Enterprise tax risk prediction method and device, computer device, and storage medium

Abstract

This invention discloses a method, apparatus, computer equipment, and storage medium for predicting corporate tax risks. The method includes: acquiring tax data of the enterprise to be predicted; performing feature engineering processing on the tax data to obtain a processing result; inputting the processing result into a prediction model to predict a risk value, thereby obtaining a prediction result; determining a risk level based on the prediction result and a set risk threshold; and outputting the risk level. The prediction model is obtained by training a multi-task learning network using a full sample set consisting of a risk sample dataset and a non-risk sample dataset. By implementing the method of this invention, the ability to identify tax risks can be improved, and the accuracy and comprehensiveness of predictions can be enhanced.

Description

Technical Field

[0001] This invention relates to methods for predicting tax risks, and more specifically to methods, apparatus, computer equipment, and storage media for predicting corporate tax risks. Background Technology

[0002] Current technologies for predicting corporate tax risks primarily rely on traditional financial auditing and manual inspection methods. However, these methods have significant drawbacks that affect the effectiveness and accuracy of risk assessment. Specifically, traditional methods typically depend on self-reported financial data from companies, which may be inaccurate or incomplete. Furthermore, these methods often overlook publicly available external information, such as business registration data, legal records, and industry trends. This lack of information integration makes it difficult to comprehensively understand a company's financial situation during risk assessment, potentially leading to biased results. Manual auditing processes are typically time-consuming and costly, making it difficult to monitor and assess the tax risks of a large number of companies in real time. This inefficiency makes it difficult for auditors to identify potential problems in a timely manner, potentially missing crucial warning signs before tax risks materialize. Traditional auditing methods have limitations in identifying potential tax risks in non-financial data. For example, related-party transactions and unusual changes in senior management may indicate tax risks, but traditional methods struggle to effectively mine this data, thus overlooking important clues and affecting the overall accuracy of risk assessment. Current risk assessment methods lack the effective application of big data and intelligent algorithms, failing to automatically extract and identify valuable information from massive amounts of data. This technological deficiency limits the ability to identify risks, resulting in the inability to detect and respond to potential tax risks in a timely manner.

[0003] In summary, existing technologies for predicting corporate tax risks suffer from problems such as limited data sources, low efficiency, insufficient accuracy, and lack of technological application. These factors collectively affect the comprehensiveness and effectiveness of risk assessment.

[0004] Therefore, it is necessary to develop a new method to improve the ability to identify tax risks and enhance the accuracy and comprehensiveness of forecasts. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method, apparatus, computer equipment and storage medium for predicting corporate tax risks.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a method for predicting corporate tax risks, comprising:

[0007] Obtain tax data for companies to be predicted;

[0008] The tax data of the enterprise to be predicted is subjected to feature engineering to obtain the processing result;

[0009] The processing results are input into the prediction model to predict the risk value, so as to obtain the prediction result;

[0010] The risk level is determined based on the prediction results and the set risk threshold.

[0011] Output the aforementioned risk level;

[0012] The prediction model is obtained by training a multi-task learning network using a full sample set consisting of risky sample datasets and non-risk sample datasets.

[0013] The further technical solution is as follows: the process of performing feature engineering on the tax data of the enterprise to be predicted to obtain the processing result includes:

[0014] The tax data of the enterprise to be predicted is converted into continuous features (WOE) to obtain the transformation result;

[0015] Named entity recognition is performed on the tax data of the enterprise to be predicted in order to obtain key information;

[0016] The tax data of the enterprise to be predicted is processed according to set rules to obtain the processing result;

[0017] The conversion results, key information, and processing results are integrated into a processing result.

[0018] The further technical solution is as follows: Named entity recognition is performed on the tax data of the enterprise to be predicted to obtain key information, including:

[0019] The tax data of the enterprises to be predicted are used to perform named entity recognition using a deep learning model based on BiLSTM-CRF to obtain key information.

[0020] The further technical solution is as follows: the prediction model includes an input layer, a feature fusion layer, a gating network layer, an expert network layer, and an output layer.

[0021] The further technical solution is as follows: inputting the processing result into the prediction model to predict the risk value, so as to obtain the prediction result, includes:

[0022] The processing results are then input into the prediction model;

[0023] The processing results are then fused to obtain a fused result;

[0024] Based on the fusion result, the gating network layer determines the expert network layer to participate in the processing.

[0025] The fusion results are processed using a defined expert network layer to determine the type of tax risk;

[0026] The tax risk types output by the expert network are integrated, and the predicted probability of tax risk is calculated through the Softmax layer to obtain the prediction result.

[0027] The further technical solution is as follows: the step of determining the expert network layer to participate in the processing based on the fusion result by the gated network layer includes:

[0028] The gated network layer assigns tasks to the corresponding expert network layers based on the characteristics of the fusion result, wherein each expert network layer focuses on a certain type of tax risk.

