Enterprise credit brand value evaluation method and system based on big data
By combining big data and Markov decision models with Q-learning methods, the limitations of traditional corporate credit assessment based on a single data source and its lack of dynamic adaptability are addressed. This enables a comprehensive assessment of corporate credit and brand value, and provides dynamic monitoring and early warning support.
Patent Information
- Application Number
- CN202410969967.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-19
- Publication Date
- 2024-11-12
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
Traditional corporate credit assessment methods rely on a single data source, which cannot comprehensively reflect the credit status of enterprises, have weak dynamic adaptability, and cannot effectively combine credit assessment results with brand value assessment. They also lack dynamic monitoring and early warning support for corporate credit risk.
This paper adopts a big data-based corporate credit brand value assessment method. Through multi-data source collection and preprocessing, a Markov decision model is constructed. The Q-learning method is used for action evaluation and parameter optimization. Combined with financial, market performance, social reputation and credit risk indicators, a corporate credit brand value assessment system is constructed.
It improves the comprehensiveness and accuracy of corporate credit assessment, dynamically adapts to changes in corporate credit, enhances the precision of brand value assessment, provides decision support for corporate credit management and brand building, and establishes a dynamic monitoring and early warning mechanism.
Smart Images

Figure CN118941141A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of deep learning technology, and in particular to a method and system for evaluating corporate credit brand value based on big data. Background Art
[0002] With the advent of the big data era, corporate credit assessment and brand value assessment have become hot topics in corporate management and academic research. Traditional credit assessment methods are often limited to the internal data of a single company, such as finance and operations, and it is difficult to fully reflect the credit risk status of the company. Brand value assessment also mainly relies on external data such as market research and consumer surveys, and lacks in-depth consideration of the internal credit status of the company.
[0003] The evaluation of corporate credit brand value has the following problems:
[0004] (1) Limitations of a single data source: Traditional credit assessment methods usually rely on a single data source and cannot fully reflect the credit status of an enterprise;
[0005] (2) Weak ability to dynamically adapt to changes in corporate credit status: unable to continuously receive feedback from companies and optimize assessment strategies, making it difficult to improve the ability to predict corporate credit risk;
[0006] (3) Insufficient decision-making support for credit management and brand building: Existing credit assessment results cannot be effectively combined with corporate brand value assessment, making it difficult to provide comprehensive decision-making support for corporate credit management and brand building;
[0007] (4) Industry credit risk management needs: Regulatory authorities need more accurate and reliable corporate credit data to strengthen dynamic monitoring and early warning of industry credit risks.
[0008] How to conduct a comprehensive assessment of a company's credit brand value is a technical problem that needs to be solved. Summary of the invention
[0009] The technical task of the present invention is to address the above shortcomings and provide a method and system for evaluating the corporate credit brand value based on big data to solve the technical problem of how to comprehensively evaluate the corporate credit brand value.
[0010] In a first aspect, the present invention provides a method for evaluating corporate credit brand value based on big data, comprising the following steps:
[0011] Data collection and processing: Collect enterprise data from multiple data sources, preprocess the enterprise data, build a data set based on the preprocessed enterprise data, and segment the enterprise data in the data set into training set, validation set, and test set;
[0012] Problem construction: Construct a Markov decision model for corporate credit brand value assessment. For the Markov decision process, define the multi-dimensional characteristics of the corporate credit brand as the state, define optional credit assessment decisions as actions, and define rewards based on the actual credit situation of the enterprise;
[0013] Model optimization: For the Markov decision model, action evaluation and parameter optimization are performed through the Q-learning method.
[0014] Preferably, the enterprise data is preprocessed, including the following steps:
[0015] Clean enterprise data to remove missing values, outliers and duplicate data;
[0016] Standardize enterprise data from different sources and in different data formats;
[0017] Desensitize enterprise data;
[0018] Based on the needs of enterprise brand value assessment, feature selection and feature engineering are performed on enterprise data to extract effective features, including feature combination and feature conversion.
