, and the seller, Supplier E. A purchase business relationship is established, and the applicable tax relationship between this transaction event and the 13% VAT on raw materials entity is also established. All extracted entity and relationship instances constitute the enterprise financial and tax knowledge set. The enterprise financial and tax knowledge set contains hundreds to thousands of entity nodes and several times that number of relationship edges.

[0036] Design the graph structure information and define the organization form of the knowledge graph, including node attributes, edge attributes, graph hierarchical structure, indexing method, etc. For example, determine which attribute fields each entity node should contain (such as the taxpayer identification number and registered capital of the enterprise), and which attributes each relation edge should contain, such as transaction amount and occurrence time.

[0037] Based on graph structure information, knowledge fusion is performed on the enterprise financial and tax knowledge set. The extracted knowledge set is integrated according to the designed graph structure, solving problems such as entity alignment and attribute merging, forming a unified graph representation, resulting in an initial financial and tax knowledge graph model. This model contains basic entities and relationships, but may still contain logical contradictions or errors. For example, in the knowledge set, Company B's invoice records show a buyer, B Trading Co., Ltd., while Company A's general ledger summary mentions receiving payment from Company B. Through string similarity matching and taxpayer identification number comparison, it is determined that B Trading Co., Ltd. and Company B refer to the same enterprise entity. Therefore, the two nodes are merged, and their respective relationships are unified onto the merged node. At the same time, attributes of the same entity from different sources are merged, such as the taxpayer identification number of Company B obtained from the invoice, which is added to the enterprise node. After fusion, all entities and relationships are loaded into the graph database, forming the initial financial and tax knowledge graph model, which already possesses the basic network structure of cross-enterprise financial and tax knowledge.

[0038] Conflict detection, or consistency check, is performed on the initial financial and tax knowledge graph model to identify contradictions and errors, including data conflicts and semantic conflicts. Data conflicts refer to inconsistent attribute values ​​for the same entity, such as the registered capital of the same company being 10 million yuan and 20 million yuan in different data sources; or the amount of the same transaction event not matching the invoice amount. Semantic conflicts refer to relationship definitions that violate business logic, such as in a company-purchase-company relationship, the purchaser and seller are the same company, which is equivalent to self-buying and selling, which does not conform to normal business logic; or a transaction event simultaneously applying two mutually exclusive tax categories.

[0039] For detected conflicts, specifically tax-related knowledge conflicts, the initial tax knowledge graph model is enhanced and optimized through reasoning. Conflicting knowledge is corrected, supplemented, or deleted, and new knowledge is derived using reasoning rules, thereby improving the quality of the initial tax knowledge graph model and obtaining the enterprise tax knowledge graph model. For example, for data conflicts, such as inconsistent amounts, a rule can be set: the amount of the invoice node takes precedence, because the invoice is the original voucher, thus correcting the amount attribute of the transaction event. For semantic conflicts, such as the buyer equaling the seller, further reasoning is performed to query whether the transaction has a contract number. If the contract shows it as an internal transfer within the group rather than an external purchase, the business type label may be incorrect, and the relationship type is corrected from purchase to internal transfer; if it cannot be reasonably explained, the relationship is marked for manual verification.

[0040] The enterprise financial and tax knowledge graph model uses entities as nodes and relationships as edges to connect financial and tax information scattered across multiple sources of data, forming a machine-understandable and reasonable knowledge representation. The enterprise financial and tax knowledge graph model includes basic information about multiple target enterprises, accounting subject systems, applicable tax items and rates, relevant financial and tax rules and regulations, as well as various transaction events between enterprises and their financial and tax processing rules.

[0041] The constructed knowledge graph model is exported from the development environment and deployed to the graph database server in the production environment. During startup or operation, the cross-enterprise financial and tax robot establishes a connection with the graph database through the configured database connection interface, loading the entire enterprise financial and tax knowledge graph model into its memory or cache, or accessing it in real time via API. After loading, the cross-enterprise financial and tax robot possesses all the financial and tax knowledge contained in the knowledge graph, and can understand the subject system, transaction relationships, applicable tax rates, compliance requirements, etc. of various enterprises. When a user submits a cross-enterprise financial and tax processing task, the cross-enterprise financial and tax robot can perform knowledge query, path reasoning, and rule matching based on the loaded knowledge graph, thereby generating a processing solution that conforms to the actual situation and the latest rules of each enterprise.

[0042] Based on the attribute data of the multi-target enterprises, a profile analysis is performed to construct a multi-target enterprise profile, while the updated financial and tax regulations and guidelines are crawled in real time.

[0043] Furthermore, this application also includes the following steps: designing an enterprise profile tagging system, the enterprise profile tagging system including basic attribute tags, financial attribute tags, tax attribute tags and business attribute tags; calculating tags on the attribute data of the multi-target enterprises according to the enterprise profile tagging system to obtain multi-enterprise profile tag data; and constructing multi-target enterprise profiles by performing profile modeling analysis and verification correction based on the multi-enterprise profile tag data.