[0029] The present invention also provides a corporate tax risk prediction device, comprising:

[0030] The data acquisition unit is used to acquire tax data of the enterprise to be predicted.

[0031] The feature engineering processing unit is used to perform feature engineering processing on the tax data of the enterprise to be predicted in order to obtain the processing result;

[0032] The prediction unit is used to input the processing results into the prediction model to predict the risk value, so as to obtain the prediction result;

[0033] A risk level determination unit is used to determine the risk level based on the prediction results and a set risk threshold.

[0034] An output unit is used to output the risk level;

[0035] The prediction model is obtained by training a multi-task learning network using a full sample set consisting of risky sample datasets and non-risk sample datasets.

[0036] The further technical solution is as follows: the feature engineering processing unit includes:

[0037] The transformation subunit is used to perform continuous feature WOE conversion on the tax data of the enterprise to be predicted in order to obtain the transformation result;

[0038] The identification subunit is used to perform named entity recognition on the tax data of the enterprise to be predicted in order to obtain key information;

[0039] The processing subunit is used to process the tax data of the enterprise to be predicted according to set rules to obtain the processing result;

[0040] The integration subunit is used to integrate the conversion results, key information, and processing results into a processing result.

[0041] The present invention also provides a computer device, the computer device including a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the above-described method.

[0042] The present invention also provides a storage medium storing a computer program that, when executed by a processor, implements the above-described method.

[0043] The advantages of this invention compared to existing technologies are as follows: After acquiring the tax data of the enterprise to be predicted, this invention performs feature engineering to extract key features. Then, the processing results are input into a prediction model to predict risk values, ensuring that the model, trained through a multi-task learning network on a full sample set, possesses stronger recognition capabilities. Based on the prediction results and a set risk threshold, the risk level is determined and finally output. This process not only improves the ability to identify tax risks but also enhances the accuracy and comprehensiveness of predictions.

[0044] The present invention will be further described below with reference to the accompanying drawings and specific embodiments. Attached Figure Description

[0045] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0046] Figure 1 This is a schematic diagram illustrating an application scenario of the enterprise tax risk prediction method provided in this embodiment of the invention.

[0047] Figure 2 A flowchart illustrating the enterprise tax risk prediction method provided in this embodiment of the invention;

[0048] Figure 3 A schematic block diagram of an enterprise tax risk prediction device provided in an embodiment of the present invention;

[0049] Figure 4 A schematic block diagram of a computer device provided for an embodiment of the present invention. Detailed Implementation

[0050] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0051] It should be understood that, when used in this specification and the appended claims, the terms "comprising" and "including" indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0052] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0053] It should also be further understood that the term "and / or" as used in this specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0054] Please see Figure 1 and Figure 2 , Figure 1 This is a schematic diagram illustrating an application scenario of the enterprise tax risk prediction method provided in an embodiment of the present invention. Figure 2 This is a schematic flowchart illustrating the enterprise tax risk prediction method provided in this embodiment of the invention. The method is applied in a server. The server interacts with the terminal to collect tax data of the enterprise to be predicted, ensuring the data's integrity and representativeness; it performs feature engineering on the data, including WOE transformation of continuous features and named entity recognition, to extract key information; it sets rules to process the data, forming a multi-dimensional feature set to enrich the input information; the processing results are input into a multi-task learning prediction model to improve recognition capabilities; a gating network layer is used in the model to select appropriate expert networks and perform targeted processing according to task characteristics; the outputs of each expert network are integrated, and a Softmax layer is used to calculate the tax risk prediction probability; different risk levels are output according to a set risk threshold to ensure comprehensiveness; and the prediction accuracy and model adaptability are improved by continuously updating the sample dataset and model training.

[0055] Figure 2 This is a flowchart illustrating the enterprise tax risk prediction method provided in an embodiment of the present invention. Figure 2 As shown, the method includes the following steps S110 to S150.

[0056] S110. Obtain tax data for the company to be predicted.

[0057] In this embodiment, the tax data of the enterprise to be predicted typically includes the following aspects:

[0058] Financial statement data, such as the balance sheet, income statement, and cash flow statement, reflects a company's financial condition and operating results.

[0059] Tax filing information: This includes tax returns for previous years, tax amounts, tax types, and filing dates.

[0060] Transaction records: Includes detailed transaction information such as sales, purchases, and expenses, which helps analyze the compliance and rationality of transactions.

[0061] Industry characteristics: Information such as tax policies and average tax burden of the industry in which the company operates, to facilitate industry comparison.

[0062] Basic company information, such as registered capital, business scope, shareholder structure, and management information, affects the company's tax risks.

[0063] Historical audit records: Previous audit results and rectification status can help identify potential tax risks.

[0064] Compliance data includes the company's compliance management measures, internal control system and its implementation.

[0065] External environmental factors: such as economic conditions, policy changes, market dynamics, etc., which may affect tax risks.

[0066] S120. Perform feature engineering on the tax data of the enterprise to be predicted to obtain the processing result.