[0019] Preferably, for the Markov decision model, action evaluation and parameter optimization are performed by Q-learning method, including the following steps:
[0020] Building a learning model based on a convolutional neural network model, the learning model includes an action selection model and an action evaluation model with the same structure, and initializing the action selection model and the action evaluation model;
[0021] Initialization state t , based on the greedy strategy, select action a t , the greedy strategy is: each time a new action is selected with a probability of ε, and the action corresponding to the maximum Q value output by the action selection model is selected with a probability of 1-ε, and the value range of ε is (0,1);
[0022] Execute action a t Get rewards t ;
[0023] Status updated to s t+1 ;
[0024] {s t ,a t ,r t ,s t+1} is stored as a four-tuple in the experience pool;
[0025] Determine whether the target number of times has been reached. If not, reselect action a based on the greedy strategyt If yes, update the network parameters of the action selection model and the action evaluation model. When updating the network parameters of the action evaluation model, copy the updated network parameters of the action selection model to the action evaluation model. When updating the network parameters of the action selection model, the agent randomly extracts n data from the experience pool, updates the parameters of the action selection model, and reduces the value of the exploration probability ε in the greedy strategy.
[0026] Determine whether the termination condition is met, if not, reinitialize state s t , if so, output the optimal strategy.
[0027] Preferably, the learning model includes an input layer, a hidden layer and an output layer;
[0028] The input of the action selection model is the state s t , the output is the Q value corresponding to n actions, respectively {Q(s t ,a1),Q(s t ,a2),…,Q(s t ,a n )};
[0029] The input of the action evaluation model is the state s t+1 , the output is the Q value corresponding to n actions, respectively {Q(s t+1 ,a1),Q(s t+1 ,a2),…,Q(s t+1 ,a n )};
[0030] With r t+1 +γQ(s t+1 ,a i ), i = 1, 2, ..., n as the actual value to revise the action selection model, r t+1 Indicates that from state s t is the real-time excitation after reaching the state, and γ is the delay coefficient.
[0031] As a preference, when defining rewards based on the actual credit situation of the enterprise, set enterprise indicators and set target values or intervals for each enterprise indicator. When the enterprise indicator meets the target value or interval, a positive reward is given. When the enterprise indicator deviates from the target value or interval, a negative penalty is imposed.
[0032] Corporate indicators include financial indicators, market performance indicators, social reputation indicators and credit risk indicators. Among them, financial indicators include debt-to-asset ratio, current ratio and net profit, market performance indicators include market share, operating income growth rate and customer satisfaction, social reputation indicators include public opinion evaluation, social responsibility fulfillment and environmental protection, and credit risk indicators include default records, litigation situations and regulatory penalties.
[0033] In a second aspect, the present invention provides a corporate credit brand value assessment system based on big data, which is used to implement corporate brand value assessment through a corporate credit brand value assessment method based on big data as described in any one of the first aspects, wherein the system includes a data collection and processing module, a problem construction module, and a model optimization module;
[0034] The data collection and processing module is used to perform the following: collect enterprise data from multiple data sources, perform data preprocessing on the enterprise data, construct a data set based on the preprocessed enterprise data, and perform data segmentation on the enterprise data in the data set, dividing the data set into a training set, a validation set, and a test set;
[0035] The problem building module is used to perform the following: construct a Markov decision model for corporate credit brand value assessment. For the Markov decision process, define the multi-dimensional characteristics of the corporate credit brand as the state, define the optional credit assessment decision as the action, and define the reward according to the actual credit situation of the enterprise;
[0036] The model optimization module is used to perform the following: For the Markov decision model, action evaluation and parameter optimization are performed through the Q-learning method.
[0037] Preferably, the data collection and processing module is used to perform the following pre-processing of the enterprise data:
[0038] Clean enterprise data to remove missing values, outliers and duplicate data;
[0039] Standardize enterprise data from different sources and in different data formats;
[0040] Desensitize enterprise data;
[0041] Based on the needs of enterprise brand value assessment, feature selection and feature engineering are performed on enterprise data to extract effective features, including feature combination and feature conversion.