[0044] Specifically, a corporate profile tagging system is designed to comprehensively describe various characteristics of a company, divided into four main categories: basic attribute tags, financial attribute tags, tax attribute tags, and business attribute tags. Basic attribute tags reflect the company's basic information and static characteristics, such as company name, unified social credit code, registered address, establishment date, registered capital, industry, company type, and number of employees. Financial attribute tags reflect the company's financial status and operating results, typically calculated based on financial statement data, such as total assets, net assets, operating revenue, net profit, debt-to-equity ratio, current ratio, and return on equity. Tax attribute tags reflect the company's tax registration, tax payment behavior, and tax characteristics, such as taxpayer type, tax credit rating (A / B / C / D), main tax types, whether tax incentives are enjoyed, and average tax burden rate over the past three years. Business attribute tags reflect the company's business model, business scope, and transaction characteristics, such as main business type, supply chain role, main customer groups, transaction frequency, average transaction amount, and main product / service categories. Basic attribute tags are mainly derived from business registration information and basic company files, used to identify the company's basic identity and industry affiliation. Financial attribute tags are derived from a company's balance sheet, profit and loss statement, and other financial statements, used to assess the company's operating scale and solvency. Tax attribute tags are derived from tax registration information and tax declaration records, used to understand the company's tax status, credit level, and preferential rules enjoyed. Business attribute tags are derived from business data such as sales contracts, purchase orders, and invoice details, used to depict the company's business model and transaction characteristics. Each tag must have a clear definition, data type, value range, and calculation rules. For example, the debt-to-equity ratio is calculated as total liabilities at the end of the period / total assets at the end of the period × 100%, with a value range of 0% to 100%. For example, basic attribute tags include company ID, company name, years of establishment, registered capital, industry, number of employees, and registered region; financial attribute tags include annual operating revenue, net profit, debt-to-equity ratio, current ratio, and return on equity; tax attribute tags include taxpayer type, tax credit rating, main tax types, high-tech enterprise status, R&D expense deduction eligibility, and annual actual tax burden rate; business attribute tags include main business type, supply chain role, main customer types, number of annual transactions, average transaction amount per transaction, and main product / service categories.

[0045] Following the designed enterprise profile tagging system, tag calculations are performed on the attribute data of multiple target enterprises. Relevant attribute data for each enterprise is extracted, potentially supplementing missing business registration information through third-party data services. Automated scripts or data processing tools are written to perform calculations for each enterprise and each tag individually. The calculation process must consider the timeliness of the data; for example, financial tags typically use data from the most recent complete fiscal year, while tax tags require real-time updates to the latest tax credit rating. After calculation, a set of tag values ​​is obtained for each enterprise. The multi-enterprise profile tagging data is a dataset consisting of a set of tag values ​​for each target enterprise after all tag calculations have been completed.

[0046] This process involves profiling and modeling multi-enterprise profile data using a tag system to construct a multi-dimensional enterprise profile model. Clustering algorithms are employed to categorize enterprises with similar characteristics, identifying typical enterprise types. Simultaneously, weights are assigned to different tags based on expert experience to calculate a comprehensive score for each enterprise, such as assessing its tax risk level or credit rating. After modeling, the generated profiles are validated. The profile results are confirmed with the enterprise's financial manager, comparing the profile tags with the actual business situation; or cross-validation is performed using data of the same type from different sources, such as comparing the main customer group tags calculated based on invoice data with customer lists provided by the business department. If discrepancies are found, the cause must be traced, which may include unreasonable tag calculation rules, incorrect original data, or a need to adjust the profile model. Corrections are made promptly, and recalculations or modeling are performed. Ultimately, an accurate and reliable multi-target enterprise profile is generated for each target enterprise. The multi-target enterprise profile is a multi-dimensional feature description generated for each target enterprise, presented in the form of a tag set. For example, Company M was established on March 15, 2010, with a registered capital of 50 million yuan, industry code C34, 320 employees, operating revenue of 250 million yuan, net profit of 18.75 million yuan, total assets of 210 million yuan, total liabilities of 123.06 million yuan, current ratio of 1.8, return on net assets of 21.5%, is a general taxpayer, has an annual tax credit rating of A, its main taxes are value-added tax and corporate income tax, it has a valid high-tech enterprise qualification, enjoys additional deduction for R&D expenses, and its actual tax burden rate (corporate income tax / operating revenue) is approximately 2.5%. Company M is labeled as a mature manufacturing high-tech enterprise.

