RPA-based finance and tax data processing method and system
Through RPA intelligent agents and financial and tax data analysis models, combined with verification graph structures and knowledge graphs, verification strategies are dynamically selected to solve the problem of low financial and tax data verification accuracy in existing technologies, achieving higher verification accuracy and reducing human intervention.
Patent Information
- Application Number
- CN202510864542.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-06-26
- Publication Date
- 2025-10-14
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
Existing technologies rely on fixed rule sets to process financial and tax data and are unable to adapt to a variety of financial and tax data needs, resulting in low verification accuracy and the need for human intervention.
Financial and tax data are acquired through RPA intelligent agents and financial and tax data analysis models are used. Combined with the verification graph structure construction layer and the financial and tax relationship knowledge graph, verification strategies are dynamically selected to construct semantically enhanced financial and tax structured data, and a self-attention mechanism is executed for verification.
It achieves higher accuracy in financial and tax data verification, reduces human intervention, and adapts to the verification needs of different financial and tax data scenarios.
Smart Images
Figure CN120781090A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of data processing, in particular to a tax data processing method and system based on RPA. BACKGROUND
[0002] With the deepening of digital transformation, the automation and intelligentization of tax data processing are increasingly prominent, and tax data has strict compliance requirements, frequent policy changes and complex business logic, which puts higher requirements on data processing technology. At present, the tax data processing technology in the market mainly relies on fixed rule sets for data verification and processing, but the tax data processing method based on fixed rule sets cannot adapt to various tax data and still needs human intervention. SUMMARY
[0003] The present application obtains the tax data to be checked and the associated tax data through the RPA agent, and then verifies the tax data to be checked through the tax data analysis model to analyze whether the tax data to be checked meets the tax situation; and sets a verification graph structure construction layer in the tax data analysis model to dynamically select the appropriate verification strategy under the guidance of the tax relationship knowledge graph, no longer limited to fixed rule sets, and more suitable for the tax data to be checked, thereby obtaining higher verification accuracy.
[0004] The present application provides a tax data processing method based on RPA, comprising:
[0005] The RPA agent obtains the tax data to be checked, and the RPA agent corresponding to the API obtains the associated tax data of the tax data to be checked, and performs preprocessing operations on the tax data to be checked and the associated tax data to obtain corresponding tax structured data and associated tax structured data, and sends the tax structured data and the associated tax structured data into the tax data analysis model for processing, and outputs the tax data analysis label;
[0006] The tax data analysis model includes a verification graph structure construction layer, a tax data verification layer and a tax data analysis label output layer, wherein the verification graph structure construction layer is used to construct a verification graph structure according to the tax structured data and the tax relationship knowledge graph; the tax data verification layer is used to perform a verification task on the tax structured data according to the verification graph structure and the associated tax structured data, and construct a tax data analysis label; and the tax data analysis label output layer is used to output the tax data analysis label.
[0007] As a preferred aspect, the verification graph structure construction layer constructs a verification graph structure according to the tax structured data and the tax relationship knowledge graph, specifically including the following steps:
[0008] The semantic information fusion network built in the verification graph structure construction layer is used to process the tax structured data, and semantic enhanced tax structured data is constructed.
[0009] Based on the tax structured data, the tax relationship knowledge graph is queried, and relevant tax relationship knowledge triples are output. Then, all relevant tax relationship knowledge triples are combined into a tax relationship knowledge feature graph. Based on the tax relationship knowledge feature graph, a self-attention mechanism is performed on the tax structured data to construct a tax relationship analysis vector. In the process of performing the self-attention mechanism, a value vector and a key vector are constructed based on the tax structured data, and a query vector is constructed based on the tax relationship knowledge feature graph.
[0010] Then, the tax relationship analysis vector is subjected to a full connection operation to output a tax relationship node set. The tax relationship node set includes a plurality of tax relationship nodes. The tax relationship node set is mapped to the tax relationship knowledge graph to segment out a verification graph structure.
