Intelligent finance and tax management method and system based on vector driving and storage medium
By vectorizing and stitching together internal financial and tax data and external public opinion data, and combining them with an attention mechanism for risk prediction, the problem of traditional models being unable to capture external dynamic factors in real time has been solved, thus achieving accurate prediction and tracing of financial and tax risks.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- HENAN WEITIAN ELECTRONIC TECHNOLOGY CO LTD
- Filing Date
- 2025-11-11
- Publication Date
- 2026-08-04
AI Technical Summary
Traditional fiscal and tax risk prediction models struggle to capture external dynamic factors in real time, and the prediction results lack interpretability and traceability, leading to poor fiscal and tax management performance.
By acquiring internal financial and tax data and external public opinion data, vectorizing them, and concatenating them into a joint vector, attention weights are obtained using an attention mechanism, and risk prediction and source tracing are performed in conjunction with a fully connected neural network.
It enables accurate prediction and precise tracing of corporate financial and tax risks, thereby improving the effectiveness of risk management.
Smart Images
Figure CN121504640B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of financial and tax risk management, specifically to a vector-driven intelligent financial and tax management method, system, and storage medium. Background Technology
[0002] Financial and tax management, especially risk forecasting, is directly related to the survival bottom line of enterprises, which can prevent them from going bankrupt due to cash flow disruption or insolvency. At the same time, by optimizing resource allocation and ensuring sound operation, it provides key basis for strategic decision-making, thereby improving the enterprise's operating efficiency, strategic flexibility and overall value, and ultimately enhancing the enterprise's resilience and sustainable growth capabilities in a complex and ever-changing market environment.
[0003] Traditional fiscal and tax risk prediction models are mostly based on static fiscal and tax indicator ratios (such as debt-to-equity ratio and current ratio), which make it difficult to capture external dynamic factors such as market sentiment, supply chain changes, and public opinion in real time. Moreover, these models are mostly "black boxes," and the prediction results lack interpretability and traceability, resulting in lagging risk control measures and insufficient decision support, which in turn leads to poor performance in fiscal and tax management. Summary of the Invention
[0004] To address the technical problem that traditional financial and tax risk prediction results lack interpretability and traceability, leading to poor effectiveness in financial and tax management, this invention aims to provide a vector-driven intelligent financial and tax management method, system, and storage medium. The specific technical solution adopted is as follows: This invention proposes a vector-driven intelligent financial and tax management method, the method comprising: Obtain internal financial and tax data of the enterprise within a preset first time period prior to the current time point, and external public opinion data of the enterprise within a preset second time period; All the financial and tax data within the preset first time period are vectorized to obtain the financial and tax status vector of the enterprise at the current time point; all the public opinion data within the preset second time period are vectorized to obtain the public opinion environment vector of the enterprise at the current time point. The tax and financial status vector and the public opinion environment vector are concatenated to obtain the joint vector of the enterprise at the current time point; the joint vector is processed using an attention mechanism to obtain the attention weight of each dimension of the joint vector at the current time point; based on the joint vector at the current time point and the attention weight of each dimension, the fusion risk representation vector at the current time point is obtained. Based on the fusion risk representation vector at the current time point, and combined with the attention weight of each dimension, the financial and tax risks of enterprises are predicted, and a risk report is generated.
[0005] Furthermore, the vector representing the enterprise's financial and tax status at the current point in time includes: Standardize and preprocess all the financial and tax data information within the preset first time period to obtain the standard financial and tax values of different financial and tax indicators at each time point within the preset first time period. The matrix formed by the standard financial and tax values of all financial and tax indicators at all time points within the preset first time period is used as the time series feature matrix of the current time point. The rows of the time series feature matrix represent financial and tax indicators, the columns represent time points, and the elements in the time series feature matrix are the standard financial and tax values. The temporal feature matrix at the current time point is input into the Long Short-Term Memory (LSTM) network, and the hidden state output by the LSTM network is used as the enterprise's financial and tax status vector at the current time point.