[0067] In this embodiment, the processing result refers to the comprehensive data output obtained after performing feature engineering on the tax data of the enterprise to be predicted. Specifically, it includes the following parts:

[0068] WOE transformation results: Continuous features transformed by WOE (Weight of Evidence) can better reflect the relationship between variables and target variables, thus improving the predictive ability of the model.

[0069] Key information: Useful information extracted through named entity recognition, such as company name, tax terms, and transaction type, helps identify important factors affecting tax risks.

[0070] Processing results: Features obtained after processing data based on set rules, such as tax indicators classified or aggregated according to specific standards, can reveal trends and patterns in the data.

[0071] Ultimately, these processing results are integrated to form a rich feature set, providing high-quality input data for subsequent predictive models, thereby improving the accuracy and effectiveness of tax risk prediction.

[0072] In one embodiment, step S120 described above may include steps S121 to S124.

[0073] S121. Perform continuous feature WOE conversion on the tax data of the enterprise to be predicted to obtain the conversion result.

[0074] In this embodiment, the conversion result refers to the result obtained after the data continuous feature is converted to WOE.

[0075] Specifically, WOE (Weight of Evidence) processing is a method for handling continuous features, aiming to improve the convergence and predictive ability of the model. By discretizing continuous variables and converting them into WOE values, the distribution of the variables can be transformed into a form that better suits the model's requirements.

[0076] Discretization: Divide continuous variables (such as years of establishment, registered capital) into N equal intervals. This process can reduce the model's sensitivity to assumptions about the distribution of continuous variables.

[0077] WOE Calculation: The WOE value is calculated based on the ratio of positive to negative samples in each bin. The formula is: ,in, Let x represent the number of positive samples in the i-th bucket. Let x represent the number of negative samples in the i-th bucket. Let x represent the number of positive samples in the full sample. This indicates the number of negative samples in the full sample.

[0078] After WOE transformation, the x variable is converted into n feature variables. ,use Replace x in the model.

[0079] The transformation improves the interpretability of the model, reduces its complexity, better handles imbalanced samples, and enhances its convergence, making it suitable as input for deep learning models.

[0080] S122. Named entity recognition is performed on the tax data of the enterprise to be predicted to obtain key information.

[0081] Specifically, the tax data of the enterprise to be predicted is used to perform named entity recognition using a deep learning model based on BiLSTM-CRF to obtain key information.

[0082] In this embodiment, Named Entity Recognition (NER) technology is used to extract key business terms from the text content of a company name to better identify the company's business scope, industry category, and potential risks. NER technology is typically used to identify named entities in text, such as personal names, organization names, and place names. In this invention, NER is introduced to automatically identify industry-related keywords in the company name, such as industry type, geographical identifier, and business scope. These keywords help improve the accuracy and comprehensiveness of the company's risk prediction model.

[0083] The NER recognition module can automatically extract the following key information from company names:

[0084] Industry keywords: such as "technology", "trade", "manufacturing", "investment", etc., to identify the industry that the company is involved in.

[0085] Geographic identifiers: such as "City A", "City B", "Province C", etc., to identify the geographical location or place of registration of a company.

[0086] Business type: such as "import and export", "e-commerce", "consulting services", etc., to identify the company's business scope and business direction.

[0087] This module performs NER based on the BiLSTM-CRF deep learning model, and its core steps are as follows:

[0088] Data preprocessing: Basic processing of company names is performed, including removing special characters, word segmentation, and standardizing capitalization, to ensure consistency in model processing.

[0089] Model Training: The NER model was trained using a dataset of company names containing labeled entities. The labeled entity types included "industry category," "geographic identifier," and "business type." Combining BiLSTM (Bidirectional Long Short-Term Memory) and CRF (Conditional Random Field) techniques, the model effectively identifies keyword entities in company names.

[0090] Keyword identification and extraction: The trained NER model automatically identifies key information in company names. For example, "A City Technology Co., Ltd." is identified as {Region: A City, Industry: Technology}, while "B City E-commerce Group" is identified as {Region: B City, Business Type: E-commerce}.

[0091] The identified keywords will undergo standardization to ensure their uniqueness and operability within the system. Each keyword will be assigned a unique code for subsequent processing and matching. For example:

[0092] "Industry Category: Technology" may be coded as {10:1001};

[0093] "Region: City A" may be coded as {20:2001};

[0094] "Business Type: E-commerce" may be coded as {30:3001}.

[0095] This standardization not only ensures consistency of various keywords within the system but also improves indexing and query efficiency during data processing. Automated key information extraction enhances data processing efficiency, provides a more accurate foundation for enterprise classification and risk assessment, and offers structured information support for subsequent data analysis and model building.

[0096] S123. The tax data of the enterprise to be predicted is processed according to the set rules to obtain the processing result.

[0097] In this embodiment, at this stage, various features are identified and processed, including basic information features, industry features, shareholder and senior management features, operational features, change record features, and blacklist features.

