Financial risk assessment and prediction system and method based on AI

Abstract

The invention provides a financial risk assessment and prediction system and method based on AI, and relates to the technical field of risk assessment and prediction, social media public opinion information is acquired, and a public opinion association graph of a target enterprise is constructed according to each piece of public opinion data in the social media public opinion information; extracting all abnormal comment features in the social media public opinion information, performing abnormal behavior node extraction on the public opinion association graph based on all the abnormal comment features to obtain a plurality of abnormal behavior nodes, and determining abnormal public opinion features according to the node dissimilation degree of each abnormal behavior node; obtaining abnormal behavior information of the target enterprise, and performing feature extraction on the abnormal behavior information to obtain abnormal behavior features; and performing behavior prediction through the abnormal public opinion features and the abnormal behavior features to obtain a behavior prediction result of the target enterprise, and evaluating the financial risk of the target enterprise by using the behavior prediction result. According to the method, the features in the social media public opinion data can be effectively extracted, so that overfitting of the prediction model is avoided.

Description

Technical Field

[0001] The present application relates to the technical field of risk assessment and prediction, and more specifically, to an AI-based financial risk assessment and prediction system and method. Background Art

[0002] AI-based financial risk assessment and prediction is an advanced method that uses artificial intelligence technology to analyze massive amounts of financial data, mine potential risks, and predict future risk trends. Its core is to provide enterprises and financial institutions with efficient and accurate risk assessment tools through a data-driven approach combined with the complexity and dynamism of the financial market.

[0003] However, in AI-based financial risk assessment and prediction, financial data sources are scattered and have different standards. Social media public opinion data is highly unstructured and requires complex natural language processing technology. In addition, financial risk prediction requires the model to be sensitive to abnormal events, but at the same time, it must also avoid overfitting that leads to a decrease in generalization ability. Therefore, how to effectively extract features from social media public opinion data to avoid overfitting of prediction models has become a difficult problem faced by the industry. Summary of the invention

[0004] The present application provides an AI-based financial risk assessment and prediction system and method, which can effectively extract features from social media public opinion data to avoid overfitting of the prediction model.

[0005] In a first aspect, the present application provides an AI-based financial risk assessment and prediction method, the prediction method comprising the following steps: Obtaining social media public opinion information related to the target enterprise in the financial market, and constructing a public opinion correlation graph of the target enterprise based on each piece of public opinion data in the social media public opinion information; Extract all abnormal comment features from the social media public opinion information, extract abnormal behavior nodes from the public opinion association graph based on all abnormal comment features, obtain multiple abnormal behavior nodes, and then determine the abnormal public opinion features for the target enterprise according to the node alienation degree of each abnormal behavior node; Acquire abnormal behavior information of a target enterprise, perform feature extraction on the abnormal behavior information, and obtain abnormal behavior features of the target enterprise; Behavior prediction is performed through the abnormal public opinion characteristics and the abnormal behavior characteristics to obtain the behavior prediction results of the target enterprise, and the financial risk of the target enterprise is evaluated using the behavior prediction results.

[0006] In this embodiment, social media public opinion information related to the target enterprise in the financial market is obtained through a data crawler tool.

[0007] In this embodiment, it also includes: performing data cleaning on the social media public opinion information, and then converting the social media public opinion information into structured data.

[0008] In this embodiment, constructing a public opinion association graph of a target enterprise based on each piece of public opinion data in the social media public opinion information specifically includes: Using natural language processing to extract entities from each piece of public opinion data in the social media public opinion information, to obtain entities for each piece of public opinion data; Using a pre-trained relationship extraction model, respectively extract entity relationships from each piece of public opinion data in the social media public opinion information to obtain the entity relationship of each piece of public opinion data; Build a public opinion correlation graph for the target enterprise through all entities and all entity relationships.