[0006] Furthermore, the vector of public opinion environment of the enterprise at the current point in time includes: Natural language processing is performed on all the public opinion data information within the preset second time period to obtain multiple structured records of public opinion within the preset second time period, wherein the structured records of public opinion include public opinion values of different public opinion indicators; The public opinion values of the same public opinion indicator in all the structured records of public opinion within the preset second time period are statistically aggregated to obtain the public opinion environment vector of the enterprise at the current time point, wherein each dimension of the public opinion environment vector corresponds to a public opinion indicator.
[0007] Furthermore, obtaining the joint vector of the enterprises at the current time point includes: By concatenating the financial and tax status vector and the public opinion environment vector, a joint vector of the enterprise at the current point in time can be obtained.
[0008] Furthermore, the attention weights for each dimension of the joint vector obtained at the current time point include: The attention mechanism includes a weight matrix parameter and a bias vector parameter. The product of the weight matrix parameter and the joint vector at the current time point is used as the initial attention vector at the current time point. The sum of the initial attention vector and the bias vector parameter is used as the attention score vector at the current time point. The element values of each dimension of the attention score vector are normalized to obtain the attention weight of each dimension of the joint vector at the current time point, wherein the sum of the attention weights of all dimensions is equal to the value 1.
[0009] Furthermore, the method for obtaining the fusion risk representation vector at the current time point includes: The product of the element value of each dimension of the joint vector at the current time point and the attention weight is used as the element value of each dimension of the fusion risk representation vector at the current time point, thereby obtaining the fusion risk representation vector at the current time point.
[0010] Furthermore, the prediction of the enterprise's financial and tax risks and the generation of risk reports include: The fused risk representation vector at the current time point is input into a fully connected neural network classifier, which outputs the probability values of different risk levels for the enterprise at the current time point. The correlation between each dimension and each financial and tax indicator of the financial and tax status vector is analyzed to obtain the relevant financial and tax indicators for each dimension of the financial and tax status vector. In the joint vector at the current time point, select the first number of dimensions with the largest attention weight as key dimensions. Among all key dimensions, the key dimensions belonging to the fiscal and tax status vector are selected as fiscal and tax key dimensions, and the key dimensions belonging to the public opinion environment vector are selected as public opinion key dimensions. The relevant financial and tax indicators and attention weights corresponding to each key financial and tax dimension, as well as the public opinion indicators and attention weights corresponding to each key public opinion dimension, are filled into the report template to generate a risk report.
[0011] Furthermore, the relevant financial and tax indicators for each dimension of the obtained financial and tax status vector include: Take any dimension of the fiscal and tax status vector as the target dimension, and use the Pearson correlation coefficient over time between the element value of the target dimension of the fiscal and tax status vector at historical time points and the standard fiscal and tax value of each fiscal and tax indicator as the correlation coefficient between the target dimension of the fiscal and tax status vector and each fiscal and tax indicator. For the target dimension of the fiscal and tax status vector, the second number of fiscal and tax indicators with the highest correlation coefficient are selected as the relevant fiscal and tax indicators for the target dimension of the fiscal and tax status vector.
[0012] The present invention also proposes a vector-driven intelligent financial and tax management system, the system comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement any one of the steps of a vector-driven intelligent financial and tax management method.
[0013] The present invention also proposes a computer storage medium storing a computer program that is executed to implement any of the steps of a vector-driven intelligent financial and tax management method.