[0098] Basic information features, such as the unified social credit code, company name, and registered capital, are used to describe the basic information of a company.

[0099] Industry characteristics: The industry code to which a company belongs can help analyze the industry background and market competition.

[0100] Shareholder and executive characteristics: Information such as shareholder stability and the number of management personnel can help assess a company's governance structure and management level.

[0101] Operating characteristics: Characteristics such as the date of establishment and registered capital can help determine the financial health of a company.

[0102] Change record features: Records changes in the enterprise, providing a basis for dynamic risk assessment.

[0103] Blacklist features: By checking whether a company, its legal representative, shareholders, and senior executives are on a blacklist, it is helpful to quickly identify potentially high-risk companies.

[0104] Industry average characteristics: Provides average data within the industry to help relatively evaluate a company's performance.

[0105] By constructing multi-dimensional features, the accuracy and comprehensiveness of the risk assessment model are improved. This facilitates subsequent model training and prediction, enhancing the model's robustness. A systematic information framework is provided, aiding in subsequent decision analysis.

[0106] In addition, the detailed processing rules for the features are shown in Tables 1 to 7.

[0107] Table 1

[0108] Field English Name Field Chinese name describe taxpayer_id Unified Social Credit Code org_encode Organization Code company_name Company Name After NER processing, it is fed into a self-developed deep learning network. register_capital Registered capital register_province Registered Province register_city Registered City operating_status Operating status

[0109] Table 2

[0110] ent_type Enterprise Type ent_type_sta Enterprise type (after mapping) taxpayer_industry1 Authorized enterprises in the first-level industry taxpayer_industry2 Authorized enterprises in secondary industries taxpayer_industry3 Authorized Enterprises Level 3 Industry taxpayer_industry4 Authorized Enterprises Level 4 Industry shareholders_stability Shareholder stability number_managers Number of managers

[0111] Table 3

[0112]

[0113] Table 4

[0114]

[0115] Table 5

[0116]

[0117] Table 6

[0118]

[0119] Table 7

[0120]

[0121] S124. Integrate the conversion result, key information and processing result into a processing result.

[0122] The conversion results, key information, and processing results are combined into a processing result.

[0123] S130. Input the processing result into the prediction model to predict the risk value, so as to obtain the prediction result.

[0124] In this embodiment, the prediction result refers to the probability value of the corresponding type of corporate tax risk.

[0125] The prediction model is obtained by training a multi-task learning network using a full sample set consisting of risky sample datasets and non-risk sample datasets.

[0126] The prediction model includes an input layer, a feature fusion layer, a gated network layer, an expert network layer, and an output layer.

[0127] The training process for the prediction model is as follows:

[0128] First, collect publicly available data, such as the list of companies involved in major illegal cases published by the tax authorities; then, periodically (T+1 update) collect the list of companies involved in major tax violations published by the tax authorities, and organize the data according to two types: tax risk and risk description.

[0129] In addition, data on real tax violations accumulated by the company over a long period of time is collected. This data is only used during the model training phase and is not used during the model inference phase.

[0130] The list of Class A taxpayers published by the tax authorities is collected periodically (updated T+1) to obtain the set LA.

[0131] After combining publicly available data with data on real tax violations accumulated by the company over a long period of time, the resulting list of companies will include: the registered addresses of the companies, which will form a blacklist of addresses, denoted as set H1; the legal representatives of the companies, which will form a blacklist of legal representatives, denoted as set H2; the shareholders of the companies, which will form a blacklist of shareholders, denoted as set H3; and the senior executives of the companies, which will form a blacklist of senior executives, denoted as set H4.

[0132] Next, publicly available data and data on real tax violations accumulated by companies over a long period are merged to form dataset A1. Then, companies with an operating status of "active" or "in operation" are selected to obtain dataset A2. If the same company has multiple violations, they will be recorded as multiple training samples, but it must be ensured that the same company does not appear simultaneously in the training set, test set, and validation set. Bad samples are classified according to the 10 types listed in Tables 1-7; if a company matches any of these types, it is considered to have tax risk.

[0133] From all enterprises in the "existing" or "operating" state, enterprises in dataset A1 are removed to form dataset B1. Next, stratified sampling is performed based on dimensions such as enterprise size, industry, and region. Good samples are selected from dataset B1 to ensure the representativeness and balance of the samples, and finally dataset B2 is obtained.

[0134] Datasets A2 and B2 are merged to obtain the full sample set C1. During training, C1 is split into training, testing, and validation sets to ensure that the same company does not appear in multiple datasets simultaneously. Ultimately, the full sample set C1 is the union of datasets A2 and B2.

[0135] Feature engineering was performed on the full sample set C1, as described in step S120, and will not be repeated here. To ensure data security and model accuracy, all features were taken from the last statistical period before the tax risk occurred. A feature extraction process was designed to ensure that each piece of data could be automatically obtained and processed from public resources. This included using natural language processing techniques to parse complex text data, using xgbooxt to evaluate and select the importance of features, performing missing value imputation, outlier detection, and data normalization.