[0009] In this embodiment, extracting all abnormal comment features in the social media public opinion information specifically includes: Obtain the target company’s social media text database; Determine the text similarity between each piece of public opinion data in the social media public opinion information and the social media text database respectively; The public opinion data with text similarity higher than the preset threshold is regarded as abnormal public opinion data, and then all abnormal public opinion data are obtained; The abnormal comment features of each abnormal public opinion data are extracted respectively, and then all abnormal comment features in the social media public opinion information are obtained.

[0010] In this embodiment, the abnormal behavior node extraction is performed on the public opinion association graph based on all abnormal comment features, and the entity node corresponding to each abnormal comment feature in the public opinion association graph is used as the abnormal behavior node.

[0011] In this embodiment, a feature extraction model based on machine learning is used to extract features from the abnormal behavior information.

[0012] In this embodiment, behavior prediction is performed by using the abnormal public opinion features and the abnormal behavior features to obtain the behavior prediction results of the target enterprise, specifically including: Get a pre-trained LSTM network model; The abnormal public opinion features and the abnormal behavior features are input into the long short-term memory network model to perform behavior prediction, and obtain the behavior prediction results of the target enterprise.

[0013] In this embodiment, using the behavior prediction results to evaluate the financial risk of the target enterprise is to input the behavior prediction results into a risk assessment model to evaluate the financial risk of the target enterprise, thereby obtaining a financial risk assessment result for the target enterprise.

[0014] In a second aspect, the present application provides an AI-based financial risk assessment and prediction system for executing an AI-based financial risk assessment and prediction method, the prediction system comprising: A correlation graph construction module is used to obtain social media public opinion information related to the target enterprise in the financial market, and to construct a public opinion correlation graph of the target enterprise based on each piece of public opinion data in the social media public opinion information; A public opinion feature extraction module is used to extract all abnormal comment features from the social media public opinion information, extract abnormal behavior nodes from the public opinion association graph based on all abnormal comment features, obtain multiple abnormal behavior nodes, and then determine the abnormal public opinion features for the target enterprise according to the node alienation degree of each abnormal behavior node; A behavior feature extraction module is used to obtain abnormal behavior information of a target enterprise, perform feature extraction on the abnormal behavior information, and obtain abnormal behavior features of the target enterprise; The risk assessment module is used to perform behavior prediction based on the abnormal public opinion characteristics and the abnormal behavior characteristics, obtain the behavior prediction results of the target enterprise, and use the behavior prediction results to assess the financial risks of the target enterprise.

[0015] The technical solution provided by the embodiments disclosed in this application has the following beneficial effects: By acquiring social media public opinion information related to the target enterprise in the financial market, a public opinion association graph of the target enterprise is constructed according to each public opinion data in the social media public opinion information; all abnormal comment features in the social media public opinion information are extracted, and abnormal behavior nodes are extracted from the public opinion association graph based on all abnormal comment features to obtain multiple abnormal behavior nodes, and then the abnormal public opinion features for the target enterprise are determined according to the node alienation degree of each abnormal behavior node; the abnormal behavior information of the target enterprise is acquired, and features are extracted from the abnormal behavior information to obtain the abnormal behavior features of the target enterprise; behavior prediction is performed through the abnormal public opinion features and the abnormal behavior features to obtain the behavior prediction results of the target enterprise, and the financial risks of the target enterprise are evaluated using the behavior prediction results.

[0016] It can be seen that in this application, firstly, by constructing a public opinion association graph, the public opinion relationship network of the target enterprise in the financial market can be visualized, and the insight and analysis capabilities of public opinion data can be significantly enhanced. When extracting data features, the public opinion association graph can provide high-quality structured information, improve the generalization ability of the model, and effectively avoid the overfitting problem; then, by extracting abnormal comment features, constructing abnormal behavior nodes, and determining abnormal public opinion features according to the degree of alienation of the nodes, the prediction accuracy and handling capabilities of public opinion crises can be effectively improved; further, by obtaining the abnormal behavior information of the target enterprise and performing feature extraction, it can help the target enterprise establish a more sensitive and highly accurate risk prediction model; finally, by using abnormal public opinion features and abnormal behavior features to predict behavior, and using the prediction results to evaluate the financial risks of the enterprise, it can effectively identify potential risks in advance and help the enterprise adjust its strategy in a timely manner.