[0014] The present invention has the following beneficial effects: This invention addresses the challenge of traditional financial and tax risk prediction models in capturing dynamic external market factors in real time, resulting in a lack of interpretability and traceability in predictions and consequently poor effectiveness in financial and tax management. Therefore, this invention first collects internal financial and tax data from a pre-defined first time period prior to the current point in time, and external public opinion data from a pre-defined second time period. This multi-source data can then be used to accurately predict the company's financial and tax risks and trace their origins. Furthermore, to facilitate subsequent neural network processing, both financial and tax data and public opinion data are vectorized, transforming these two fundamentally different data types into a single data format, thus facilitating subsequent vector fusion and... The model calculates and then concatenates the financial and tax status vector and the public opinion environment vector to obtain the joint vector at the current time point. This provides complete contextual information for the subsequent attention mechanism, allowing it to simultaneously see all characteristics of the internal financial and tax status and the external market public opinion. The acquired attention weights reflect the importance of each dimension's information to risk assessment, improving the accuracy of subsequent risk prediction and risk tracing. The attention weights of each dimension are then applied to the joint vector at the current time point. Based on the acquired fused risk representation vector, combined with the attention weights of each dimension, the financial and tax risks of the enterprise are predicted, and a risk report is generated. This achieves accurate prediction of the enterprise's financial and tax risks while also enabling precise tracing of the enterprise's financial and tax risks. Attached Figure Description
[0015] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention 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 only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0016] Figure 1 This is a flowchart of a vector-driven intelligent financial and tax management method provided in one embodiment of the present invention. Detailed Implementation
[0017] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a vector-driven intelligent financial and tax management method, system, and storage medium proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.
[0018] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.
[0019] The following description, in conjunction with the accompanying drawings, details a specific solution for a vector-driven intelligent financial and tax management method, system, and storage medium provided by this invention.
[0020] Please see Figure 1 The diagram illustrates a flowchart of a vector-driven intelligent financial and tax management method according to an embodiment of the present invention, the method comprising: Step S1: Obtain the company's internal financial and tax data for the first time period prior to the current time point and the external public opinion data for the second time period.
[0021] Enterprises typically use financial and tax management systems, such as ERP systems, to manage all company resources and business processes. Therefore, in this embodiment of the invention, the enterprise's internal financial and tax data information within a preset first time period prior to the current time point is first retrieved from the financial and tax management system. The preset first time period typically ranges from 12 to 24 months, with each month as the unit of time. In one embodiment of the invention, the preset first time period is set to 12 months. The specific value of the preset first time period can also be set by the implementer according to the specific implementation scenario, and is not limited here. The financial and tax data information is usually presented in tabular form and contains a large number of financial and tax indicators.
[0022] Since the external market sentiment environment can significantly impact a company's financial and tax situation—for example, negative news about a company in external media can directly affect its tax revenue—this embodiment of the invention also requires the use of web crawlers and other methods to collect external public opinion data from news media, social media, and related websites within a preset second time period. The preset second time period typically ranges from 1 to 3 months. In one embodiment of the invention, the preset second time period is set to 1 month. The specific value of the preset second time period can also be set by the implementer according to the specific implementation scenario and is not limited here. The public opinion data is typically webpage text data.
[0023] Step S2: Vectorize all financial and tax data within the preset first time period to obtain the financial and tax status vector of the enterprise at the current time point; vectorize all public opinion data within the preset second time period to obtain the public opinion environment vector of the enterprise at the current time point.
[0024] Typically, a company's internal financial and tax data contains a large number of financial and tax indicators, including not only basic indicators such as operating revenue, net profit, return on assets, and current ratio, but also derived indicators such as year-on-year and month-on-month comparisons, like the year-on-year growth rate of revenue and the month-on-month growth rate of net profit. In contrast, external public opinion data is mostly text data. The two data formats are different. To facilitate subsequent neural network processing, financial and tax data and public opinion data are further vectorized, thereby transforming these two completely different types of data into the same data format, which is convenient for subsequent vector fusion and model calculation.
[0025] In this embodiment of the invention, all financial and tax data within a preset first time period are first vectorized to obtain the financial and tax status vector of the enterprise at the current time point. This allows the numerous financial and tax indicators within the enterprise to be quantified into a fixed-length vector that contains rich historical information and can reflect the enterprise's financial and tax status.