[0136] A multi-task learning network is constructed, which is based on the Mixture of Experts (MoE) architecture and applied to multi-task learning models to effectively handle various tax risk prediction scenarios. This network includes an input layer, a feature fusion layer, a gating network layer, an expert network layer, and an output layer. The complete process from the input layer through feature fusion, gating, and expert network processing, to the output layer, is illustrated in the diagram. This diagram reflects the core architecture of the MoE model, including the allocation of expert networks and the dynamic adjustment of gating weights. The MoE architecture's gating mechanism dynamically allocates processing tasks to different expert networks. Each expert network handles a specific type of tax risk, and their outputs are then merged into the fusion layer. The multiple tax risk predictions from the expert networks are then integrated into a comprehensive output, which is finally processed by a softmax layer to output the overall tax risk probability.

[0137] The next step is model training, a crucial step involving training a multi-task learning model based on the Mixture of Experts (MoE) architecture using the prepared full sample set C1 (including the bad sample dataset A2 and the good sample dataset B2). The detailed steps are as follows:

[0138] Data preprocessing: Clean the full sample set C1, including handling missing values, detecting outliers, and standardizing data, to ensure that the data quality is suitable for model training.

[0139] Feature selection and transformation: Based on the results of feature engineering, features with strong predictive power for corporate tax risks are selected. Simultaneously, text data is converted into a format that the model can process using natural language processing techniques.

[0140] Sample splitting: The entire sample set C1 is randomly divided into a training set, a validation set, and a test set. The training set is used for model learning, the validation set is used to adjust model parameters to prevent overfitting, and the test set is used to evaluate the model's generalization ability. Ensure that the same company does not appear in different datasets simultaneously.

[0141] Model architecture setup: Configure a Mixture of Experts architecture, designing multiple expert networks and a gating mechanism. Each expert network focuses on learning the characteristic patterns of a specific type of tax risk, while the gating mechanism is responsible for dynamically selecting the most suitable expert network based on the input features.

[0142] Model training: Optimize model parameters using gradient descent. During training, continuously monitor the performance on the validation set and apply techniques such as early stopping to avoid overfitting. Additionally, methods like weight decay and dropout can be used to improve the model's generalization ability.

[0143] Hyperparameter tuning: Utilize automated hyperparameter optimization techniques (such as grid search and Bayesian optimization) to find the optimal model parameters, including learning rate, batch size, and number of experts.

[0144] Model Evaluation and Selection: Evaluate model performance on the test set, with particular attention to the predictive accuracy and recall of different types of tax risks. Ultimately, select the model configuration that performs best across multiple metrics.

[0145] In one embodiment, step S130 described above may include steps S131 to S135.

[0146] S131. Input the processing result into the prediction model.

[0147] In this stage, the processing results obtained in the previous step (which may be data after feature extraction or preprocessing results) are first passed to the prediction model. The prediction model can be a machine learning or deep learning-based model, designed for further analysis and prediction.

[0148] Directly inputting cleaned and characterized data into the model can improve the accuracy and efficiency of subsequent predictions. This forms a systematic data processing chain, reducing the need for manual intervention and thus improving consistency.

[0149] S132. The processing results are fused to obtain a fused result.

[0150] In this embodiment, data from different sources or with different characteristics are fused. This step is usually achieved through weighted averaging, splicing, or other fusion techniques.

[0151] Integrating information from different sources can provide a more comprehensive perspective and reduce the bias of information. Data fusion can improve the robustness and generalization ability of models and reduce the noise that may come from a single data source.

[0152] S133. Based on the fusion result, the gating network layer determines the expert network layer to participate in the processing.

[0153] Specifically, the gated network layer assigns tasks to corresponding expert network layers based on the characteristics of the fusion result, wherein each expert network layer focuses on a certain type of tax risk.

[0154] The gating network layer dynamically determines which expert networks will participate in the final prediction based on the characteristics of the fusion results. This process includes evaluating the suitability of each expert and assigning the task to the most relevant expert network.

[0155] Each expert network is trained for a specific type of tax risk, enabling the model to handle different situations tailored to individual circumstances and improve prediction accuracy. By selecting the most suitable expert network, computational resources are reduced while processing speed is increased.

[0156] S134. The fusion results are processed using a defined expert network layer to determine the type of tax risk.

[0157] In this embodiment, the expert network selected by the gating network layer processes the fusion results and outputs the type of tax risk. This step typically involves forward propagation and the use of activation functions.

[0158] Expert networks, by focusing on specific types of tasks, are able to identify and classify tax risks more quickly. The specialized training of each expert network enables it to perform better on specific tasks, further improving the accuracy of predictions.

[0159] S135. Integrate the tax risk types output by the expert network, and calculate the tax risk prediction probability through the Softmax layer to obtain the prediction result.