[0011] As a preferred aspect, the semantic information fusion network built in the verification graph structure construction layer is used to process the tax structured data, and semantic enhanced tax structured data is constructed. Specifically, the following steps are included:
[0012] The tax structured data is segmented by a sliding window to obtain a plurality of tax structured data segments. The size of the sliding window is C, and the step size is S.
[0013] All tax structured data segments are spliced from top to bottom according to the segmentation order to construct a tax structured data feature graph. Then, the tax structured data feature graph is input into the semantic information fusion network to process the tax structured data, and semantic enhanced tax structured data is constructed. The semantic information fusion network is based on a Transformer model.
[0014] As a preferred aspect, the tax data verification layer is used to perform a verification task on the tax structured data based on the verification graph structure and the associated tax structured data, and a tax data analysis label is constructed. Specifically, the following steps are included:
[0015] For each associated tax structured data, the following operations are performed. The selected associated tax structured data is denoted as target associated tax structured data. Then, the tax structured data and the target associated tax structured data are spliced to construct a verification analysis vector. A tax relationship correlation feature matrix is constructed based on the verification graph structure, the tax structured data, and the target associated tax structured data. A self-attention mechanism is performed on the verification analysis vector based on the tax relationship correlation feature matrix, and then a full connection operation is performed to construct a local verification score. In the process of performing the self-attention mechanism, a value vector and a key vector are constructed based on the verification analysis vector, and a query vector is constructed based on the tax relationship correlation feature matrix.
[0016] The partial check scores are weighted and summed to obtain a check score, and a tax data analysis label is determined based on the check score.
[0017] As a preferred aspect, a tax relationship correlation feature matrix is constructed based on the check graph structure, the tax structured data and the target associated tax structured data, and specifically includes the following steps:
[0018] The tax structured data and the target associated tax structured data are abstracted into corresponding analysis tax nodes and target tax nodes, and the analysis tax nodes and the target tax nodes both correspond to entities in the tax relationship knowledge graph, a tax relationship correlation feature matrix is constructed, the size of the tax relationship correlation feature matrix is MxN, M is the total number of analysis tax nodes, N is the total number of target tax nodes, the (i, j) element of the tax relationship correlation feature matrix stores a feature vector corresponding to an entity relationship between the i-th analysis tax node and the j-th target tax node, i=1, 2, 3, …, M, j=1, 2, 3, …, N.
[0019] As a preferred aspect, the tax data analysis model is trained, and specifically includes the following steps:
[0020] A plurality of tax data analysis training samples are obtained, the tax data analysis training samples include the tax data to be checked and the corresponding tax structured data and associated tax structured data of the associated tax data, the tax data analysis training samples are labeled by tax data analysis labels, all the labeled tax data analysis training samples are combined to form a tax data analysis training set, the tax data analysis model is trained by the tax data analysis training set, it is determined whether the training condition is met, if the training condition is met, the trained tax data analysis model is output; otherwise, the tax data analysis model is continuously trained by the tax data analysis training set.
[0021] The application also provides a tax data processing system based on RPA, which comprises:
[0022] The data acquisition module is configured to acquire the tax data to be checked by the RPA agent, acquire the associated tax data of the tax data to be checked by the API corresponding to the RPA agent, and perform preprocessing on the tax data to be checked and the associated tax data to obtain the corresponding tax structured data and associated tax structured data.
[0023] The tax data analysis module is configured to input the tax structured data and the associated tax structured data into the tax data analysis model for processing, and output a tax data analysis label.
[0024] The financial tax data analysis model comprises a verification graph structure construction layer, a financial tax data verification layer and a financial tax data analysis label output layer, wherein the verification graph structure construction layer is configured to construct a verification graph structure according to financial tax structured data and a financial tax relationship knowledge graph; the financial tax data verification layer is configured to perform a verification task on the financial tax structured data according to the verification graph structure and the associated financial tax structured data, and construct a financial tax data analysis label; and the financial tax data analysis label output layer is configured to output the financial tax data analysis label.