[0026] Preferably, in one embodiment of the present invention, the method for obtaining the enterprise's financial and tax status vector at the current point in time specifically includes: First, financial and tax data are generally presented in tabular form, which contains financial and tax values of different financial and tax indicators at different points in time. Since the dimensions of different financial and tax indicators are different, in order to eliminate the influence of dimensions, this embodiment of the invention first performs standardization preprocessing on all financial and tax data within a preset first time period to obtain standard financial and tax values of different financial and tax indicators at each point in time within the preset first time period. The standardization preprocessing of data is a technical means well known to those skilled in the art and will not be described in detail here.
[0027] Then, the matrix composed of the standard fiscal and tax values of all fiscal and tax indicators at all time points within the preset first time period is used as the time series feature matrix of the current time point. In the time series feature matrix, the rows represent fiscal and tax indicators, the columns represent time points, and the elements in the time series feature matrix are standard fiscal and tax values.
[0028] The time-series feature matrix at the current time point is input into a Long Short-Term Memory (LSTM) network, and the hidden state output by the LSTM network is used as the enterprise's financial and tax status vector at the current time point. The LSTM network reads the data in the time-series feature matrix step by step, learns the change pattern of individual indicators and the correlation between indicators, and updates its internal hidden state. The final hidden state output by the LSTM network contains all the important information of the enterprise's financial and tax data within the preset first time period. The LSTM network is a well-known technique in the art and will not be described in detail here. In other embodiments of the present invention, neural network models such as Transformer can also be used to process the time-series feature matrix at the current time point, which is not limited here.
[0029] It should be noted that each dimension of the financial and tax status vector obtained after vectorizing financial and tax data does not have a clear business meaning. Each dimension does not correspond to a specific financial and tax indicator, but is an abstract representation of the company's overall financial and tax situation that is learned by the model. All dimensions together represent the company's overall financial and tax situation.
[0030] Since the public opinion environment in the external market directly affects the financial and tax status of enterprises, this embodiment of the invention also needs to vectorize all public opinion data information within the preset second time period to obtain the public opinion environment vector of the enterprise at the current time point. Subsequently, the financial and tax status vector and the public opinion environment vector can be combined to accurately predict and precisely trace the financial and tax risks of enterprises.
[0031] Preferably, in one embodiment of the present invention, the method for obtaining the public opinion environment vector of an enterprise at the current point in time specifically includes: First, since public opinion data is presented in text form, natural language processing can be performed on all public opinion data within the preset second time period to obtain multiple structured records of public opinion within the preset second time period. Among them, the structured records of public opinion include public opinion values of different public opinion indicators. Public opinion indicators include, for example, sentiment scores, word frequency of specific keywords (such as "supply chain" and "management"), number of negative news items, etc. Natural language processing is a technical means well known to those skilled in the art, and will not be elaborated here.
[0032] Then, statistical aggregation is performed on the public opinion values of the same public opinion indicator in all structured records within the preset second time period to obtain the public opinion environment vector of the enterprise at the current time point. Each dimension of the public opinion environment vector corresponds to a public opinion indicator. The specific process is as follows: one or more aggregation functions are preset. For a certain public opinion indicator, a specific aggregation function is used to calculate the public opinion value of that indicator in all structured records within the preset second time period. The resulting function value of that public opinion indicator is then used as the value of one dimension of the public opinion environment vector. For example, the set aggregation function can be an averaging function. After the sentiment score in the structured records is processed using this function, the mean of the sentiment scores of all structured records within the preset second time period can be obtained, and this mean is used as the value of one dimension of the public opinion environment vector.
[0033] It should be noted that, unlike the fiscal and tax status vector, the dimensions in the public opinion environment vector are specific and interpretable, with each dimension having a specific meaning. For example, a lower average sentiment score for a certain dimension in the public opinion environment vector indicates a lower overall sentiment in the external market, while a higher average word frequency for supply chain keywords for a certain dimension in the public opinion environment vector indicates a higher level of public attention and discussion about the overall supply chain. This feature can be used to accurately trace the external influencing factors of corporate financial and tax risks.