[0160] In this embodiment, the tax risk types output by all expert networks are ultimately integrated, and then the predicted probability for each tax risk type is calculated using a Softmax layer. This process ensures the interpretability of the final results.

[0161] Using a Softmax layer provides predicted probabilities for each tax risk, making the decision-making process more transparent and interpretable. The probabilities help decision-makers understand the relative likelihood of different risks, thus enabling better risk management and resource allocation.

[0162] By implementing the above steps, the entire forecasting process can efficiently and accurately identify and assess a company's tax risks, thereby improving its decision support capabilities.

[0163] S140. Determine the risk level based on the prediction results and the set risk threshold.

[0164] In this embodiment, the tax risks of enterprises are assessed based on set thresholds. Typically, the assessment results will provide both a risk probability value and a risk level. The specific classifications are as follows:

[0165] Level A risk: Score above 90;

[0166] Category B risk: Score between 70 and 90;

[0167] Level C risk: Score between 50 and 70;

[0168] Level D risk: Score below 50;

[0169] These thresholds can be dynamically adjusted according to actual business needs.

[0170] S150, Output the stated risk level.

[0171] In this embodiment, the risk level is sent to the terminal for display.

[0172] The aforementioned method for predicting corporate tax risks involves acquiring the tax data of the company to be predicted, then performing feature engineering to extract key features. The processed results are then input into a prediction model to predict risk values, ensuring the model is trained using a multi-task learning network across the entire sample set, thus possessing stronger recognition capabilities. Based on the prediction results and a set risk threshold, the risk level is determined and finally output. This process not only improves the ability to identify tax risks but also enhances the accuracy and comprehensiveness of the predictions.

[0173] Figure 3 This is a schematic block diagram of a corporate tax risk prediction device 300 provided in an embodiment of the present invention. Figure 3 As shown, corresponding to the above-described enterprise tax risk prediction method, the present invention also provides an enterprise tax risk prediction device 300. This enterprise tax risk prediction device 300 includes a unit for executing the above-described enterprise tax risk prediction method, and the device can be configured in a server. Specifically, please refer to... Figure 3 The enterprise tax risk prediction device 300 includes a data acquisition unit 301, a feature engineering processing unit 302, a prediction unit 303, a risk level determination unit 304, and an output unit 305.

[0174] The data acquisition unit 301 is used to acquire the tax data of the enterprise to be predicted; the feature engineering processing unit 302 is used to perform feature engineering processing on the tax data of the enterprise to be predicted to obtain the processing result; the prediction unit 303 is used to input the processing result into the prediction model to predict the risk value to obtain the prediction result; the risk level determination unit 304 is used to determine the risk level according to the prediction result and the set risk threshold; the output unit 305 is used to output the risk level; wherein, the prediction model is obtained by training a multi-task learning network with a full sample set composed of risk sample dataset and non-risk sample dataset.

[0175] In one embodiment, the feature engineering processing unit 302 includes:

[0176] The transformation subunit is used to perform continuous feature WOE conversion on the tax data of the enterprise to be predicted to obtain the transformation result; the identification subunit is used to perform named entity recognition on the tax data of the enterprise to be predicted to obtain key information; the processing subunit is used to process the tax data of the enterprise to be predicted according to set rules to obtain the processing result; and the integration subunit is used to integrate the transformation result, key information and processing result into a processing result.

[0177] In one embodiment, the prediction unit 303 includes:

[0178] The system comprises: an input subunit for inputting the processing results into the prediction model; a fusion subunit for fusing the processing results to obtain a fusion result; a determination subunit for determining the expert network layers to participate in the processing based on the fusion result using the gated network layer; a processing subunit for processing the fusion result using the determined expert network layers to determine the tax risk type; and an integration subunit for integrating the tax risk types output by the expert networks, calculating the tax risk prediction probability through a Softmax layer to obtain the prediction result.

[0179] In one embodiment, the determining subunit is used by the gated network layer to assign tasks to corresponding expert network layers based on the characteristics of the fusion result, wherein each expert network layer focuses on a certain type of tax risk.

[0180] It should be noted that those skilled in the art can clearly understand that the specific implementation process of the above-mentioned enterprise tax risk prediction device 300 and each unit can be referred to the corresponding description in the foregoing method embodiments. For the sake of convenience and brevity, it will not be repeated here.

[0181] The aforementioned enterprise tax risk prediction device 300 can be implemented as a computer program, which can, for example... Figure 4 It runs on the computer device shown.

[0182] Please see Figure 4 , Figure 4 This is a schematic block diagram of a computer device provided in an embodiment of this application. The computer device 500 can be a server, wherein the server can be a standalone server or a server cluster composed of multiple servers.

[0183] See Figure 4 The computer device 500 includes a processor 502, a memory, and a network interface 505 connected via a system bus 501. The memory may include a non-volatile storage medium 503 and internal memory 504.

[0184] The non-volatile storage medium 503 may store an operating system 5031 and a computer program 5032. The computer program 5032 includes program instructions that, when executed, cause the processor 502 to perform a corporate tax risk prediction method.