[0025] The present application has the following advantages:
[0026] The present application has the following advantages: BRIEF DESCRIPTION OF DRAWINGS
[0027] Figure 1 The present application has the following advantages:
[0028] Figure 2 The present application has the following advantages: DETAILED DESCRIPTION
[0029] In order to better understand the technical solutions in the present application, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application.
[0030] Embodiment 1, a financial tax data processing method based on RPA, comprising:
[0031] The RPA agent obtains the tax data to be checked, which specifically refers to tax-related invoices such as value-added tax invoices and sales unified invoices, for reflecting the tax situation of the company, and obtains the tax data associated with the tax data to be checked through the API corresponding to the RPA agent. The associated tax data refers to the tax data associated with the tax data to be checked in the historical records, such as the tax data associated with the sales unified invoice including the purchase order data and the receipt order data, etc. The tax data to be checked and the associated tax data are preprocessed to obtain corresponding tax structured data and associated tax structured data. The tax structured data and the associated tax structured data corresponding to the tax data to be checked and the associated tax data are input into the tax data analysis model for processing, and the tax data analysis label is output. The tax data analysis label includes normal and abnormal, and the tax data analysis label can reflect whether the tax data to be checked conforms to the tax situation. It should be noted that the tax data associated with the tax data to be checked is realized based on the rule matching of the invoice. The operator sets the matching rule in advance, for example, the sales invoice can match the invoice of the supplier, and the matching is performed through the name of the supplier.
[0032] The preprocessing operation includes OCR recognition and word embedding operation. The OCR recognition is used to extract the content on the paper invoice as text data, and the word embedding operation is used to convert the corresponding words on the invoice into numerical content. The word embedding can use Word2Vec and Glove technical means.
[0033] Referring to Figure 1 The tax data analysis model includes a verification graph structure construction layer, a tax data verification layer, and a tax data analysis label output layer. The verification graph structure construction layer is used to construct a verification graph structure according to the tax structured data and the tax relationship knowledge graph. The verification graph structure can dynamically select a verification strategy according to the tax data to be checked, instead of a fixed rule set, which can better fit the tax data to be checked, thereby obtaining higher verification accuracy. The tax relationship knowledge graph stores tax relationship knowledge triples, i.e., entity-entity relationship-entity triples, and the tax relationship knowledge triples represent the tax relationship corresponding to the tax situation, such as invoice-order consistency and sales data consistency. The tax data verification layer is used to perform a verification task on the tax structured data according to the verification graph structure and the associated tax structured data, and construct a tax data analysis label. Under the guidance of the verification strategy corresponding to the verification graph structure, each associated tax data is used to verify the tax data to be checked, to determine whether the tax data to be checked conforms to the tax situation of the historical records, and to determine the tax data analysis label corresponding to the tax data to be checked based on the verification result. The tax data analysis label output layer is used to output the tax data analysis label.
[0034] The construction mode of the tax relationship knowledge graph is that an operator constructs a series of entity-entity relationship-entity triples according to the actual tax relationship of the company and expert experience, and then all the entity-entity relationship-entity triples are combined to form the tax relationship knowledge graph;
[0035] The application obtains the tax data to be checked and the associated tax data through the RPA intelligent agent, and then verifies the tax data to be checked through the tax data analysis model to analyze whether the tax data to be checked conforms to the tax situation; and a verification graph structure construction layer is set in the tax data analysis model, so that suitable verification strategies are dynamically selected under the guidance of the tax relationship knowledge graph, instead of being limited to a fixed rule set, which can better fit the tax data to be checked, thereby obtaining higher verification accuracy.