[0034] Step S3: Concatenate the financial and tax status vector and the public opinion environment vector to obtain the joint vector of the enterprise at the current time point; process the joint vector using an attention mechanism to obtain the attention weight of each dimension of the joint vector at the current time point; obtain the fusion risk representation vector at the current time point based on the joint vector at the current time point and the attention weight of each dimension.
[0035] The above process completes the vectorization of financial and tax information and public opinion information. Since the financial and tax risks of an enterprise are not only affected by its own various financial and tax indicators, but also by the external market public opinion, it is not possible to predict and trace financial and tax risks solely through the enterprise's internal financial and tax information. It is also necessary to combine the external market public opinion environment factors. Therefore, this embodiment of the invention concatenates the financial and tax status vector and the public opinion environment vector to obtain the joint vector of the enterprise at the current time point, providing complete contextual information for the subsequent attention mechanism, allowing it to see all the characteristics of the internal financial and tax status and the external market public opinion at the same time, thereby improving the accuracy of enterprise financial and tax risk prediction and the precision of risk tracing.
[0036] Preferably, in one embodiment of the present invention, the method for obtaining the joint vector of an enterprise at the current time point specifically includes: By connecting the financial and tax status vector and the public opinion environment vector end to end, we can obtain the joint vector of the enterprise at the current time point. For example, if A represents the financial and tax status vector and B represents the public opinion environment vector, then the joint vector of the enterprise at the current time point is C = (A, B), or it can also be expressed as C = (B, A), without limitation here.
[0037] Since each dimension of the joint vector has different levels of importance for enterprise risk prediction, the attention mechanism enables the model to focus on the signal most relevant to the current risk, effectively filtering out noise and thus making a more accurate judgment. Therefore, this embodiment of the invention introduces an attention mechanism to process the joint vector and obtain the attention weight of each dimension of the joint vector at the current time point. The larger the attention weight of a certain dimension, the greater the importance of that dimension feature in the prediction of financial and tax risks. Subsequently, the attention weight can be combined to complete the accurate tracing of enterprise financial and tax risks.
[0038] Preferably, in one embodiment of the present invention, the method for obtaining the attention weights of each dimension of the joint vector at the current time point specifically includes: The attention mechanism includes weight matrix parameters and bias vector parameters. These two parameters can be obtained by training a deep learning model that uses the attention mechanism (such as the Transformer model). The training process of the attention network model is its self-learning process. The product of the weight matrix parameters and the joint vector at the current time point is used as the initial attention vector at the current time point. The sum of the initial attention vector and the bias vector parameters is used as the attention score vector at the current time point. The larger the element value of each dimension in the attention score vector, the more important that dimension feature is in the prediction of corporate financial and tax risks.
[0039] Then, the element values of each dimension of the attention score vector are normalized to limit the calculation results to... Within the range, the attention weights of each dimension of the joint vector at the current time point are obtained, wherein the sum of the attention weights of all dimensions is equal to the value 1. In one embodiment of the present invention, the normalization of the element values of each dimension of the attention score vector can be implemented using the Softmax function or other related functions, which is not limited here.
[0040] After obtaining the joint vector at the current time point and the attention weight of each dimension, the fusion risk representation vector at the current time point can be obtained based on the joint vector at the current time point and the attention weight of each dimension. Subsequently, the financial and tax risks of enterprises can be accurately predicted based on the fusion risk representation vector at the current time point.
[0041] Preferably, in one embodiment of the present invention, the method for obtaining the fusion risk representation vector at the current time point specifically includes: The product of the element value of each dimension of the joint vector at the current time point and the attention weight is used as the element value of each dimension of the fusion risk representation vector at the current time point, thereby obtaining the fusion risk representation vector at the current time point.