[0047] Simultaneously, a web crawler program is configured to perform scheduled tasks to access designated financial and tax information source websites, automatically fetching the latest regulations, guidelines, interpretations, and announcements, and performing preliminary analysis to extract key information such as titles, document numbers, issuing authorities, release dates, and text. To ensure the completeness and accuracy of the information, the crawled text also undergoes format cleaning and deduplication. The structured regulations data is then stored in a dedicated regulations database, marked with the retrieval time for later use. Updated financial and tax regulations and guidelines are the latest normative documents issued by the financial and tax departments, including adjustments to tax rules, changes in accounting standards, and updates to tax declaration requirements. These are automatically obtained from official websites, announcements, and other channels using real-time crawling technology.

[0048] By designing and implementing a corporate profiling and tagging system, a comprehensive description of a company's financial and tax status can be achieved, supporting cross-enterprise financial and tax management, optimization, and risk analysis. Through corporate profiling modeling and analysis, a company's financial and tax potential, risk points, and optimization space can be identified. Real-time crawling of updated financial and tax regulations ensures that the company's financial and tax data always meets the latest requirements, reducing compliance risks.

[0049] Based on the aforementioned tax and financial regulations and guidelines and the target enterprise profile, the enterprise tax and financial knowledge graph model is updated using reinforcement learning to generate an updated enterprise tax and financial knowledge graph model.

[0050] Furthermore, this application also includes the following steps: standardizing and semantically parsing the aforementioned tax and finance update regulations and guidelines to obtain tax and finance update regulations and rules information; associating and matching the target enterprise profile with the tax and finance update regulations and rules information to obtain an enterprise-related regulations and rules set; and updating the enterprise tax and finance knowledge graph model based on the enterprise-related regulations and rules set through knowledge impact quantification and reinforcement learning to generate an enterprise tax and finance knowledge graph update model.

[0051] Specifically, the updated tax and financial regulations and guidelines crawled in real time undergo standardization and semantic parsing. Upon receiving the raw text of the regulations from the crawler, text cleaning is performed to remove irrelevant content such as webpage tags, headers, footers, and attachment descriptions, and to standardize character encoding and text format. Natural language processing (NLP) technology is then used to perform syntactic analysis, named entity recognition, and key information extraction on the cleaned text. This process parses key information from the tax and financial regulations and guidelines, identifying structured fields such as the regulation name, document number, issuing authority, issuance date, effective date, applicable industry scope, applicable enterprise type, and specific clause content (e.g., tax rate adjustments, deduction ratio changes, and changes in filing deadlines). All extracted rule information is organized according to a pre-set rule template to form a tax and financial regulations rule information database. Each rule contains a clear condition-action pair. The tax and financial regulations rule information is the information within the standardized and semantically parsed tax and financial regulations and guidelines, including the applicable conditions (e.g., applicable industry, enterprise type), effective date, and specific content.

[0052] The parsed rule information is compared with the tags in the enterprise profile to filter out enterprises affected by the rule, thus generating a corresponding enterprise-related rule set for each enterprise—that is, a collection of all new financial and tax rules related to that enterprise. Each rule in the financial and tax update rule information is traversed, and its applicable conditions are read. These conditions may include industry code ranges, enterprise size thresholds, and whether specific qualifications are required. These conditions are then converted into query statements, and the tag data of all target enterprises is retrieved in the enterprise profile database. For each rule, a list corresponding to the rule and enterprise ID is generated, recording which enterprises the rule affects. Conversely, for each enterprise, all matched rules are summarized and sorted according to the rule's effective date and priority to form the enterprise's related rule set. The enterprise-related rule set is a specific set of tax rules that matches the enterprise's financial and tax status, obtained by matching the enterprise's profile data with the rule information of the financial and tax update regulations.

[0053] For each rule in the enterprise-related regulations rule set, an impact quantification analysis is performed. Graph query language is used to locate entities and relationships related to the rule within the existing knowledge graph. For example, if a rule involves R&D expense deduction, all transaction event nodes associated with the R&D expense deduction regulations and guidelines node, as well as the enterprise nodes associated with these nodes, are queried. The number of nodes and relationships requiring modification is counted, and the scope of the rule's impact is assessed. The importance weight of the rule is calculated based on factors such as its effective date and the number of affected enterprises, serving as the basis for reward signals in reinforcement learning updates.

[0054] A reinforcement learning mechanism is employed to dynamically adjust the knowledge graph based on the impact quantification results and rule importance, ultimately generating an updated enterprise financial and tax knowledge graph model—an optimized version incorporating the latest regulations. For clearly added rules, corresponding regulation / guideline nodes are added to the knowledge graph, and their correct attributes are set. For modifications to rule content, such as tax rate changes, the corresponding old regulation node is located, its attributes are updated, or it is marked as invalid, while a new version of the regulation node is added. For expansions of the applicable objects of rules, new applicable relationships are added between relevant enterprise nodes and regulation nodes. For the repeal of old rules, the corresponding relationships and nodes are deleted or invalidated. Consistency of knowledge must be maintained during the update process. For example, for the same entity affected by multiple rules simultaneously, conflict resolution is performed according to rule priority or effective time. After a series of update operations, the updated enterprise financial and tax knowledge graph model is output, fully integrating the latest financial and tax regulations and matching the personalized profile characteristics of each enterprise.