[0042] Preferably, for the Markov decision model, the model optimization module is used to perform the following action evaluation and parameter optimization by Q-learning method:
[0043] Building a learning model based on a convolutional neural network model, the learning model includes an action selection model and an action evaluation model with the same structure, and initializing the action selection model and the action evaluation model;
[0044] Initialization state t , based on the greedy strategy, select action a t, the greedy strategy is: each time a new action is selected with a probability of ε, and the action corresponding to the maximum Q value output by the action selection model is selected with a probability of 1-ε, and the value range of ε is (0,1);
[0045] Execute action a t Get rewards t ;
[0046] Status updated to s t+1 ;
[0047] {s t ,a t ,r t ,s t+1} is stored as a four-tuple in the experience pool;
[0048] Determine whether the target number of times has been reached. If not, reselect action a based on the greedy strategy t If yes, update the network parameters of the action selection model and the action evaluation model. When updating the network parameters of the action evaluation model, copy the updated network parameters of the action selection model to the action evaluation model. When updating the network parameters of the action selection model, the agent randomly extracts n data from the experience pool, updates the parameters of the action selection model, and reduces the value of the exploration probability ε in the greedy strategy.
[0049] Determine whether the termination condition is met, if not, reinitialize state s t , if so, output the optimal strategy.
[0050] Preferably, the learning model includes an input layer, a hidden layer and an output layer;
[0051] The input of the action selection model is the state s t , the output is the Q value corresponding to n actions, respectively {Q(s t ,a1),Q(s t ,a2),…,Q(s t ,a n )};
[0052] The input of the action evaluation model is the state s t+1 , the output is the Q value corresponding to n actions, respectively {Q(s t+1 ,a1),Q(s t+1 ,a2),…,Q(s t+1 ,a n )};
[0053] With r t+1 +γQ(s t+1 ,a i ), i = 1, 2, ..., n as the actual value to revise the action selection model, rt+1 Indicates that from state s t is the real-time excitation after reaching the state, and γ is the delay coefficient.
[0054] As a preference, when defining rewards based on the actual credit situation of the enterprise, set enterprise indicators and set target values or intervals for each enterprise indicator. When the enterprise indicator meets the target value or interval, a positive reward is given. When the enterprise indicator deviates from the target value or interval, a negative penalty is imposed.
[0055] Corporate indicators include financial indicators, market performance indicators, social reputation indicators and credit risk indicators. Among them, financial indicators include debt-to-asset ratio, current ratio and net profit, market performance indicators include market share, operating income growth rate and customer satisfaction, social reputation indicators include public opinion evaluation, social responsibility fulfillment and environmental protection, and credit risk indicators include default records, litigation situations and regulatory penalties.
[0056] The enterprise credit brand value evaluation method and system based on big data of the present invention have the following advantages:
[0057] 1. Improve the comprehensiveness and accuracy of enterprise credit assessment, enhance the ability to dynamically adapt to changes in enterprise credit assessment, receive feedback from enterprises in real time and optimize assessment strategies, and improve the ability to predict enterprise credit risks;
[0058] 2. Improve the accuracy of brand value assessment. Combining corporate credit assessment with brand value assessment can more accurately reflect the brand value of the company and provide decision-making support for corporate brand building;
[0059] 3. Establish a dynamic monitoring and early warning mechanism to dynamically monitor and warn of corporate credit status, promptly detect changes in credit risk, and provide risk management support for enterprises and regulatory authorities. BRIEF DESCRIPTION OF THE DRAWINGS
[0060] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0061] The present invention is further described below in conjunction with the accompanying drawings.
[0062] Figure 1 This is a flowchart of a method for evaluating corporate credit brand value based on big data in Example 1. DETAILED DESCRIPTION
[0063] The present invention is further described below in conjunction with the accompanying drawings and specific embodiments so that those skilled in the art can better understand the present invention and implement it. However, the embodiments are not intended to limit the present invention. In the absence of conflict, the embodiments of the present invention and the technical features in the embodiments may be combined with each other.
[0064] The embodiment of the present invention provides a method and system for evaluating the corporate credit brand value based on big data, which are used to solve the technical problem of how to comprehensively evaluate the corporate credit brand value.
[0065] Embodiment 1:
[0066] The present invention provides a corporate credit brand value evaluation method based on big data, which includes three steps: data collection and processing, problem construction and model optimization.