[0042] Among them, the fused risk representation vector is no longer a simple accumulation of raw data, but a vector that has been intelligently weighted and highlights key information, which can improve the accuracy of subsequent risk prediction.
[0043] Step S4: Based on the fusion risk representation vector at the current time point and combined with the attention weight of each dimension, predict the financial and tax risks of the enterprise and generate a risk report.
[0044] By obtaining the fusion risk representation vector at the current time point and the attention weight of each dimension, the fusion risk representation vector at the current time point can be input into the neural network model to predict the financial and tax risks of enterprises. At the same time, by combining the attention weight of each dimension, the enterprise risks can be accurately traced back to their source, thereby generating a risk report.
[0045] Preferably, in one embodiment of the present invention, the method for predicting the financial and tax risks of an enterprise and generating a risk report specifically includes: The fused risk representation vector at the current time point is input into a fully connected neural network classifier, which outputs the probability values of different risk levels for the enterprise at the current time point. In one embodiment of the present invention, the risk levels are divided into three levels: low risk, medium risk, and high risk. If the output of the fully connected neural network classifier is (0.1, 0.2, 0.7), it means that the probability of the enterprise's financial and tax situation being low is 0.1, the probability of medium risk is 0.2, and the probability of high risk is 0.7. This indicates that the enterprise's financial and tax situation is likely to be high, and the enterprise needs to make timely adjustments.
[0046] When enterprises make adjustments to their business or management to address risks, they need to identify the sources of the risks in order to determine the specific direction and details of the adjustments and make the best response decisions. Factors that cause financial and tax risks to enterprises may come from the enterprise's own financial and tax situation, or from the influence of public opinion in the external market, or a combination of both. Therefore, it is also necessary to trace the sources of the factors that cause enterprise risks.
[0047] Unlike the public opinion environment vector, each dimension of the financial and tax status vector represents an abstract feature, and each dimension itself does not have a clear business meaning. Therefore, it is necessary to analyze the correlation between each dimension of the financial and tax status vector and each financial and tax indicator to obtain the relevant financial and tax indicators for each dimension of the financial and tax status vector, so as to achieve accurate tracing of the internal source of corporate financial and tax risks.
[0048] Preferably, in one embodiment of the present invention, the method for obtaining the relevant financial and tax indicators for each dimension of the financial and tax status vector specifically includes: Using any dimension of the fiscal and tax status vector as the target dimension, the Pearson correlation coefficient over time between the element values of the target dimension of the fiscal and tax status vector at historical time points and the standard fiscal and tax values of each fiscal and tax indicator is used as the correlation coefficient between the target dimension of the fiscal and tax status vector and each fiscal and tax indicator. Specifically, the element values of the target dimension of the fiscal and tax status vector at each historical time point are arranged in chronological order to obtain a time series sequence of the target dimension of the fiscal and tax status vector. Then, the standard fiscal and tax values of each fiscal and tax indicator at each historical time point are arranged to obtain a time series sequence of each fiscal and tax indicator. The Pearson correlation coefficient between the time series sequence of the target dimension of the fiscal and tax status vector and the time series sequence of each fiscal and tax indicator is then used as the correlation coefficient between the target dimension of the fiscal and tax status vector and each fiscal and tax indicator. In other embodiments of this invention, existing Spearman rank correlation coefficients or Kendall rank correlation coefficients can also be used to achieve correlation analysis, which is not limited here.
[0049] For the target dimension of the fiscal and tax status vector, the second preset number of fiscal and tax indicators with the highest correlation coefficient are used as the relevant fiscal and tax indicators of the target dimension of the fiscal and tax status vector. The relevant fiscal and tax indicators of each dimension of the fiscal and tax status vector can be obtained by the same method described above. The value of the second preset number is in the range of 2 to 5. In one embodiment of the present invention, the second preset number is set to 3, that is, the three fiscal and tax indicators with the highest correlation coefficient are used as the relevant fiscal and tax indicators of the target dimension of the fiscal and tax status vector. The specific value of the second preset number can also be set by the implementer according to the specific implementation scenario, and is not limited here.