[0185] The processor 502 provides computing and control capabilities to support the operation of the entire computer device 500.

[0186] The internal memory 504 provides an environment for the operation of the computer program 5032 in the non-volatile storage medium 503. When the computer program 5032 is executed by the processor 502, the processor 502 can execute a corporate tax risk prediction method.

[0187] This network interface 505 is used for network communication with other devices. Those skilled in the art will understand that... Figure 4 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device 500 to which the present application is applied. The specific computer device 500 may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0188] The processor 502 is used to run a computer program 5032 stored in the memory to perform the following steps:

[0189] Obtain the tax data of the enterprise to be predicted; perform feature engineering on the tax data to obtain the processing result; input the processing result into the prediction model to predict the risk value to obtain the prediction result; determine the risk level based on the prediction result and a set risk threshold; output the risk level.

[0190] The prediction model is obtained by training a multi-task learning network using a full sample set consisting of risky sample datasets and non-risk sample datasets.

[0191] The prediction model includes an input layer, a feature fusion layer, a gated network layer, an expert network layer, and an output layer.

[0192] In one embodiment, when the processor 502 performs feature engineering processing on the tax data of the enterprise to be predicted to obtain the processing result, it specifically implements the following steps:

[0193] The tax data of the enterprise to be predicted is converted into continuous features (WOE) to obtain the transformation result; the tax data of the enterprise to be predicted is subjected to named entity recognition to obtain key information; the tax data of the enterprise to be predicted is processed according to set rules to obtain the processing result; the transformation result, key information and processing result are integrated into the processing result.

[0194] In one embodiment, when the processor 502 performs named entity recognition on the tax data of the enterprise to be predicted to obtain key information, it specifically implements the following steps:

[0195] The tax data of the enterprises to be predicted are used to perform named entity recognition using a deep learning model based on BiLSTM-CRF to obtain key information.

[0196] In one embodiment, when the processor 502 implements the step of inputting the processing result into the prediction model to predict the risk value and obtain the prediction result, it specifically implements the following steps:

[0197] The processing results are input into the prediction model; the processing results are fused to obtain a fused result; the gating network layer determines the expert network layer to participate in the processing based on the fused result; the determined expert network layer processes the fused result to determine the tax risk type; the tax risk types output by the expert network are integrated and the tax risk prediction probability is calculated through the Softmax layer to obtain the prediction result.

[0198] In one embodiment, when implementing the step of determining the participating expert network layer based on the fusion result by the gated network layer, the processor 502 specifically implements the following steps:

[0199] The gated network layer assigns tasks to the corresponding expert network layers based on the characteristics of the fusion result, wherein each expert network layer focuses on a certain type of tax risk.

[0200] It should be understood that in the embodiments of this application, the processor 502 may be a central processing unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor.

[0201] It will be understood by those skilled in the art that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program includes program instructions and can be stored in a storage medium, which is a computer-readable storage medium. The program instructions are executed by at least one processor in the computer system to implement the process steps of the embodiments of the above methods.

[0202] Therefore, the present invention also provides a storage medium. This storage medium can be a computer-readable storage medium. The storage medium stores a computer program, wherein when executed by a processor, the computer program causes the processor to perform the following steps:

[0203] Obtain the tax data of the enterprise to be predicted; perform feature engineering on the tax data to obtain the processing result; input the processing result into the prediction model to predict the risk value to obtain the prediction result; determine the risk level based on the prediction result and a set risk threshold; output the risk level.

[0204] The prediction model is obtained by training a multi-task learning network using a full sample set consisting of risky sample datasets and non-risk sample datasets.

[0205] The prediction model includes an input layer, a feature fusion layer, a gated network layer, an expert network layer, and an output layer.

[0206] In one embodiment, when the processor executes the computer program to perform feature engineering processing on the tax data of the enterprise to be predicted to obtain the processing result, it specifically implements the following steps:

[0207] The tax data of the enterprise to be predicted is converted into continuous features (WOE) to obtain the transformation result; the tax data of the enterprise to be predicted is subjected to named entity recognition to obtain key information; the tax data of the enterprise to be predicted is processed according to set rules to obtain the processing result; the transformation result, key information and processing result are integrated into the processing result.

[0208] In one embodiment, when the processor executes the computer program to perform named entity recognition on the tax data of the enterprise to be predicted in order to obtain key information, it specifically implements the following steps:

[0209] The tax data of the enterprises to be predicted are used to perform named entity recognition using a deep learning model based on BiLSTM-CRF to obtain key information.

[0210] In one embodiment, when the processor executes the computer program to implement the step of inputting the processing result into the prediction model for risk value prediction to obtain the prediction result, it specifically implements the following steps:

[0211] The processing results are input into the prediction model; the processing results are fused to obtain a fused result; the gating network layer determines the expert network layer to participate in the processing based on the fused result; the determined expert network layer processes the fused result to determine the tax risk type; the tax risk types output by the expert network are integrated and the tax risk prediction probability is calculated through the Softmax layer to obtain the prediction result.