[0067] Step S100: Data collection and processing: Collect enterprise data from multiple data sources, perform data preprocessing on the enterprise data, build a data set based on the preprocessed enterprise data, and perform data segmentation on the enterprise data in the data set, dividing the data set into a training set, a validation set, and a test set.
[0068] In this embodiment, the enterprise data is preprocessed, including the following steps:
[0069] (1) Clean the enterprise data to remove missing values, outliers and duplicate data;
[0070] (2) Standardize enterprise data from different sources and in different data formats;
[0071] (3) Desensitize enterprise data;
[0072] (4) Based on the needs of corporate brand value assessment, feature selection and feature engineering are performed on corporate data to extract effective features, including feature combination and feature transformation.
[0073] This step of the present embodiment is divided into the following processes: First, data collection. Collect the financial, operational, credit and other related data of the enterprise from multiple enterprise credit data sources (such as banks, taxation, industry and commerce, etc.), collect data related to the enterprise brand, such as brand awareness, reputation, market share, etc., to ensure the timeliness and integrity of the data, and cover the historical information of the enterprise as much as possible. Second, data cleaning. Clean the collected data, including processing missing values, outliers, duplicate data, etc., to ensure the accuracy and consistency of the data, and provide a good foundation for subsequent data analysis and modeling. Third, data standardization. Standardize data from different sources and formats to ensure the uniformity of data formats. According to the characteristics of the evaluation indicators, adopt appropriate standardization methods, such as z-score standardization, min-max standardization, etc. Fourth, data desensitization. Desensitize the collected enterprise data to protect the privacy and commercial secrets of the enterprise. Use encryption, anonymization and other technical means to ensure the security of data during sharing and use. Fifth, feature engineering. According to the needs of credit brand value assessment, feature selection and engineering are performed on the data to extract effective features, including feature combination, feature conversion and other technologies to improve the prediction performance of the model. Sixth, data segmentation. Divide the collected data into training set, validation set and test set for subsequent model training and evaluation. Ensure that the data distribution between the training set, validation set and test set is as consistent as possible to avoid data bias.
[0074] Step S200 Problem construction: Construct a Markov decision model for corporate credit brand value evaluation. For the Markov decision process, define the multi-dimensional characteristics of the corporate credit brand as the state, define optional credit evaluation decisions as actions, and define rewards based on the actual credit situation of the enterprise.
[0075] Since different institutions have the same type of characteristics of different user groups or different types of characteristics of the same user groups, it is necessary to first build a simulation model that simulates the enterprise operating environment and define the state space, including multi-dimensional characteristics such as enterprise financial indicators, market performance, and social reputation. Secondly, define optional credit assessment decision actions. For example, define rating actions, and make credit ratings of AAA, AA, A, BBB, BB, etc. according to the current situation and future forecasts of the enterprise; define suggestion actions, and give corresponding improvement suggestions when the enterprise needs to enhance financial management, improve market performance, and improve social reputation; define report actions, generate enterprise credit assessment reports, and summarize the analysis results. Finally, design a reward function based on the actual credit status of the enterprise. Set the corresponding target value or interval, give positive rewards when the enterprise indicators meet the standards, and give negative penalties when the indicators deviate from the targets. The indicator settings include but are not limited to financial indicators such as debt-to-asset ratio, current ratio, and net profit, market performance indicators such as market share, operating income growth rate, and customer satisfaction, social reputation indicators such as public opinion evaluation, social responsibility fulfillment, and environmental protection, and credit risk indicators such as the enterprise's default record, litigation, and regulatory penalties.
[0076] Step S300: Model optimization: For the Markov decision model, action evaluation and parameter optimization are performed using the Q-learning method.