[0036] The verification graph structure construction layer constructs a verification graph structure according to the tax structured data and the tax relationship knowledge graph, and specifically includes the following steps:
[0037] The semantic information fusion network built in the verification graph structure construction layer processes the tax structured data to construct semantic enhanced tax structured data, extracts the semantic relationship between the content context in the tax structured data, so that the obtained semantic enhanced tax structured data can more characteristically describe the corresponding tax data to be checked, and then the subsequent query to the tax relationship knowledge graph is more accurate;
[0038] Based on the tax structured data, the tax relationship knowledge graph is queried, and relevant tax relationship knowledge triples are output. It should be noted that the content in the tax structured data generally includes invoice code, number, issue date, buyer name / tax number, seller name / tax number, commodity / service name, specification, unit, quantity, unit price, amount, tax rate, tax amount, total price with tax, etc. These words can be abstracted as entity concepts and can be matched with entities in the tax relationship knowledge graph, and a successful match is considered to be relevant. All relevant tax relationship knowledge triples are combined to form a tax relationship knowledge feature graph. Based on the tax relationship knowledge feature graph, a self-attention mechanism is performed on the tax structured data to construct a tax relationship analysis vector. The tax relationship analysis vector can reflect the adaptive tax relationship of the tax data to be checked. In the process of performing the self-attention mechanism, the tax structured data is used to construct a corresponding value vector and a key vector, and the tax relationship knowledge feature graph is used to construct a corresponding query vector.
[0039] The financial and tax relationship analysis vector is then fully connected, and the financial and tax relationship node set is output. The financial and tax relationship node set includes a plurality of financial and tax relationship nodes, and the financial and tax relationship node set is mapped to the financial and tax relationship knowledge graph to segment the verification graph structure. It should be noted that the full connection operation outputs the correlation probability corresponding to all financial and tax relationship nodes. The financial and tax relationship nodes correspond to all entities in the financial and tax relationship knowledge graph. If the correlation probability is higher than the pre-set threshold, the financial and tax relationship node is added to the financial and tax relationship node set;
[0040] It should be noted that the self-attention mechanism is based on the self-attention mechanism in the Transformer model. Generally, a value vector V, a key vector K, and a query vector Q are constructed based on the input vector. The self-attention mechanism is implemented through the formula: G = softmax(QK T / D 0.5 )T V, where T is the matrix transpose operation, D is the dimension size of the key vector, and G is the vector output by the self-attention mechanism.
[0041] The semantic information fusion network built-in the verification graph structure construction layer processes the financial and tax structured data, and constructs semantic enhanced financial and tax structured data, including the following steps:
[0042] The financial and tax structured data is segmented by a sliding window to obtain a plurality of financial and tax structured data segments. The size of the sliding window is C, and the step size is S. The specific value of C is determined by the operator, and the value of S is generally C-1.
[0043] All financial and tax structured data segments are spliced from top to bottom according to the segmentation order to construct a financial and tax structured data feature map. The financial and tax structured data feature map is sent to the semantic information fusion network to process the financial and tax structured data, and construct semantic enhanced financial and tax structured data. The semantic information fusion network is based on the Transformer model and is used to analyze the context dependency relationship in the financial and tax structured data.
[0044] The financial and tax data verification layer verifies the financial and tax structured data based on the verification graph structure and the related financial and tax structured data, and constructs a financial and tax data analysis label, including the following steps:
[0045] For each associated tax structured data, the following is performed, the selected associated tax structured data is recorded as the target associated tax structured data, the tax structured data and the target associated tax structured data are spliced to construct a verification analysis vector, a tax relationship association feature matrix is constructed based on the verification graph structure, the tax structured data and the target associated tax structured data, and after performing a self-attention mechanism on the verification analysis vector based on the tax relationship association feature matrix, a full connection operation is performed to construct a local verification score. During the self-attention mechanism process, a corresponding value vector and key vector are constructed based on the verification analysis vector, and a corresponding query vector is constructed based on the tax relationship association feature matrix;
[0046] All local verification scores are weighted and summed to obtain a verification score, and a tax data analysis label is determined based on the verification score. The specific determination method is that when the verification score is higher than the score threshold set by the operator, the corresponding normal tax data analysis label is output, and when the verification score is not higher than the score threshold set by the operator, the corresponding abnormal tax data analysis label is output. The weight during the weighted summation is generally used as a hyperparameter and can be set through a group optimization algorithm;
[0047] Based on the verification graph structure, the tax structured data and the target associated tax structured data, a tax relationship association feature matrix is constructed, which includes the following steps:
[0048] The tax structured data and the target associated tax structured data are abstracted into corresponding analysis tax nodes and target tax nodes, and the analysis tax nodes and the target tax nodes correspond to entities in the tax relationship knowledge graph. A tax relationship association feature matrix is constructed. The size of the tax relationship association feature matrix is MxN, M is the total number of analysis tax nodes, N is the total number of target tax nodes, the i-th row and j-th column of the tax relationship association feature matrix stores the feature vector corresponding to the entity relationship between the i-th analysis tax node and the j-th target tax node, i=1, 2, 3, …, M, j=1, 2, 3, …, N. It should be noted that the entity relationship between the i-th analysis tax node and the j-th target tax node is derived from the verification graph structure, and when the i-th analysis tax node and the j-th target tax node do not have an entity relationship, the value stored in the i-th row and j-th column of the tax relationship association feature matrix is 0.