[0055] By matching updated tax regulations with enterprise profile data, we ensure that enterprises always comply with the latest tax rules, reducing tax risks. Through knowledge-driven quantification and reinforcement learning updates, we optimize enterprise tax management processes and adapt to rule changes in real time. Automated rule matching and reinforcement learning updates significantly improve the efficiency of cross-enterprise tax management, reduce manual intervention, and enhance decision-making quality.

[0056] The system acquires cross-enterprise invoice information to be processed and then uses the cross-enterprise financial and tax robot to perform collaborative optimization processing on the cross-enterprise invoice information based on the enterprise financial and tax knowledge graph update model.

[0057] Furthermore, this application also includes the following steps: decomposing the cross-enterprise financial and tax robot into a collaborative task set to obtain a set of enterprise financial and tax collaborative tasks, which includes invoice processing, knowledge query, rule generation, and collaborative accounting; and performing collaborative optimization processing on the cross-enterprise invoice information to be processed based on the enterprise financial and tax knowledge graph update model according to the set of enterprise financial and tax collaborative tasks.

[0058] Furthermore, this application also includes the following steps: identifying and verifying the cross-enterprise invoice information to be processed according to the enterprise financial and tax collaborative task set to obtain cross-enterprise invoice information; performing knowledge query and rule generation on the cross-enterprise invoice information based on the enterprise financial and tax knowledge graph update model to determine the enterprise related accounting rule set; and using the enterprise related accounting rule set to perform collaborative accounting optimization processing on the cross-enterprise invoice information to obtain cross-enterprise financial and tax collaborative accounting results.

[0059] Furthermore, this application also includes the following steps: performing knowledge query on the cross-enterprise invoice information based on the enterprise financial and tax knowledge graph update model to obtain a cross-enterprise related financial and tax knowledge set; performing dynamic rule derivation to generate the cross-enterprise related financial and tax knowledge set to determine the enterprise related accounting rule set.

[0060] Specifically, a cross-enterprise financial and tax robot is an intelligent software system deployed in the cloud or on-premises within an enterprise. It incorporates an enterprise financial and tax knowledge graph update model and is capable of handling financial and tax tasks involving multiple enterprises. It can receive invoice information input by users, automatically execute a series of operations, and ultimately output collaborative accounting results. The overall processing objective of the cross-enterprise financial and tax robot is decomposed according to functional logic into a series of independently executable, sequentially connected sub-tasks, resulting in a set of collaborative enterprise financial and tax tasks, including invoice processing, knowledge retrieval, rule generation, and collaborative accounting. The bill processing stage involves identifying, verifying, and formatting bills to be processed, extracting key information such as bill type, amount, date, parties involved in the transaction, and tax amount. Knowledge retrieval is based on an updated enterprise financial and tax knowledge graph model, retrieving relevant financial and tax knowledge based on bill information, such as the entities involved (enterprises, accounts, tax items, regulations) and their relationships. Rule generation, based on the retrieved knowledge and the specific circumstances of the current bill, dynamically derives accounting rules applicable to the transaction, such as the applicable accounting subjects, tax rates, and rule clauses to be followed. Collaborative accounting applies the generated rules to the bill information, performing cross-enterprise accounting, tax calculations, cost allocation, and other operations to obtain the final collaborative accounting result.

[0061] After receiving cross-enterprise invoice information from users, the cross-enterprise financial and tax robot breaks down the overall processing objective into four sequentially executed sub-tasks: invoice processing, knowledge query, rule generation, and collaborative accounting, using a set of enterprise financial and tax collaborative tasks. A task execution queue is then established. Each sub-task has clearly defined input and output interfaces, with the output of the previous task serving as the input for the next. After decomposition, the robot sequentially calls the processing modules of each sub-task. The cross-enterprise financial and tax robot uses an OCR engine to perform text recognition on invoice images or PDFs; if the invoice is already structured data, it directly parses it. After recognition, key fields are extracted, including invoice code, invoice number, invoice date, amount, tax, buyer's name and taxpayer identification number, seller's name and taxpayer identification number, name of goods or taxable services, quantity, and unit price. Simultaneously, the cross-enterprise financial and tax robot performs compliance verification on the invoices, such as verifying the authenticity of the invoice through an invoice verification interface, checking whether the invoice amount matches the tax amount, and verifying whether the invoice date is within a reasonable range. For invoices involving multiple parties, such as a single invoice covering both the sale of goods and transportation services, with transportation provided by a third party, it is necessary to extract detailed line information to identify different business types and their corresponding counterparties. After processing, the original invoice is transformed into a clearly structured and accurate cross-enterprise invoice information document, containing multiple lines of transaction details. This cross-enterprise invoice information is structured information extracted from the original invoice after the invoice processing task, including invoice code, invoice number, invoice date, amount, tax amount, names and taxpayer identification numbers of both the buyer and seller, and details of the goods or services.