[0077] As a specific implementation, for the Markov decision model, action evaluation and parameter optimization are performed through the Q-learning method, including the following steps:
[0078] (1) Building a learning model based on a convolutional neural network model, the learning model includes an action selection model and an action evaluation model with the same structure, and initializing the action selection model and the action evaluation model;
[0079] (2) Initialization state s t , based on the greedy strategy, select action a t , the greedy strategy is: each time a new action is selected with a probability of ε, and the action corresponding to the maximum Q value output by the action selection model is selected with a probability of 1-ε, and the value range of ε is (0,1);
[0080] (3) Execute action a t Get rewards t ;
[0081] (4) The status is updated to s t+1 ;
[0082] (5) {s t ,a t ,r t ,s t+1} is stored as a four-tuple in the experience pool;
[0083] (6) Determine whether the target number of times has been reached. If not, reselect action a based on the greedy strategy t If yes, update the network parameters of the action selection model and the action evaluation model. When updating the network parameters of the action evaluation model, copy the updated network parameters of the action selection model to the action evaluation model. When updating the network parameters of the action selection model, the agent randomly extracts n data from the experience pool, updates the parameters of the action selection model, and reduces the value of the exploration probability ε in the greedy strategy.
[0084] (7) Determine whether the termination condition is met. If not, reinitialize state s t , if so, output the optimal strategy.
[0085] This embodiment is based on the reinforcement learning framework of the ε-greedy strategy. The corporate credit brand value evaluation model can dynamically balance between exploration and utilization, continuously optimize the evaluation strategy, and improve the understanding and prediction ability of the corporate credit status. When using the ε-greedy strategy for decision-making, the exploration action is usually randomly selected with probability ε, and the current optimal action is selected with probability 1-ε. The Q-learning algorithm is used to continuously update the expected cumulative reward, and finally an optimal state-action value function Q(s,a) is obtained, where the update rule is: Q(s,a)=Q(s,a)+α[r+γ*max_a'Q(s',a')-Q(s,a)], α is the learning rate, and γ is the discount factor. After that, training is performed based on the neural network model, and the exploration probability ε in the ε-greedy strategy is dynamically adjusted, and gradually reduced to improve the utilization efficiency.
[0086] This embodiment can feed back the evaluation results to the enterprise to help it understand its own credit status and brand value, and provide decision support for corporate credit management and brand building. The evaluation results are provided to the regulatory authorities to provide data support for industry credit risk management. A dynamic monitoring and early warning mechanism is established to timely detect changes in the credit status of enterprises and provide support for risk management.
[0087] Embodiment 2:
[0088] The present invention discloses an enterprise credit brand value evaluation system based on big data, comprising a data collection and processing module, a problem construction module and a model optimization module.
[0089] The data collection and processing module is used to perform the following: collect enterprise data from multiple data sources, preprocess the enterprise data, build a data set based on the preprocessed enterprise data, and split the enterprise data in the data set, dividing the data set into a training set, a validation set, and a test set.
[0090] As a specific implementation, the data collection and processing module is used to perform the following pre-processing of enterprise data:
[0091] (1) Clean the enterprise data to remove missing values, outliers and duplicate data;
[0092] (2) Standardize enterprise data from different sources and in different data formats;
[0093] (3) Desensitize enterprise data;
[0094] (4) Based on the needs of corporate brand value assessment, feature selection and feature engineering are performed on corporate data to extract effective features, including feature combination and feature transformation.
[0095] In this embodiment, this module is used to perform the following operations: First, data collection. Collect financial, operational, credit and other related data of enterprises from multiple enterprise credit data sources (such as banks, taxation, industry and commerce, etc.), collect enterprise brand-related data, such as brand awareness, reputation, market share, etc., to ensure the timeliness and integrity of the data, and cover the historical information of the enterprise as much as possible. Second, data cleaning. Clean the collected data, including processing missing values, outliers, duplicate data, etc., to ensure the accuracy and consistency of the data, and provide a good foundation for subsequent data analysis and modeling. Third, data standardization. Standardize data from different sources and formats to ensure the uniformity of data formats. According to the characteristics of the evaluation indicators, adopt appropriate standardization methods, such as z-score standardization, min-max standardization, etc. Fourth, data desensitization. Desensitize the collected enterprise data to protect the privacy and commercial secrets of the enterprise. Use encryption, anonymization and other technical means to ensure the security of data during sharing and use. Fifth, feature engineering. According to the needs of credit brand value assessment, feature selection and engineering are performed on the data to extract effective features, including feature combination, feature conversion and other technologies to improve the prediction performance of the model. Sixth, data segmentation. Divide the collected data into training set, validation set and test set for subsequent model training and evaluation. Ensure that the data distribution between the training set, validation set and test set is as consistent as possible to avoid data bias.