[0049] The tax data analysis model is trained, which includes the following steps:
[0050] A plurality of fiscal and tax data analysis training samples are obtained, the fiscal and tax data analysis training samples include fiscal and tax data to be checked and corresponding fiscal and tax structured data and associated fiscal and tax structured data of the associated fiscal and tax data. It should be noted that the fiscal and tax data to be checked and the associated fiscal and tax data are generated in actual company operation, and the fiscal and tax data analysis training samples are labeled by fiscal and tax data analysis labels. The fiscal and tax data analysis labels are obtained after human inspection, and the normal corresponding check score is 1 and the abnormal corresponding check score is 0. All labeled fiscal and tax data analysis training samples form a fiscal and tax data analysis training set. The fiscal and tax data analysis model is trained through the fiscal and tax data analysis training set, and whether the training condition is met is determined. The training condition is generally that the accuracy of the fiscal and tax data analysis model meets the expectation. If the training condition is met, the trained fiscal and tax data analysis model is output; otherwise, the fiscal and tax data analysis model is continuously trained through the fiscal and tax data analysis training set.
[0051] In embodiment 2, a fiscal and tax data processing system based on RPA is provided, as shown in Figure 2 , comprising:
[0052] The data acquisition module is configured to obtain the fiscal and tax data to be checked by the RPA agent. The fiscal and tax data specifically refers to the fiscal and tax related invoices such as value-added tax special invoices and sales unified invoices, which are used to reflect the fiscal and tax situation of the company. The RPA agent is configured to obtain the associated fiscal and tax data of the fiscal and tax data to be checked through the corresponding API. The associated fiscal and tax data refers to the fiscal and tax data associated with the fiscal and tax data to be checked in the historical record, such as the associated fiscal and tax data of the sales unified invoice including the purchase order data and the receipt order data. The fiscal and tax data to be checked and the associated fiscal and tax data are preprocessed to obtain corresponding fiscal and tax structured data and associated fiscal and tax structured data.
[0053] The fiscal and tax data analysis module is configured to send the corresponding fiscal and tax structured data and the associated fiscal and tax structured data of the fiscal and tax data to be checked and the associated fiscal and tax data into the fiscal and tax data analysis model for processing, and output the fiscal and tax data analysis label. The fiscal and tax data analysis label includes normal and abnormal, and the fiscal and tax data analysis label can reflect whether the fiscal and tax data to be checked meets the fiscal and tax situation.