[0017] In summary, the technical solution adopted in this application can effectively extract features from social media public opinion data to avoid overfitting of the prediction model. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 It is a flowchart of the AI-based financial risk assessment and prediction method provided in this application; Figure 2 It is a schematic diagram of the process of constructing a public opinion association graph of a target enterprise according to the application; Figure 3 It is based on the extraction of all abnormal comment features in social media public opinion information provided by this application Schematic diagram of the process; Figure 4 This is a module structure diagram of the AI-based financial risk assessment and prediction system provided in this application. DETAILED DESCRIPTION

[0019] The technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application.

[0020] The embodiment of the present application provides an AI-based financial risk assessment and prediction system and method, the core of which is to obtain social media public opinion information related to the target enterprise in the financial market, and construct a public opinion association graph of the target enterprise based on each public opinion data in the social media public opinion information; extract all abnormal comment features in the social media public opinion information, and extract abnormal behavior nodes from the public opinion association graph based on all abnormal comment features to obtain multiple abnormal behavior nodes, and then determine the abnormal public opinion features of the target enterprise according to the node alienation degree of each abnormal behavior node; obtain the abnormal behavior information of the target enterprise, extract features from the abnormal behavior information, and obtain the abnormal behavior features of the target enterprise; perform behavior prediction through the abnormal public opinion features and the abnormal behavior features to obtain the behavior prediction results of the target enterprise, and use the behavior prediction results to evaluate the financial risks of the target enterprise. The above scheme can effectively extract features from social media public opinion data to avoid overfitting of the prediction model.

[0021] Embodiment 1: In order to better understand the above technical solution, the above technical solution will be described in detail below in conjunction with the accompanying drawings and specific implementation methods. Figure 1 As shown in FIG. 1 , this figure is an exemplary flow chart of an AI-based financial risk assessment and prediction method according to this embodiment of the present application. The prediction method includes the following steps: In step S1, social media public opinion information related to the target enterprise in the financial market is obtained, and a public opinion association graph of the target enterprise is constructed based on each piece of public opinion data in the social media public opinion information.

[0022] In specific implementation, social media public opinion information related to the target enterprise in the financial market can be obtained through data crawling tools. It should be noted that in this application, social media public opinion information includes multiple public opinion data, and public opinion data is a data text composed of entities, events and actions.

[0023] In this embodiment, after obtaining the social media public opinion information, it also includes: performing data cleaning on the social media public opinion information, and then converting the social media public opinion information into structured data. In specific implementation, the existing data cleaning technology can be used to clean the social media public opinion information to ensure the accuracy of the social media public opinion information, so that the social media public opinion information after data cleaning can be converted into structured data, which is convenient for subsequent correlation analysis.

[0024] Preferably, in this embodiment, a public opinion association graph of the target enterprise is constructed based on each piece of public opinion data in the social media public opinion information. Figure 2 As shown, this figure is a schematic diagram of the process of constructing a public opinion association graph of a target enterprise in some embodiments of the present application. In this embodiment, constructing a public opinion association graph of a target enterprise can be implemented by the following steps: First, in step S11, natural language processing is used to extract entities from each piece of public opinion data in the social media public opinion information to obtain an entity of each piece of public opinion data; Then, in step S12, the pre-trained relationship extraction model is used to extract entity relationships from each piece of public opinion data in the social media public opinion information to obtain the entity relationship of each piece of public opinion data; Finally, in step S13, a public opinion association graph of the target enterprise is constructed through all entities and all entity relationships.