[0062] Using the company name and taxpayer identification number from the invoice information as query criteria, the corresponding company entity nodes are located in the enterprise financial and tax knowledge graph update model, and profile tags for these companies are obtained. Based on the names of goods or services in the invoice details, the corresponding tax item entities are queried, and applicable tax rates and tax calculation rules are obtained. Accounting standards and tax regulations related to the transaction type are retrieved, including complete information on multiple company nodes, attributes of multiple tax item nodes, and multiple relationship paths, and integrated into a cross-enterprise related financial and tax knowledge set. This cross-enterprise related financial and tax knowledge set is all knowledge related to the current invoice information retrieved from the enterprise financial and tax knowledge graph update model through a knowledge query task.

[0063] After acquiring a cross-enterprise related tax knowledge set, the cross-enterprise tax robot needs to perform dynamic rule derivation. Invoice information and the related knowledge set are input into the rule engine, which has a series of built-in tax-related reasoning rules. For example, if the traded goods belong to a certain industry and the enterprise is a general taxpayer, the applicable tax rate is 13%; if the enterprise is a high-tech enterprise and the transaction involves R&D, it may enjoy additional deductions for R&D expenses; if a transaction includes both the sale of goods and transportation services, and the transportation services are provided by a third party, the freight costs need to be separately accounted for and included in sales expenses or raw material costs, depending on the contract agreement. Based on each transaction in the invoice details, combined with the profiles of the involved enterprises, these rules are matched and instantiated one by one. For example, for the sales of goods details, an accounting rule is generated to debit accounts receivable: buyer, credit main business revenue, and credit taxes payable - VAT payable, and the tax rate is determined to be 13%. For the transportation service details, if the transportation service is prepaid by the seller but actually borne by the buyer, a rule is generated to debit other receivables: prepaid freight, and credit bank deposit. All generated rules are categorized by enterprise, forming a set of enterprise-related accounting rules. This set includes customized accounting directions and amount calculation methods for each enterprise. The enterprise-related accounting rule set is a collection of specific accounting rules generated after rule derivation, applicable to the current invoice and the enterprises involved. It may include accounting rules (debit accounts, credit accounts), tax rules (tax rates, tax calculation methods), and allocation rules (the proportion of expenses each enterprise should bear), etc.

[0064] Based on the enterprise-related accounting rule set, collaborative accounting optimization is performed on cross-enterprise invoice information. This involves applying the enterprise-related accounting rule set to automate the financial and tax processing of cross-enterprise invoice information, including generating accounting vouchers, calculating tax payable for each enterprise, and adjusting internal accounts receivable and payable. The final result is a cross-enterprise collaborative financial and tax accounting report, including separate accounting entries for each enterprise, summarized tax returns, and internal transaction reconciliation results, meeting the financial and tax needs of all involved enterprises. The cross-enterprise collaborative financial and tax accounting result represents the final compliance check result for multiple enterprises in terms of taxation and finance, and can be directly imported into each enterprise's financial system to complete the accounting processing for this cross-enterprise transaction. Based on a dynamically updated enterprise financial and tax knowledge graph, the robot can accurately obtain the latest profiles and rules of each enterprise, thereby generating accounting rules that conform to the actual situation and latest requirements of each enterprise. The cross-enterprise financial and tax robot can efficiently coordinate the financial and tax processing tasks of multiple enterprises, ensuring the consistency and accuracy of cross-enterprise tax processing.

[0065] In summary, the knowledge collaborative optimization method for cross-enterprise financial and tax robots provided in this application has the following technical effects: It collects multi-source financial and tax datasets from multiple target enterprises, performs knowledge extraction and modeling on these datasets to generate an enterprise financial and tax knowledge graph model, and loads this model into the cross-enterprise financial and tax robot; it performs profile analysis based on the attribute data of the multiple target enterprises to construct multi-target enterprise profiles, while simultaneously crawling updated financial and tax regulations and guidelines in real time; it updates the enterprise financial and tax knowledge graph model using reinforcement learning based on the updated regulations and guidelines and the target enterprise profiles, generating an updated enterprise financial and tax knowledge graph model; it acquires cross-enterprise invoice information to be processed, and the cross-enterprise financial and tax robot performs collaborative optimization processing on this information based on the updated enterprise financial and tax knowledge graph model. In other words, by collecting multi-source financial and tax data, a corporate financial and tax knowledge graph model is constructed. Based on corporate attribute data, a tag system is designed, tag values ​​are calculated, and model verification is performed to form a multi-objective corporate profile. Updated financial and tax regulations and guidelines are crawled, and the updated regulations are associated and matched with the corporate profile to quantify the impact of knowledge. Reinforcement learning is used to incrementally update the knowledge graph. After receiving the invoices to be processed, the task is broken down, and knowledge query and rule generation are performed based on the updated knowledge graph. Finally, collaborative optimization processing is completed to improve the efficiency and accuracy of cross-enterprise financial and tax collaborative processing.