[0096] The problem building module is used to perform the following: construct a Markov decision model for corporate credit brand value assessment, define the multi-dimensional characteristics of the corporate credit brand as the state, define optional credit assessment decisions as actions, and define rewards based on the actual credit situation of the enterprise.
[0097] Since different institutions have the same type of characteristics of different user groups or different types of characteristics of the same user groups, it is necessary to first build a simulation model that simulates the enterprise operating environment and define the state space, including multi-dimensional characteristics such as enterprise financial indicators, market performance, and social reputation. Secondly, define optional credit assessment decision actions. For example, define rating actions, and make credit ratings of AAA, AA, A, BBB, BB, etc. according to the current situation and future forecasts of the enterprise; define suggestion actions, and give corresponding improvement suggestions when the enterprise needs to enhance financial management, improve market performance, and improve social reputation; define report actions, generate enterprise credit assessment reports, and summarize the analysis results. Finally, design a reward function based on the actual credit status of the enterprise. Set the corresponding target value or interval, give positive rewards when the enterprise indicators meet the standards, and give negative penalties when the indicators deviate from the targets. The indicator settings include but are not limited to financial indicators such as debt-to-asset ratio, current ratio, net profit, market performance indicators such as market share, operating income growth rate, customer satisfaction, social reputation indicators such as public opinion evaluation, social responsibility fulfillment, and environmental protection, and credit risk indicators such as the enterprise's default record, litigation, and regulatory penalties.
[0098] The model optimization module is used to perform the following: For the Markov decision model, action evaluation and parameter optimization are performed through the Q-learning method.
[0099] As a specific implementation, for the Markov decision model, the model optimization module is used to perform action evaluation and parameter optimization through the Q-learning method as follows:
[0100] (1) Building a learning model based on a convolutional neural network model, the learning model includes an action selection model and an action evaluation model with the same structure, and initializing the action selection model and the action evaluation model;
[0101] (2) Initialization state s t , based on the greedy strategy, select action a t , the greedy strategy is: each time a new action is selected with a probability of ε, and the action corresponding to the maximum Q value output by the action selection model is selected with a probability of 1-ε, and the value range of ε is (0,1);
[0102] (3) Execute action a t Get rewards t ;
[0103] (4) The status is updated to s t+1 ;
[0104] (5) {s t ,a t ,r t ,s t+1} is stored as a four-tuple in the experience pool;
[0105] (6) Determine whether the target number of times has been reached. If not, reselect action a based on the greedy strategy t If yes, update the network parameters of the action selection model and the action evaluation model. When updating the network parameters of the action evaluation model, copy the updated network parameters of the action selection model to the action evaluation model. When updating the network parameters of the action selection model, the agent randomly extracts n data from the experience pool, updates the parameters of the action selection model, and reduces the value of the exploration probability ε in the greedy strategy.
[0106] (7) Determine whether the termination condition is met. If not, reinitialize state s t , if so, output the optimal strategy.
[0107] This embodiment is based on the reinforcement learning framework of the ε-greedy strategy. The corporate credit brand value evaluation model can dynamically balance between exploration and utilization, continuously optimize the evaluation strategy, and improve the understanding and prediction ability of the corporate credit status. When using the ε-greedy strategy for decision-making, the exploration action is usually randomly selected with probability ε, and the current optimal action is selected with probability 1-ε. The Q-learning algorithm is used to continuously update the expected cumulative reward, and finally an optimal state-action value function Q(s,a) is obtained, where the update rule is: Q(s,a)=Q(s,a)+α[r+γ*max_a'Q(s',a')-Q(s,a)], α is the learning rate, and γ is the discount factor. After that, training is performed based on the neural network model, and the exploration probability ε in the ε-greedy strategy is dynamically adjusted, and gradually reduced to improve the utilization efficiency.