[0025] In specific implementation, first, natural language processing can be used to extract entities from each piece of public opinion data in social media public opinion information, that is, a pre-trained named entity recognition (NER) model can be used to process each piece of public opinion data, so as to extract the entities of each piece of public opinion data, such as organizations and names; then, a pre-trained relationship extraction model can be used to extract entity relationships from each piece of public opinion data in social media public opinion information, that is, a pre-trained relationship extraction model can be used to process each piece of public opinion data (for example, using BERT+relation classification head or OpenAI GPT model), so as to extract the relationship between the entity and the target enterprise for each pair of entities, such as cooperative relationship, competitive relationship, investment relationship and influence relationship; finally, a public opinion association graph of the target enterprise can be constructed through all entities and all entity relationships, that is, the entities are taken as nodes in the graph, and the entity relationships are taken as edges in the graph. The data correlation between the public opinion data corresponding to the two connected entities can be calculated by cosine similarity. The data correlation represents the degree of association between the entities. The data correlation is taken as the value of the connected edge, so as to construct a public opinion association graph of the target enterprise.

[0026] It should be noted that by constructing a public opinion correlation graph, the public opinion relationship network of the target enterprise in the financial market can be visualized, significantly enhancing the insight and analysis capabilities of public opinion data. When extracting data features, the public opinion correlation graph can provide high-quality structured information, reduce redundancy and noise, and improve the generalization ability of the model, thereby effectively avoiding overfitting problems and improving the reliability and accuracy of financial risk assessment predictions.

[0027] In step S2, all abnormal comment features in the social media public opinion information are extracted, and abnormal behavior nodes are extracted from the public opinion association graph based on all abnormal comment features to obtain multiple abnormal behavior nodes, and then the abnormal public opinion features for the target enterprise are determined according to the node alienation degree of each abnormal behavior node.

[0028] Preferably, in this embodiment, all abnormal comment features of the social media public opinion information are extracted, and reference Figure 2As shown, this figure is a schematic diagram of the process of extracting all abnormal comment features from social media public opinion information in some embodiments of the present application. In this embodiment, extracting all abnormal comment features from social media public opinion information can be achieved by using the following steps: First, in step S21, a social media text database of the target enterprise is obtained; Then, in step S22, the text similarity between each piece of public opinion data in the social media public opinion information and the social media text database is determined respectively; Then, in step S23, the public opinion data with text similarity higher than a preset threshold is taken as abnormal public opinion data, and then all abnormal public opinion data are obtained; Finally, in step S24, the abnormal comment features of each abnormal public opinion data are extracted respectively, and then all abnormal comment features in the social media public opinion information are obtained.

[0029] In specific implementation, first, the social media text database of the target enterprise can be obtained through the management system of the target enterprise; then, the text similarity between each piece of public opinion data in the social media public opinion information and the social media text database can be determined separately by means of Euclidean distance, and the text similarity represents the degree of text similarity between the public opinion data and the social media text database; further, the public opinion data with text similarity higher than a preset threshold can be regarded as abnormal public opinion data, wherein the threshold can be preset through historical experience and data analysis, and all abnormal public opinion data can be obtained through the above method.

[0030] In addition, in the specific implementation, the abnormal comment features of each abnormal public opinion data can be extracted separately; wherein, the abnormal comment feature is a behavioral feature used to indicate the presence of abnormal comments in the abnormal public opinion data. The larger the abnormal comment feature, the higher the abnormal risk level of the abnormal public opinion data. First, a suitable machine learning algorithm can be selected to extract abnormal comment features according to the characteristics of the data and business needs. The feature extraction model based on machine learning selected in this application is a support vector machine. In actual implementation, other feature extraction models based on machine learning can also be selected, which are not limited here. Then, for each abnormal public opinion data, the abnormal public opinion data is input into the feature extraction model for feature extraction, so as to obtain the abnormal comment features of the abnormal public opinion data. All abnormal comment features in the social media public opinion information can be obtained in the above manner.