[0050] In the current time point, select the first number of dimensions with the largest attention weight from all dimensions of the joint vector as key dimensions. Among all key dimensions, the key dimensions belonging to the fiscal and tax status vector are selected as fiscal and tax key dimensions, and the key dimensions belonging to the public opinion environment vector are selected as public opinion key dimensions. Generally speaking, the dimension of the joint vector is more than 100, so the value range of the first number is usually kept between 20 and 50. In one embodiment of the present invention, the first number is set to 30. The specific value of the first number can also be set by the implementer according to the specific implementation scenario, and is not limited here.
[0051] By filling the relevant financial and tax indicators and attention weights corresponding to each key financial and tax dimension, as well as the public opinion indicators and attention weights corresponding to each key public opinion dimension, into the report template, a risk report is generated. The risk report clearly presents various internal factors (i.e., relevant financial and tax indicators corresponding to the key financial and tax dimensions) and external factors (i.e., public opinion indicators corresponding to the key public opinion dimensions) related to this corporate financial and tax risk phenomenon, as well as the importance of each factor (i.e., attention weight), thereby achieving accurate tracing of corporate financial and tax risks and improving the effectiveness of corporate financial and tax risk management.
[0052] One embodiment of the present invention provides a vector-driven intelligent financial and tax management system. The system includes a memory, a processor, and a computer program. The memory is used to store the corresponding computer program, and the processor is used to run the corresponding computer program. When the computer program runs in the processor, it can implement the methods described in steps S1 to S4.
[0053] One embodiment of the present invention provides a computer storage medium storing a computer program that, when executed, can implement the methods described in steps S1 to S4.
[0054] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.
[0055] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.
Claims
1. A vector-driven intelligent financial and tax management method, characterized in that, The method includes: Obtain internal financial and tax data of the enterprise within a preset first time period prior to the current time point, and external public opinion data of the enterprise within a preset second time period; Vectorization processing is performed on all the financial and tax data information within a preset first time period to obtain the financial and tax status vector of the enterprise at the current time point. Obtaining the financial and tax status vector of the enterprise at the current time point includes: standardizing and preprocessing all the financial and tax data information within the preset first time period to obtain standard financial and tax values for different financial and tax indicators at each time point within the preset first time period; using the matrix composed of the standard financial and tax values of all financial and tax indicators at all time points within the preset first time period as the temporal feature matrix for the current time point, wherein the rows of the temporal feature matrix represent financial and tax indicators, the columns represent time points, and the elements in the temporal feature matrix are the standard financial and tax values; inputting the temporal feature matrix for the current time point into a Long Short-Term Memory (LSTM) network, and using the hidden state finally output by the LSTM network as the financial and tax status vector of the enterprise at the current time point. All the public opinion data information within the preset second time period is vectorized to obtain the public opinion environment vector of the enterprise at the current time point; The tax and financial status vector and the public opinion environment vector are concatenated to obtain the joint vector of the enterprise at the current time point; the joint vector is processed using an attention mechanism to obtain the attention weight of each dimension of the joint vector at the current time point; based on the joint vector at the current time point and the attention weight of each dimension, the fusion risk representation vector at the current time point is obtained. Based on the fused risk representation vector at the current time point, and combined with the attention weights of each dimension, the financial and tax risks of the enterprise are predicted, and a risk report is generated; the prediction of the enterprise's financial and tax risks and the generation of a risk report include: The fused risk representation vector at the current time point is input into a fully connected neural network classifier, which outputs the probability values of different risk levels for the enterprise at the current time point. The correlation between each dimension of the fiscal and tax status vector and each fiscal and tax indicator is analyzed to obtain the relevant fiscal and tax indicators for each dimension of the fiscal and tax status vector. Obtaining the relevant fiscal and