[0066] Example 2: Based on the same inventive concept as the knowledge collaborative optimization method for cross-enterprise financial and tax robots in Example 1, this application also provides a knowledge collaborative optimization system for cross-enterprise financial and tax robots. Please refer to the appendix. Figure 2 The knowledge collaborative optimization system for cross-enterprise financial and tax robots includes: a knowledge extraction and modeling module 11, used to collect multi-source financial and tax datasets from multiple target enterprises, perform knowledge extraction and modeling on the multi-source financial and tax datasets, generate an enterprise financial and tax knowledge graph model, and load the enterprise financial and tax knowledge graph model into the cross-enterprise financial and tax robot; a profile analysis module 12, used to perform profile analysis based on the attribute data of the multiple target enterprises, construct profiles of the multiple target enterprises, and simultaneously crawl updated financial and tax regulations and guidelines in real time; a reinforcement learning update module 13, used to perform reinforcement learning updates on the enterprise financial and tax knowledge graph model based on the updated financial and tax regulations and guidelines and the profiles of the target enterprises, generating an updated enterprise financial and tax knowledge graph model; and a collaborative optimization processing module 14, used to acquire cross-enterprise invoice information to be processed, and to perform collaborative optimization processing on the cross-enterprise invoice information to be processed by the cross-enterprise financial and tax robot based on the updated enterprise financial and tax knowledge graph model.

[0067] Furthermore, the knowledge extraction and modeling module 11 in the knowledge collaborative optimization system for cross-enterprise financial and tax robots is also used for: constructing a data cleaning process, which includes removing duplicate data, filling missing values, handling outliers, and unifying data formats; performing data cleaning processing on the multi-source financial and tax dataset according to the data cleaning process to obtain a usable financial and tax dataset; performing text recognition extraction and data field parsing on the usable financial and tax dataset using OCR and NLP to obtain a key financial and tax dataset; and performing knowledge extraction and modeling optimization based on the key financial and tax dataset to generate an enterprise financial and tax knowledge graph model.

[0068] Furthermore, the knowledge extraction and modeling module 11 in the knowledge collaborative optimization system for cross-enterprise financial and tax robots is also used for: defining knowledge entity types according to the enterprise's financial and tax processing needs, wherein the knowledge entity types include enterprises, accounting subjects, tax items, regulations and standards, and transaction events; defining entity relationship types based on the knowledge entity types, wherein the entity relationship types include financial relationships, tax relationships, and business relationships; extracting knowledge from the key financial and tax dataset according to the knowledge entity types and entity relationship types to obtain an enterprise financial and tax knowledge set; designing graph structure information, and performing fusion modeling optimization on the enterprise financial and tax knowledge set based on the graph structure information to generate an enterprise financial and tax knowledge graph model.

[0069] Furthermore, the knowledge extraction and modeling module 11 in the knowledge collaborative optimization system for cross-enterprise financial and tax robots is also used to: perform knowledge fusion on the enterprise financial and tax knowledge set based on the graph structure information to construct an initial financial and tax knowledge graph model; perform conflict detection on the initial financial and tax knowledge graph model to obtain financial and tax conflict knowledge, which includes data conflict and semantic conflict; and perform reasoning enhancement optimization on the initial financial and tax knowledge graph model based on the financial and tax conflict knowledge to generate an enterprise financial and tax knowledge graph model.

[0070] Furthermore, the profile analysis module 12 in the knowledge collaboration optimization system for cross-enterprise financial and tax robots is also used for: designing an enterprise profile tag system, which includes basic attribute tags, financial attribute tags, tax attribute tags, and business attribute tags; calculating tags on the attribute data of the multi-target enterprises according to the enterprise profile tag system to obtain multi-enterprise profile tag data; and performing profile modeling analysis and verification correction based on the multi-enterprise profile tag data to construct multi-target enterprise profiles.

[0071] Furthermore, the reinforcement learning update module 13 in the knowledge collaborative optimization system for cross-enterprise financial and tax robots is also used for: standardizing and semantically parsing the financial and tax update regulations and guidelines to obtain financial and tax update regulation rule information; associating and matching the target enterprise profile with the financial and tax update regulation rule information to obtain an enterprise-related regulation rule set; and performing knowledge impact quantification and reinforcement learning update on the enterprise financial and tax knowledge graph model based on the enterprise-related regulation rule set to generate an enterprise financial and tax knowledge graph update model.