[0212] In one embodiment, when the processor executes the computer program to implement the step of determining the participating expert network layer based on the fusion result by the gated network layer, it specifically implements the following steps:

[0213] The gated network layer assigns tasks to the corresponding expert network layers based on the characteristics of the fusion result, wherein each expert network layer focuses on a certain type of tax risk.

[0214] The storage medium can be any computer-readable storage medium capable of storing program code, such as a USB flash drive, portable hard drive, read-only memory (ROM), magnetic disk, or optical disk.

[0215] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0216] In the several embodiments provided by this invention, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative. For example, the division of each unit is merely a logical functional division, and there may be other division methods in actual implementation. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed.

[0217] The steps in the method of this invention can be adjusted, merged, or reduced in order according to actual needs. The units in the device of this invention can be merged, divided, or reduced according to actual needs. Furthermore, the functional units in the various embodiments of this invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0218] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, a terminal, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention.

[0219] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for predicting tax risk for a business, characterized by, The method comprises the following steps: obtaining enterprise tax data to be predicted; performing feature engineering processing on the enterprise tax data to be predicted to obtain a processing result; inputting the processing result into a prediction model to perform risk value prediction and obtain a prediction result; determining a risk level according to the prediction result and a set risk threshold; outputting the risk level; wherein the prediction model is obtained by training a multi-task learning network using a full sample set composed of a risk sample data set and a non-risk sample data set; the feature engineering processing on the enterprise tax data to be predicted to obtain a processing result comprises the following steps: performing WOE on continuous features of the enterprise tax data to be predicted to obtain a conversion result; performing named entity recognition on the enterprise tax data to be predicted to obtain key information; performing rule setting processing on the enterprise tax data to be predicted to obtain a processing result; integrating the conversion result, the key information and the processing result into the processing result; the prediction model comprises an input layer, a feature fusion layer, a gate network layer, an expert network layer and an output layer; the inputting of the processing result into the prediction model to perform risk value prediction and obtain a prediction result comprises the following steps: inputting the processing result into the prediction model; performing fusion on the processing result to obtain a fusion result; determining an expert network layer participating in processing according to the fusion result by the gate network layer; processing the fusion result by the determined expert network layer to determine a tax risk type; integrating the tax risk types output by the expert network, calculating the tax risk prediction probability by a Softmax layer to obtain a prediction result.

2. The method of claim 1, wherein, the named entity recognition on the enterprise tax data to be predicted to obtain key information comprises the following steps: performing named entity recognition on the enterprise tax data to be predicted by using a deep learning model based on BiLSTM-CRF to obtain key information.

3. The method of claim 1, wherein, the determination of an expert network layer participating in processing according to the fusion result by the gate network layer comprises the following steps: the gate network layer assigns tasks to corresponding expert network layers according to the characteristics of the fusion result, wherein each expert network layer focuses on a certain type of tax risk.

4. An enterprise tax risk prediction device characterized by comprising: The method comprises the following steps: a data acquisition unit is configured to acquire enterprise tax data to be predicted; a feature engineering processing unit is configured to perform feature engineering processing on the enterprise tax data to be predicted to obtain a processing result; a prediction unit is configured to input the processing result into a prediction model to perform risk value prediction and obtain a prediction result; a risk level determination unit is configured to determine a risk level according to the prediction result and a set risk threshold; an output unit is configured to output the risk level; wherein the prediction model is obtained by training a multi-task learning network using a full sample set composed of a risk sample data set and a non-risk sample data set; the feature engineering processing unit comprises: a conversion subunit configured to perform WOE on continuous features of the enterprise tax data to be predicted to obtain a conversion result; an identification subunit configured to perform named entity recognition on the enterprise tax data to be predicted to obtain key information; The processing subunit is configured to perform rule setting processing on the enterprise tax data to be predicted to obtain a processing result. The integration subunit is configured to integrate the conversion result, the key information, and the processing result into a processing result. The prediction model comprises an input layer, a feature fusion layer, a gate network layer, an expert network layer, and an output layer. The prediction unit comprises: The input subunit is configured to input the processing result into the prediction model; the fusion subunit is configured to fuse the processing result to obtain a fusion result; the determination subunit is configured to determine, according to the fusion result, the expert network layer participating in processing from the gate network layer; the processing subunit is configured to process the fusion result by using the determined expert network layer to determine a tax risk type; and the integration subunit is configured to integrate the tax risk types output by the expert network, calculate a tax risk prediction probability by using a Softmax layer, and obtain a prediction result.

5. A computer device, comprising: The computer device comprises a memory and a processor, the memory stores a computer program, and the processor implements the method according to any one of claims 1 to 3 when executing the computer program.

6. A storage medium, characterized by The storage medium stores a computer program, and the computer program is executed by the processor to implement the method according to any one of claims 1 to 3.

Images