[0054] The finance and tax data analysis model comprises a verification graph structure construction layer, a finance and tax data verification layer and a finance and tax data analysis label output layer. The verification graph structure construction layer is configured to construct a verification graph structure according to finance and tax structured data and a finance and tax relationship knowledge graph. The verification graph structure can dynamically select a verification strategy according to the finance and tax data to be checked, instead of a fixed rule set, and can be more suitable for the finance and tax data to be checked, thereby obtaining a higher verification accuracy. The finance and tax relationship knowledge graph stores finance and tax relationship knowledge triples, i.e., entity-entity relationship-entity triples, and the finance and tax relationship knowledge triples represent the corresponding finance and tax relationship of the finance and tax situation, such as invoice-order consistency and sales data consistency. The finance and tax data verification layer is configured to perform a verification task on the finance and tax structured data according to the verification graph structure and the associated finance and tax structured data, and to construct a finance and tax data analysis label. Under the guidance of the verification strategy corresponding to the verification graph structure, each associated finance and tax data is used to verify the finance and tax data to be checked, to determine whether the finance and tax data to be checked conforms to the historical record of the finance and tax situation, and to determine the finance and tax data analysis label corresponding to the finance and tax data to be checked based on the verification result. The finance and tax data analysis label output layer is configured to output the finance and tax data analysis label.
[0055] It should be understood that, for those skilled in the art, improvements or changes can be made according to the above description, and all such improvements and changes shall fall within the protection scope of the appended claims of the present application. The parts not described in detail in the specification belong to the prior art known to those skilled in the art.
Claims
1. A method for processing financial and tax data based on RPA, characterized in that: include: The RPA agent obtains the financial and tax data to be checked, and obtains the financial and tax data associated with the financial and tax data to be checked through the RPA agent's corresponding API. The financial and tax data to be checked and the associated financial and tax data are preprocessed to obtain the corresponding financial and tax structured data and the associated financial and tax structured data. The financial and tax structured data and the associated financial and tax structured data are fed into the financial and tax data analysis model for processing, and the financial and tax data analysis labels are output. The financial and taxation data analysis model includes a verification graph structure construction layer, a financial and taxation data verification layer, and a financial and taxation data analysis label output layer. The verification graph structure construction layer is used to construct a verification graph structure based on financial and taxation structured data and the financial and taxation relationship knowledge graph; the financial and taxation data verification layer is used to perform verification tasks on financial and taxation structured data based on the verification graph structure and related financial and taxation structured data, and to construct financial and taxation data analysis labels; the financial and taxation data analysis label output layer is used to output financial and taxation data analysis labels.
2. The RPA-based fiscal and taxation data processing method according to claim 1 is characterized in that: The verification graph structure construction layer constructs the verification graph structure based on the financial and tax structured data and the financial and tax relationship knowledge graph. The specific steps include: The financial and tax structured data is processed by the semantic information fusion network built into the verification graph structure construction layer to construct semantically enhanced financial and tax structured data; Based on the financial and tax structured data, the financial and tax relationship knowledge graph is queried, and relevant financial and tax relationship knowledge triples are output. Then, all relevant financial and tax relationship knowledge triples are combined into a financial and tax relationship knowledge feature graph. Based on the financial and tax relationship knowledge feature graph, a self-attention mechanism is performed on the financial and tax structured data to construct a financial and tax relationship analysis vector. In the process of executing the self-attention mechanism, the corresponding value vector and key vector are constructed using the financial and tax structured data, and the corresponding query vector is constructed using the financial and tax relationship knowledge feature graph. Then, the finance and taxation relationship analysis vector is fully connected to output a finance and taxation relationship node set, which includes several finance and taxation relationship nodes. The finance and taxation relationship node set is mapped to the finance and taxation relationship knowledge graph to segment the verification graph structure.
3. The RPA-based fiscal and taxation data processing method according to claim 2 is characterized in that: The financial and tax structured data is processed by the semantic information fusion network built into the verification graph structure construction layer to construct semantically enhanced financial and tax structured data. The specific steps include the following: The financial and tax structured data is segmented through a sliding window to obtain several financial and tax structured data fragments. The size of the sliding window is C and the step length is S; All financial and tax structured data fragments are spliced from top to bottom in the order of segmentation to construct a financial and tax structured data feature map, and then the financial and tax structured data feature map is sent to the semantic information fusion network to process the financial and tax structured data and construct semantically enhanced financial and tax structured data. The semantic information fusion network is established based on the Transformer model.