[0072] Furthermore, the collaborative optimization processing module 14 in the knowledge collaborative optimization system for cross-enterprise financial and tax robots is also used to: decompose the collaborative tasks of the cross-enterprise financial and tax robots to obtain a set of enterprise financial and tax collaborative tasks, which includes invoice processing, knowledge query, rule generation, and collaborative accounting; and perform collaborative optimization processing on the cross-enterprise invoice information to be processed according to the enterprise financial and tax collaborative task set based on the enterprise financial and tax knowledge graph update model.

[0073] Furthermore, the collaborative optimization processing module 14 in the knowledge collaborative optimization system for cross-enterprise financial and tax robots is also used to: identify and verify the cross-enterprise invoice information to be processed according to the enterprise financial and tax collaborative task set to obtain cross-enterprise invoice information; perform knowledge query and rule generation on the cross-enterprise invoice information based on the enterprise financial and tax knowledge graph update model to determine the enterprise association accounting rule set; and use the enterprise association accounting rule set to perform collaborative accounting optimization processing on the cross-enterprise invoice information to obtain the cross-enterprise financial and tax collaborative accounting result.

[0074] Furthermore, the collaborative optimization processing module 14 in the knowledge collaborative optimization system for cross-enterprise financial and tax robots is also used to: perform knowledge query on the cross-enterprise invoice information based on the enterprise financial and tax knowledge graph update model to obtain a cross-enterprise related financial and tax knowledge set; and perform dynamic rule derivation to generate the cross-enterprise related financial and tax knowledge set to determine the enterprise related accounting rule set.

[0075] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The knowledge collaborative optimization method and specific examples for cross-enterprise financial and tax robots in the aforementioned embodiment one are also applicable to the knowledge collaborative optimization system for cross-enterprise financial and tax robots in this embodiment. Through the foregoing detailed description of the knowledge collaborative optimization method for cross-enterprise financial and tax robots, those skilled in the art can clearly understand the knowledge collaborative optimization system for cross-enterprise financial and tax robots in this embodiment. Therefore, for the sake of brevity, it will not be described in detail here.

[0076] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0077] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of this application and its equivalents, this application also intends to include such modifications and variations.

Claims

1. A knowledge collaboration optimization method for cross-enterprise financial and tax robots, characterized in that, include: Collect multi-source financial and tax datasets from multiple target enterprises, perform knowledge extraction and modeling on the multi-source financial and tax datasets, generate an enterprise financial and tax knowledge graph model, and load the enterprise financial and tax knowledge graph model into a cross-enterprise financial and tax robot; Based on the attribute data of the multi-target enterprises, a profile analysis is performed to construct a multi-target enterprise profile, while the updated financial and tax regulations and guidelines are crawled in real time. Based on the aforementioned tax and financial update regulations and guidelines and the target enterprise profile, the enterprise tax and financial knowledge graph model is updated using reinforcement learning to generate an updated enterprise tax and financial knowledge graph model. The system acquires cross-enterprise invoice information to be processed and then uses the cross-enterprise financial and tax robot to perform collaborative optimization processing on the cross-enterprise invoice information based on the enterprise financial and tax knowledge graph update model.

2. The knowledge collaborative optimization method for cross-enterprise financial and tax robots as described in claim 1, characterized in that, Generate a corporate financial and tax knowledge graph model, including: A data cleaning process is constructed, which includes removing duplicate data, filling missing values, handling outliers, and standardizing data formats. The multi-source financial and tax dataset is cleaned according to the data cleaning process described above to obtain a usable financial and tax dataset. By using OCR and NLP to perform text recognition and extraction and data field parsing on the available financial and tax dataset, key financial and tax datasets can be obtained. Based on the aforementioned key financial and tax dataset, knowledge extraction and modeling optimization are performed to generate an enterprise financial and tax knowledge graph model.

3. The knowledge collaborative optimization method for cross-enterprise financial and tax robots as described in claim 2, characterized in that, Based on the aforementioned key financial and tax dataset, knowledge extraction and modeling optimization are performed to generate an enterprise financial and tax knowledge graph model, including: Based on the enterprise's financial and tax processing needs, knowledge entity types are defined, including enterprises, accounting subjects, tax items, regulations and standards, and transaction events; Based on the knowledge entity type, entity relationship types are defined, including financial relationships, tax relationships, and business relationships. Knowledge is extracted from the key financial and tax dataset according to the knowledge entity type and the entity relationship type to obtain the enterprise financial and tax knowledge set. Design the graph structure information, and based on the graph structure information, perform fusion modeling and optimization on the enterprise financial and tax knowledge set to generate an enterprise financial and tax knowledge graph model.

