Enterprise customer risk control management method and system based on AI intelligence

Abstract

The application provides an AI intelligent-based enterprise customer risk control management method and system, relates to the technical field of risk control management, and comprises the following steps: collecting data of an enterprise customer, establishing a collected data set, including internal interaction data and external public data; capturing the behavior of the enterprise customer, establishing a behavior path atlas; obtaining the static enterprise risk of the enterprise customer as an attention factor, configuring a risk trigger guide strategy based on the internal interaction data; after the enterprise customer interacts, establishing a mapping behavior feedback, and establishing a mapping group data with the corresponding risk trigger guide strategy; identifying an abnormal channel, establishing a dynamic enterprise risk; and generating a risk control management strategy. The application solves the technical problem that the prior art mostly relies on static data for risk assessment, cannot make a timely response to rapidly changing customer behavior, fails to identify and cope with potential risk behavior, and results in poor risk control management effect.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of risk control management, in particular to an enterprise customer risk control management method and system based on AI intelligence. BACKGROUND

[0002] With the expansion of the scale of enterprise customers and the complexity of market environment, traditional risk control management methods are facing more and more challenges. Many traditional risk control methods rely on static data and historical analysis to assess the credit risk and potential risk of customers. However, the risk of enterprise customers presents dynamic, complex and multi-dimensional characteristics. With the rapid development of big data and artificial intelligence technology, traditional risk control methods cannot fully meet the needs of large-scale data analysis and real-time decision-making, cannot comprehensively and accurately analyze the behavior patterns and potential risks of customers, and cannot provide comprehensive risk control decision support. SUMMARY

[0003] The application provides an enterprise customer risk control management method and system based on AI intelligence, aiming to solve the technical problem that the prior art mostly relies on static data for risk assessment and cannot respond to rapidly changing customer behavior in a timely manner, resulting in failure to identify and respond to potential risk behavior and poor risk control management effect.

[0004] The first aspect of the application provides an enterprise customer risk control management method based on AI intelligence, which comprises the following steps: collecting data of enterprise customers, establishing a collected data set, the collected data set comprising internal interaction data and external public data; capturing enterprise customer behavior based on the collected data set, and establishing a behavior path atlas based on the behavior capture result; obtaining static enterprise risk of the enterprise customer, and configuring a risk trigger guide strategy based on the internal interaction data, taking the static enterprise risk as an attention factor; after the enterprise customer interacts based on the risk trigger guide strategy, establishing a mapping behavior feedback, and establishing the mapping behavior feedback and the corresponding risk trigger guide strategy as mapping group data; identifying abnormal channels based on the mapping group data and the behavior path atlas, and establishing dynamic enterprise risk; and generating a risk control management strategy according to the static enterprise risk and the dynamic enterprise risk.

[0005] In a second aspect, the application discloses an AI intelligent-based enterprise customer risk control management system, which is used for the AI intelligent-based enterprise customer risk control management method, and comprises a data acquisition module, a behavior capturing module, a strategy triggering module, a mapping group data establishing module, an abnormal channel identification module, and a management strategy generating module.

[0006] The one or more technical solutions provided in the application have at least the following beneficial effects:

[0007] By collecting internal interaction data and external public data, various information of enterprise customers can be comprehensively integrated, which provides multi-dimensional data support for subsequent risk control management and behavior analysis; by capturing the behavior of the collected data set, the behavior pattern of enterprise customers can be mastered in detail, and the establishment of the behavior path atlas can show the interaction behavior sequence of customers, reflect the regular and abnormal behavior of enterprise customers when interacting with the enterprise system, help the risk control team intuitively understand the pattern and changes of customer behavior, and thus effectively identify potential risk behavior; by evaluating the static enterprise risk of enterprise customers, the potential risk of enterprise customers in the long term is identified, and after the static enterprise risk is taken as an attention factor, attention is focused on the features that have the greatest impact on risk control decisions, so as to configure accurate risk trigger guide strategies, help the risk control system make appropriate responses to different risk scenarios; by mapping the interaction between the risk trigger guide strategy and the enterprise customer, the risk control strategy can be adjusted according to the actual behavior of the customer, and this feedback mechanism helps the system to update the risk assessment in real time to respond to changes in customer behavior, and the establishment of the mapping group data enables the risk control system to dynamically adjust the risk management strategy according to different customer behavior feedback, thereby improving the flexibility and adaptability of risk control management; by identifying abnormal channels in the mapping group data and the behavior path atlas, abnormal paths in customer behavior can be identified, and after identifying the abnormal channels, the dynamic enterprise risk of the customer is evaluated in real time to provide timely information for risk control decisions; by combining static enterprise risk and dynamic enterprise risk, the overall risk of enterprise customers is comprehensively evaluated, and more accurate risk control management strategies are generated, so that risk control management is more comprehensive and accurate.

[0008] The above description is only a summary of the technical solutions of the present application. In order to more clearly understand the technical means of the present application, the present application can be implemented according to the content of the specification, and in order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the following specific embodiments of the present application are described. BRIEF DESCRIPTION OF DRAWINGS

[0009] Figure 1 The AI intelligent-based enterprise customer risk control management method flowchart provided by the embodiments of the present application.