4. The RPA-based fiscal and taxation data processing method according to claim 3 is characterized in that: The fiscal and taxation data verification layer verifies the fiscal and taxation structured data based on the verification graph structure and associated fiscal and taxation structured data, and constructs fiscal and taxation data analysis tags. The specific steps include the following: For each associated fiscal and tax structured data, the following content is executed: the selected associated fiscal and tax structured data is recorded as the target associated fiscal and tax structured data, and then the fiscal and tax structured data and the target associated fiscal and tax structured data are spliced to construct a verification analysis vector. A fiscal and tax relationship association feature matrix is constructed based on the verification graph structure, the fiscal and tax structured data, and the target associated fiscal and tax structured data. A self-attention mechanism is performed on the verification analysis vector based on the fiscal and tax relationship association feature matrix, followed by a full connection operation to construct a local verification score. In the process of executing the self-attention mechanism, the corresponding value vector and key vector are constructed using the verification analysis vector, and the corresponding query vector is constructed using the fiscal and tax relationship association feature matrix. All local verification scores are weighted and summed to obtain the verification score, and the financial and tax data analysis label is determined based on the verification score.
5. The RPA-based fiscal and taxation data processing method according to claim 4 is characterized in that: A finance and taxation relationship association feature matrix is constructed based on the verification graph structure, finance and taxation structured data and target-related finance and taxation structured data, specifically including the following steps: abstracting the finance and taxation structured data and the target-related finance and taxation structured data into corresponding analysis finance and taxation nodes and target finance and taxation nodes, and both the analysis finance and taxation nodes and the target finance and taxation nodes correspond to entities in the finance and taxation relationship knowledge graph, and constructing a finance and taxation relationship association feature matrix. The size of the finance and taxation relationship association feature matrix is M×N, where M is the total number of analysis finance and taxation nodes, and N is the total number of target finance and taxation nodes. The i-th row and j-th column of the finance and taxation relationship association feature matrix stores the feature vector corresponding to the entity relationship between the i-th analysis finance and taxation node and the j-th target finance and taxation node, i=1, 2, 3,…, M, j=1, 2, 3,…, N.
6. The RPA-based fiscal and taxation data processing method according to claim 5 is characterized in that: Training the financial and tax data analysis model includes the following steps: Obtain several financial and taxation data analysis training samples, which include the financial and taxation data to be checked and the financial and taxation structured data corresponding to the related financial and taxation data, and the related financial and taxation structured data; label the financial and taxation data analysis training samples with financial and taxation data analysis labels; form a financial and taxation data analysis training set with all labeled financial and taxation data analysis training samples; train the financial and taxation data analysis model with the financial and taxation data analysis training set to determine whether the training conditions are met; if so, output the trained financial and taxation data analysis model; otherwise, continue to train the financial and taxation data analysis model with the financial and taxation data analysis training set.
7. A financial and tax data processing system based on RPA, characterized by: The system applies the RPA-based financial and tax data processing method according to any one of claims 1 to 6, including: The data acquisition module is used to obtain the financial and tax data to be checked through the RPA agent, and obtain the financial and tax data associated with the financial and tax data to be checked through the API corresponding to the RPA agent. The module pre-processes the financial and tax data to be checked and the associated financial and tax data to obtain the corresponding financial and tax structured data and the associated financial and tax structured data. The finance and taxation data analysis module is used to feed finance and taxation structured data and related finance and taxation structured data into the finance and taxation data analysis model for processing and output finance and taxation data analysis labels; The financial and taxation data analysis model includes a verification graph structure construction layer, a financial and taxation data verification layer, and a financial and taxation data analysis label output layer. The verification graph structure construction layer is used to construct a verification graph structure based on financial and taxation structured data and the financial and taxation relationship knowledge graph; the financial and taxation data verification layer is used to perform verification tasks on financial and taxation structured data based on the verification graph structure and related financial and taxation structured data, and to construct financial and taxation data analysis labels; the financial and taxation data analysis label output layer is used to output financial and taxation data analysis labels.