4. The knowledge collaborative optimization method for cross-enterprise financial and tax robots as described in claim 3, characterized in that, Based on the graph structure information, the enterprise financial and tax knowledge set is fused, modeled, and optimized to generate an enterprise financial and tax knowledge graph model, including: Based on the graph structure information, knowledge fusion is performed on the enterprise financial and tax knowledge set to construct an initial financial and tax knowledge graph model; Conflict detection is performed on the initial fiscal and tax knowledge graph model to obtain fiscal and tax conflict knowledge, which includes data conflict and semantic conflict. Based on the aforementioned knowledge of tax and financial conflicts, the initial tax and financial knowledge graph model is enhanced and optimized through reasoning to generate an enterprise tax and financial knowledge graph model.

5. The knowledge collaborative optimization method for cross-enterprise financial and tax robots as described in claim 1, characterized in that, Constructing multi-objective enterprise profiles, including: Design an enterprise profile tagging system, which includes basic attribute tags, financial attribute tags, tax attribute tags, and business attribute tags; According to the enterprise profile tagging system, the attribute data of the multi-target enterprises are tagged to obtain multi-enterprise profile tag data. Based on the multi-enterprise profile tag data, profile modeling analysis and verification correction are performed to construct multi-target enterprise profiles.

6. The knowledge collaborative optimization method for cross-enterprise financial and tax robots as described in claim 1, characterized in that, Generate an updated model for the enterprise's financial and tax knowledge graph, including: The aforementioned tax and finance update regulations and guidelines are standardized and semantically parsed to obtain tax and finance update regulations and rules information. The target enterprise profile is matched with the updated financial and tax regulations and rules information to obtain a set of enterprise-related regulations and rules; Based on the set of rules and regulations related to the enterprise, the knowledge impact quantification and reinforcement learning are applied to the enterprise financial and tax knowledge graph model to generate an updated enterprise financial and tax knowledge graph model.

7. The knowledge collaborative optimization method for cross-enterprise financial and tax robots as described in claim 1, characterized in that, The cross-enterprise financial and tax robot performs collaborative optimization processing on the cross-enterprise invoice information to be processed based on the enterprise financial and tax knowledge graph update model, including: The cross-enterprise financial and tax robot is decomposed into collaborative tasks to obtain a set of enterprise financial and tax collaborative tasks, which includes invoice processing, knowledge query, rule generation, and collaborative accounting. The cross-enterprise invoice information to be processed is collaboratively optimized based on the enterprise financial and tax collaborative task set and the enterprise financial and tax knowledge graph update model.

8. The knowledge collaborative optimization method for cross-enterprise financial and tax robots as described in claim 7, characterized in that, Based on the enterprise financial and tax collaboration task set and the enterprise financial and tax knowledge graph update model, the cross-enterprise invoice information to be processed is collaboratively optimized, including: The cross-enterprise invoice information to be processed is identified and verified according to the enterprise financial and tax collaboration task set to obtain the cross-enterprise invoice information; Based on the enterprise financial and tax knowledge graph update model, knowledge query and rule generation are performed on the cross-enterprise invoice information to determine the enterprise related accounting rule set; The cross-enterprise invoice information is processed using the enterprise-related accounting rule set to achieve collaborative accounting optimization, resulting in cross-enterprise financial and tax collaborative accounting.

9. The knowledge collaborative optimization method for cross-enterprise financial and tax robots as described in claim 8, characterized in that, Determine the set of rules for accounting related to enterprises, including: Based on the enterprise financial and tax knowledge graph update model, knowledge queries are performed on the cross-enterprise invoice information to obtain a cross-enterprise related financial and tax knowledge set. Dynamic rule derivation is performed on the cross-enterprise related financial and tax knowledge set to determine the enterprise related accounting rule set.

10. A knowledge collaboration and optimization system for cross-enterprise financial and tax robots, characterized in that: The step of implementing the knowledge collaborative optimization method for cross-enterprise financial and tax robots according to any one of claims 1 to 9, wherein the knowledge collaborative optimization system for cross-enterprise financial and tax robots comprises: The knowledge extraction and modeling module is used to collect multi-source financial and tax datasets from multiple target enterprises, perform knowledge extraction and modeling on the multi-source financial and tax datasets, generate enterprise financial and tax knowledge graph models, and load the enterprise financial and tax knowledge graph models into the cross-enterprise financial and tax robot. The profile analysis module is used to perform profile analysis based on the attribute data of the multi-target enterprises, construct multi-target enterprise profiles, and crawl updated financial and tax regulations and guidelines in real time. The reinforcement learning update module is used to perform reinforcement learning updates on the enterprise financial and tax knowledge graph model based on the financial and tax update regulations and the target enterprise profile, and generate an updated enterprise financial and tax knowledge graph model. The collaborative optimization processing module is used to acquire cross-enterprise invoice information to be processed, and to perform collaborative optimization processing on the cross-enterprise invoice information to be processed through the cross-enterprise financial and tax robot based on the enterprise financial and tax knowledge graph update model.

Images