[0031] In this embodiment, the abnormal behavior node extraction is performed on the public opinion association graph based on all abnormal comment features, and the entity node corresponding to each abnormal comment feature in the public opinion association graph is used as an abnormal behavior node, thereby obtaining multiple abnormal behavior nodes.

[0032] In this embodiment, the abnormal public opinion characteristics for the target enterprise are determined according to the node alienation degree of each abnormal behavior node. In specific implementation, first, for each abnormal behavior node, all edges connected to the abnormal behavior node can be obtained in the public opinion association graph, and the difference between the mean of the values ​​of all connected edges and 1 is used as the node alienation degree of the abnormal behavior node. The node alienation degree represents the relative abnormal change degree of the abnormal behavior node. The node alienation degree of each abnormal behavior node can be obtained in the above manner; finally, the standard deviation of the node alienation degree of all abnormal behavior nodes can be used as the abnormal public opinion characteristic for the target enterprise. The abnormal public opinion characteristic is a characteristic used to represent the degree of influence of abnormal public opinion on the target enterprise.

[0033] It should be noted that by extracting abnormal comment features, constructing abnormal behavior nodes, and determining abnormal public opinion features based on the degree of alienation of the nodes, the prediction accuracy and handling capabilities of public opinion crises can be effectively improved. It can also effectively avoid overfitting when constructing prediction models, improve the stability and generalization ability of the model, and thus provide companies with more accurate and timely public opinion risk warnings and management strategies in practical applications.

[0034] In step S3, the abnormal behavior information of the target enterprise is obtained, and features are extracted from the abnormal behavior information to obtain abnormal behavior features of the target enterprise.

[0035] In specific implementation, the abnormal behavior information of the target enterprise can be obtained through financial data monitoring, that is, analyzing abnormal fluctuations in financial statements, such as sharp changes in indicators such as income, cost, profit, liabilities, or issues raised in the audit report; it can also be through stock market behavior monitoring, that is, monitoring the sharp fluctuations in corporate stock prices and abnormal changes in trading volume, especially abnormal stock performance after certain events or announcements.

[0036] In this embodiment, a feature extraction model based on machine learning can be used to extract features from the abnormal behavior information, thereby obtaining abnormal behavior features of the target enterprise. In specific implementation, the feature extraction model based on machine learning in step S2 can be used to extract features from the abnormal behavior information, thereby obtaining abnormal behavior features of the target enterprise, which are behavioral features used to represent abnormal activities of the target enterprise.

[0037] It should be noted that by obtaining the target enterprise’s abnormal behavior information and performing feature extraction, it is possible to help the target enterprise establish a more sensitive and highly accurate risk prediction model. Especially in social media public opinion analysis, the extraction of abnormal behavior features can help the model focus on key and practical features, reduce the risk of overfitting, and improve the model’s generalization and robustness.

[0038] In step S4, behavior prediction is performed using the abnormal public opinion features and the abnormal behavior features to obtain a behavior prediction result of the target enterprise, and the financial risk of the target enterprise is evaluated using the behavior prediction result.

[0039] In this embodiment, the behavior prediction is performed by using the abnormal public opinion characteristics and the abnormal behavior characteristics to obtain the behavior prediction result of the target enterprise, which can be specifically obtained in the following manner, namely: Get a pre-trained LSTM network model; The abnormal public opinion features and the abnormal behavior features are input into the long short-term memory network model to perform behavior prediction, and obtain the behavior prediction results of the target enterprise.

[0040] In the specific implementation, first, a pre-trained long short-term memory network (LSTM) model can be obtained. The long short-term memory network is a powerful recursive neural network variant, which is particularly suitable for processing time series data or data with long-term dependencies. It predicts future behavioral trends by learning historical behavior patterns. In financial risk assessment, the long short-term memory network is particularly suitable for analyzing abnormal behaviors and public opinion data of enterprises, because financial market and social media information usually have temporal and nonlinear relationships; then, the abnormal public opinion features and abnormal behavior features can be input into the long short-term memory network model for behavior prediction. The long short-term memory network model will generate a behavior prediction result for a target enterprise, which includes the future behavior feature information of the target enterprise.