[0108] The system of this embodiment can execute the method disclosed in Embodiment 1 to implement corporate credit brand value evaluation.
[0109] The present invention has been shown and described in detail above through the accompanying drawings and preferred embodiments. However, the present invention is not limited to these disclosed embodiments. Based on the above multiple embodiments, those skilled in the art can know that the means in the above different embodiments can be combined to obtain more embodiments of the present invention, and these embodiments are also within the protection scope of the present invention.
Claims
1. A method for evaluating corporate credit brand value based on big data, characterized in that: The steps include: Data collection and processing: Collect enterprise data from multiple data sources, preprocess the enterprise data, build a data set based on the preprocessed enterprise data, and segment the enterprise data in the data set into training set, validation set, and test set; Problem construction: Construct a Markov decision model for corporate credit brand value assessment. For the Markov decision process, define the multi-dimensional characteristics of the corporate credit brand as the state, define optional credit assessment decisions as actions, and define rewards based on the actual credit situation of the enterprise; Model optimization: For the Markov decision model, action evaluation and parameter optimization are performed through the Q-learning method.
2. The enterprise credit brand value evaluation method based on big data according to claim 1 is characterized in that: Preprocessing of enterprise data includes the following steps: Clean enterprise data to remove missing values, outliers and duplicate data; Standardize enterprise data from different sources and in different data formats; Desensitize enterprise data; Based on the needs of corporate brand value assessment, feature selection and feature engineering are performed on corporate data to extract effective features. Includes feature combination and feature transformation.
3. The enterprise credit brand value evaluation method based on big data according to claim 1 is characterized in that: For the Markov decision model, action evaluation and parameter optimization are performed using the Q-learning method, which includes the following steps: A learning model is constructed based on a convolutional neural network model. The learning model includes an action selection model and an action evaluation model with the same structure. The action selection model and the action evaluation model are initialized. Initialization state t , based on the greedy strategy, select action a t , the greedy strategy is: each time a new action is selected with a probability of ε, and the action corresponding to the maximum Q value output by the action selection model is selected with a probability of 1-ε, and the value range of ε is (0,1); Execute action a t Get rewards t ; Status updated to s t+1 ; {s t ,a t ,r t ,s t+1 } is stored as a four-tuple in the experience pool; Determine whether the target number of times has been reached. If not, reselect action a based on the greedy strategy t If yes, update the network parameters of the action selection model and the action evaluation model. When updating the network parameters of the action evaluation model, copy the updated network parameters of the action selection model to the action evaluation model. When updating the network parameters of the action selection model, the agent randomly extracts n data from the experience pool, updates the parameters of the action selection model, and reduces the value of the exploration probability ε in the greedy strategy. Determine whether the termination condition is met, if not, reinitialize state s t , if so, output the optimal strategy.
4. The enterprise credit brand value evaluation method based on big data according to claim 3 is characterized in that: The learning model includes input layer, hidden layer and output layer; The input of the action selection model is the state s t , the output is the Q value corresponding to n actions, respectively {Q(s t ,a1),Q(s t ,a2),…,Q(s t ,a n )}; The input of the action evaluation model is the state s t+1 , the output is the Q value corresponding to n actions, respectively {Q(s t+1 ,a1),Q(s t+1 ,a2),…,Q(s t+1 ,a n )}; With r t+1 +γQ(s t+1 ,a i ), i = 1, 2, ..., n as the actual value to revise the action selection model, r t+1 Indicates that from state s t is the real-time excitation after reaching the state, and γ is the delay coefficient.
5. The enterprise credit brand value evaluation method based on big data according to claim 3 is characterized in that: When defining rewards based on the actual credit situation of the enterprise, set enterprise indicators and set target values or ranges for each enterprise indicator. When the enterprise indicator meets the target value or range, positive rewards are given. When the enterprise indicator deviates from the target value or range, negative penalties are imposed. Corporate indicators include financial indicators, market performance indicators, social reputation indicators and credit risk indicators. Among them, financial indicators include debt-to-asset ratio, current ratio and net profit, market performance indicators include market share, operating income growth rate and customer satisfaction, social reputation indicators include public opinion evaluation, social responsibility fulfillment and environmental protection, and credit risk indicators include default records, litigation situations and regulatory penalties.