tax indicators for each dimension of the fiscal and tax status vector includes: taking any dimension of the fiscal and tax status vector as the target dimension, and using the Pearson correlation coefficient over time between the element values of the target dimension of the fiscal and tax status vector at historical time points and the standard fiscal and tax values of each fiscal and tax indicator as the correlation coefficient between the target dimension of the fiscal and tax status vector and each fiscal and tax indicator. Specifically, the process involves: following the chronological order, analyzing the target dimension of the fiscal and tax status vector at each historical time point... The element values of the dimension are arranged to obtain a time series sequence of the target dimension of the fiscal and tax status vector. Then, the standard fiscal and tax values of each fiscal and tax indicator at each historical time point are arranged to obtain a time series sequence of each fiscal and tax indicator. The Pearson correlation coefficient between the time series sequence of the target dimension of the fiscal and tax status vector and the time series sequence of each fiscal and tax indicator is used as the correlation coefficient between the target dimension of the fiscal and tax status vector and each fiscal and tax indicator. For the target dimension of the fiscal and tax status vector, the second preset number of fiscal and tax indicators with the largest correlation coefficients are used as the relevant fiscal and tax indicators of the target dimension of the fiscal and tax status vector. In the joint vector at the current time point, select the first number of dimensions with the largest attention weight as key dimensions. Among all key dimensions, the key dimensions belonging to the fiscal and tax status vector are selected as fiscal and tax key dimensions, and the key dimensions belonging to the public opinion environment vector are selected as public opinion key dimensions. The relevant financial and tax indicators and attention weights corresponding to each key financial and tax dimension, as well as the public opinion indicators and attention weights corresponding to each key public opinion dimension, are filled into the report template to generate a risk report.
2. The vector-driven intelligent financial and tax management method according to claim 1, characterized in that, The vector of the company's public opinion environment at the current point in time includes: Natural language processing is performed on all the public opinion data information within the preset second time period to obtain multiple structured records of public opinion within the preset second time period, wherein the structured records of public opinion include public opinion values of different public opinion indicators; The public opinion values of the same public opinion indicator in all the structured records of public opinion within the preset second time period are statistically aggregated to obtain the public opinion environment vector of the enterprise at the current time point, wherein each dimension of the public opinion environment vector corresponds to a public opinion indicator.
3. The vector-driven intelligent financial and tax management method according to claim 1, characterized in that, The joint vector of the enterprises at the current time point includes: By concatenating the financial and tax status vector and the public opinion environment vector, a joint vector of the enterprise at the current point in time can be obtained.
4. The vector-driven intelligent financial and tax management method according to claim 1, characterized in that, The attention weights for each dimension of the joint vector obtained at the current time point include: The attention mechanism includes a weight matrix parameter and a bias vector parameter. The product of the weight matrix parameter and the joint vector at the current time point is used as the initial attention vector at the current time point. The sum of the initial attention vector and the bias vector parameter is used as the attention score vector at the current time point. The element values of each dimension of the attention score vector are normalized to obtain the attention weight of each dimension of the joint vector at the current time point, wherein the sum of the attention weights of all dimensions is equal to the value 1.
5. The vector-driven intelligent financial and tax management method according to claim 1, characterized in that, The obtained fusion risk representation vector at the current time point includes: The product of the element value of each dimension of the joint vector at the current time point and the attention weight is used as the element value of each dimension of the fusion risk representation vector at the current time point, thereby obtaining the fusion risk representation vector at the current time point.
6. A vector-driven intelligent financial and tax management system, the system comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements a vector-driven intelligent financial and tax management method as described in any one of claims 1-5.
7. A computer storage medium, characterized in that, The storage medium stores a computer program that is executed to implement a vector-driven intelligent financial and tax management method as described in any one of claims 1-5.