[0041] In this embodiment, using the behavior prediction results to evaluate the financial risk of the target enterprise is to input the behavior prediction results into the risk assessment model to evaluate the financial risk of the target enterprise, thereby obtaining the financial risk assessment results of the target enterprise. In specific implementation, the risk assessment model determines the risk level faced by the enterprise based on established rules and machine learning algorithms (such as decision trees, support vector machines, etc.). For example, the financial risk assessment results of the target enterprise can be the current risk level (such as high risk, medium risk or low risk) and the specific risk type (market risk, credit risk, operational risk, etc.).

[0042] It should be noted that by using abnormal public opinion characteristics and abnormal behavior characteristics to predict behavior, and using the prediction results to assess the company's financial risks, potential risks can be effectively identified in advance and help companies adjust their strategies in a timely manner.

[0043] It can be seen that in this application, firstly, by constructing a public opinion association graph, the public opinion relationship network of the target enterprise in the financial market can be visualized, and the insight and analysis capabilities of public opinion data can be significantly enhanced. When extracting data features, the public opinion association graph can provide high-quality structured information, improve the generalization ability of the model, and effectively avoid the overfitting problem; then, by extracting abnormal comment features, constructing abnormal behavior nodes, and determining abnormal public opinion features according to the degree of alienation of the nodes, the prediction accuracy and handling capabilities of public opinion crises can be effectively improved; further, by obtaining the abnormal behavior information of the target enterprise and performing feature extraction, it can help the target enterprise establish a more sensitive and highly accurate risk prediction model; finally, by using abnormal public opinion features and abnormal behavior features to predict behavior, and using the prediction results to evaluate the financial risks of the enterprise, it can effectively identify potential risks in advance and help the enterprise adjust its strategy in a timely manner.

[0044] In summary, the technical solution adopted in this application can effectively extract features from social media public opinion data to avoid overfitting of the prediction model.

[0045] Embodiment 2: This application provides a financial risk assessment and prediction system based on AI, referring to Figure 4 As shown, this figure is a schematic diagram of the AI-based financial risk assessment and prediction system shown in this embodiment of the present application, and the prediction system includes: The association graph construction module 100 is used to obtain social media public opinion information related to the target enterprise in the financial market, and to construct a public opinion association graph of the target enterprise based on each piece of public opinion data in the social media public opinion information; The public opinion feature extraction module 200 is used to extract all abnormal comment features from the social media public opinion information, extract abnormal behavior nodes from the public opinion association graph based on all abnormal comment features, obtain multiple abnormal behavior nodes, and then determine the abnormal public opinion features for the target enterprise according to the node alienation degree of each abnormal behavior node; The behavior feature extraction module 300 is used to obtain abnormal behavior information of a target enterprise, perform feature extraction on the abnormal behavior information, and obtain abnormal behavior features of the target enterprise; The risk assessment module 400 is used to perform behavior prediction based on the abnormal public opinion characteristics and the abnormal behavior characteristics, obtain the behavior prediction results of the target enterprise, and use the behavior prediction results to assess the financial risks of the target enterprise.

[0046] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems) and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

Claims

1. A financial risk assessment and prediction method based on AI, characterized in that: The prediction method comprises the following steps: Obtaining social media public opinion information related to the target enterprise in the financial market, and constructing a public opinion correlation graph of the target enterprise based on each piece of public opinion data in the social media public opinion information; Extract all abnormal comment features from the social media public opinion information, extract abnormal behavior nodes from the public opinion association graph based on all abnormal comment features, obtain multiple abnormal behavior nodes, and then determine the abnormal public opinion features for the target enterprise according to the node alienation degree of each abnormal behavior node; Acquire abnormal behavior information of a target enterprise, perform feature extraction on the abnormal behavior information, and obtain abnormal behavior features of the target enterprise; Behavior prediction is performed through the abnormal public opinion characteristics and the abnormal behavior characteristics to obtain the behavior prediction results of the target enterprise, and the financial risk of the target enterprise is evaluated using the behavior prediction results.