6. A corporate credit brand value evaluation system based on big data, characterized in that: Used to implement enterprise brand value assessment through a method for assessing enterprise credit brand value based on big data as described in any one of claims 1 to 5, the system comprising a data collection and processing module, a problem building module and a model optimization module; The data collection and processing module is used to perform the following: collect enterprise data from multiple data sources, perform data preprocessing on the enterprise data, construct a data set based on the preprocessed enterprise data, and perform data segmentation on the enterprise data in the data set, dividing the data set into a training set, a validation set, and a test set; The problem building module is used to perform the following: construct a Markov decision model for corporate credit brand value assessment. For the Markov decision process, define the multi-dimensional characteristics of the corporate credit brand as the state, define the optional credit assessment decision as the action, and define the reward according to the actual credit situation of the enterprise; The model optimization module is used to perform the following: For the Markov decision model, action evaluation and parameter optimization are performed through the Q-learning method.
7. A corporate credit brand value evaluation system based on big data, characterized in that: The data collection and processing module is used to perform the following pre-processing of enterprise data: Clean enterprise data to remove missing values, outliers and duplicate data; Standardize enterprise data from different sources and in different data formats; Desensitize enterprise data; Based on the needs of corporate brand value assessment, feature selection and feature engineering are performed on corporate data to extract effective features. Includes feature combination and feature transformation.
8. A corporate credit brand value evaluation system based on big data, characterized in that: For the Markov decision model, the model optimization module is used to perform action evaluation and parameter optimization through the Q-learning method as follows: A learning model is constructed based on a convolutional neural network model. The learning model includes an action selection model and an action evaluation model with the same structure. The action selection model and the action evaluation model are initialized. Initialization state t , based on the greedy strategy, select action a t , the greedy strategy is: each time a new action is selected with a probability of ε, and the action corresponding to the maximum Q value output by the action selection model is selected with a probability of 1-ε, and the value range of ε is (0,1); Execute action a t Get rewards t ; Status updated to s t+1 ; {s t ,a t ,r t ,s t+1 } is stored as a four-tuple in the experience pool; Determine whether the target number of times has been reached. If not, reselect action a based on the greedy strategy t If yes, update the network parameters of the action selection model and the action evaluation model. When updating the network parameters of the action evaluation model, copy the updated network parameters of the action selection model to the action evaluation model. When updating the network parameters of the action selection model, the agent randomly extracts n data from the experience pool, updates the parameters of the action selection model, and reduces the value of the exploration probability ε in the greedy strategy. Determine whether the termination condition is met, if not, reinitialize state s t , if so, output the optimal strategy.
9. A corporate credit brand value evaluation system based on big data, characterized in that: The learning model includes input layer, hidden layer and output layer; The input of the action selection model is the state s t , the output is the Q value corresponding to n actions, respectively {Q(s t ,a1),Q(s t ,a2),…,Q(s t ,a n )}; The input of the action evaluation model is the state s t+1 , the output is the Q value corresponding to n actions, respectively {Q(s t+1 ,a1),Q(s t+1 ,a2),…,Q(s t+1 ,a n )}; With r t+1 +γQ(s t+1 ,a i ), i = 1, 2, ..., n as the actual value to revise the action selection model, r t+1 Indicates that from state s t is the real-time excitation after reaching the state, and γ is the delay coefficient.
10. A corporate credit brand value evaluation system based on big data, characterized in that: When defining rewards based on the actual credit situation of the enterprise, set enterprise indicators and set target values or ranges for each enterprise indicator. When the enterprise indicator meets the target value or range, positive rewards are given. When the enterprise indicator deviates from the target value or range, negative penalties are imposed. Corporate indicators include financial indicators, market performance indicators, social reputation indicators and credit risk indicators. Among them, financial indicators include debt-to-asset ratio, current ratio and net profit, market performance indicators include market share, operating income growth rate and customer satisfaction, social reputation indicators include public opinion evaluation, social responsibility fulfillment and environmental protection, and credit risk indicators include default records, litigation situations and regulatory penalties.