2. The AI-based financial risk assessment and prediction method according to claim 1, characterized in that: Use data crawling tools to obtain social media public opinion information related to target companies in the financial market.

3. The AI-based financial risk assessment and prediction method according to claim 1, characterized in that: Also includes: The social media public opinion information is cleaned and then converted into structured data.

4. The AI-based financial risk assessment and prediction method according to claim 1, characterized in that: Constructing a public opinion correlation graph of a target enterprise based on each piece of public opinion data in the social media public opinion information specifically includes: Using natural language processing to extract entities from each piece of public opinion data in the social media public opinion information, to obtain entities for each piece of public opinion data; Using a pre-trained relationship extraction model, respectively extract entity relationships from each piece of public opinion data in the social media public opinion information to obtain the entity relationship of each piece of public opinion data; Build a public opinion correlation graph for the target enterprise through all entities and all entity relationships.

5. The AI-based financial risk assessment and prediction method according to claim 1, characterized in that: Extracting all abnormal comment features from the social media public opinion information specifically includes: Obtain the target company’s social media text database; Determine the text similarity between each piece of public opinion data in the social media public opinion information and the social media text database respectively; The public opinion data with text similarity higher than the preset threshold is regarded as abnormal public opinion data, and then all abnormal public opinion data are obtained; The abnormal comment features of each abnormal public opinion data are extracted respectively, and then all abnormal comment features in the social media public opinion information are obtained.

6. The AI-based financial risk assessment and prediction method according to claim 1, characterized in that: The abnormal behavior node extraction of the public opinion association graph based on all abnormal comment features is to use the entity node corresponding to each abnormal comment feature in the public opinion association graph as the abnormal behavior node.

7. The AI-based financial risk assessment and prediction method according to claim 1, characterized in that: A feature extraction model based on machine learning is used to extract features from the abnormal behavior information.

8. The AI-based financial risk assessment and prediction method according to claim 1, characterized in that: The behavior prediction results of the target enterprise obtained by performing behavior prediction based on the abnormal public opinion characteristics and the abnormal behavior characteristics specifically include: Get a pre-trained LSTM network model; The abnormal public opinion features and the abnormal behavior features are input into the long short-term memory network model to perform behavior prediction, and obtain the behavior prediction results of the target enterprise.

9. The AI-based financial risk assessment and prediction method according to claim 1, characterized in that: Using the behavior prediction results to evaluate the financial risk of the target enterprise is to input the behavior prediction results into a risk assessment model to evaluate the financial risk of the target enterprise, thereby obtaining a financial risk assessment result for the target enterprise.

10. An AI-based financial risk assessment and prediction system, used to execute an AI-based financial risk assessment and prediction method as claimed in any one of claims 1 to 9, characterized in that: The prediction system comprises: A correlation graph construction module is used to obtain social media public opinion information related to the target enterprise in the financial market, and to construct a public opinion correlation graph of the target enterprise based on each piece of public opinion data in the social media public opinion information; A public opinion feature extraction module is used to extract all abnormal comment features from the social media public opinion information, extract abnormal behavior nodes from the public opinion association graph based on all abnormal comment features, obtain multiple abnormal behavior nodes, and then determine the abnormal public opinion features for the target enterprise according to the node alienation degree of each abnormal behavior node; A behavior feature extraction module is used to obtain abnormal behavior information of a target enterprise, perform feature extraction on the abnormal behavior information, and obtain abnormal behavior features of the target enterprise; The risk assessment module is used to perform behavior prediction based on the abnormal public opinion characteristics and the abnormal behavior characteristics, obtain the behavior prediction results of the target enterprise, and use the behavior prediction results to assess the financial risks of the target enterprise.

Images