[0010] Figure 2 The AI intelligent-based enterprise customer risk control management system structure diagram provided by the embodiments of the present application.

[0011] Explanation of reference numerals: data collection module 10, behavior capture module 20, strategy trigger module 30, mapping group data establishment module 40, abnormal channel identification module 50, management strategy generation module 60. DETAILED DESCRIPTION

[0012] The embodiment of the application provides an AI intelligent-based enterprise customer risk control management method and system, and solves the technical problem that most of the prior art relies on static data for risk assessment, cannot respond to rapidly changing customer behavior in time, cannot identify and cope with potential risk behavior, and cannot achieve good risk control management effect.

[0013] After introducing the basic principle of the application, various non-limiting embodiments of the application will be specifically introduced in combination with the drawings of the specification. It should be understood that the specific embodiments described herein are only used to explain the application and not to limit the application.

[0014] Embodiment one, as shown in the figure, the embodiment of the application provides an AI intelligent-based enterprise customer risk control management method, which comprises: Figure 1

[0015] Collecting data of the enterprise customer, establishing a collected data set, and the collected data set comprising internal interaction data and external public data.

[0016] Collecting relevant data of the enterprise customer, and establishing a collected data set, which contains internal interaction data and external public data, as the basis for subsequent analysis and risk control management, wherein the internal interaction data comes from the interaction between the enterprise and the customer, such as customer access record, purchase record, customer service interaction, transaction history, contract information, etc., and the internal interaction data is used to analyze the behavior pattern, demand change, transaction frequency, etc. of the customer; the external public data comes from outside the enterprise, such as comments on social media, public financial reports, news information, industry reports, etc., and the external public data can provide background information of the industry, market environment, economic situation, etc. of the customer company, helping to assess external risks.

[0017] Capturing the behavior of the enterprise customer based on the collected data set, and establishing a behavior path atlas based on the behavior capture result.

[0018] Analyzing the collected data set to capture the behavior characteristics of the enterprise customer, including the login behavior, browsing behavior, purchase behavior, service request, etc. of the user, and the goal of capture is to determine how the customer interacts with the enterprise system and identify potential risk patterns. Specifically, the customer behavior in the collected data set is arranged in time sequence to form a behavior event stream, and according to the preset behavior recognition rule, the behavior nodes, time, context, etc. are captured and extracted, for example, a specific browsing behavior, purchase behavior, etc. Each behavior is taken as a node, and the relationship between behaviors is defined as an edge, such as the order of behaviors, condition triggering, etc. to form a graph structure, and a behavior path atlas is obtained, which is used to analyze the typical behavior path of the customer and its changes.

[0019] ​Obtain a static enterprise risk of an enterprise customer, use the static enterprise risk as an attention factor, and configure a risk trigger guide strategy based on internal interaction data.

[0020] The static enterprise risk refers to the risk level of an enterprise customer at a certain point in time, which is usually obtained based on the evaluation of the financial situation, historical performance, industry risk, and other factors of the enterprise. The static enterprise risk is relatively stable and does not change much over time, providing long-term risk assessment. By analyzing the static enterprise risk, a static risk feature set is obtained, which is normalized to convert it into a feature vector suitable for AI analysis. A shallow network is used for high-dimensional embedding, and an attention score network is used to calculate the weight of each feature in risk assessment. Based on the static enterprise risk and internal interaction data, a corresponding risk trigger guide strategy is configured. The risk trigger guide strategy is activated when certain risk signs are detected, such as triggering corresponding risk management measures when transactions are frequent or abnormal.

[0021] After the enterprise customer interacts based on the risk trigger guide strategy, a mapping behavior feedback is established, and the mapping behavior feedback and the corresponding risk trigger guide strategy are established as mapping group data.

[0022] According to the risk trigger guide strategy, the enterprise customer interacts, which is triggered by the enterprise customer's actions such as clicking, purchasing, querying, etc. to trigger risk management strategies, such as automatically taking guide strategies like reminders, warnings, or limiting certain operations when certain specific behavior patterns are detected. Through real-time feedback on the enterprise customer's interaction behavior, the enterprise customer's executed behavior is recorded, and each interaction generates a mapping behavior feedback. These mapping behavior feedbacks reflect the enterprise customer's reaction when facing specific risk prompts, such as the enterprise customer choosing to continue operating after receiving a risk prompt or changing behavior according to the guide strategy. The mapping behavior feedback and the corresponding risk trigger guide strategy are established as mapping group data, which includes the relationship between the enterprise customer's mapping behavior feedback and its corresponding risk trigger guide strategy. In this way, it can be determined which strategies are effective in actual operation and which need to be adjusted.

[0023] An abnormal channel identification is performed on the mapping group data and the behavior path atlas to establish a dynamic enterprise risk.

[0024] The abnormal channel identification is performed by analyzing the mapping group data and the behavior path atlas to identify potential abnormal channels. The abnormal channel refers to a path in the enterprise customer behavior pattern that is different from the normal pattern or indicates potential risks. For example, after receiving a risk warning, the enterprise customer still chooses to ignore and continues to perform high-risk operations, indicating that the enterprise customer has a high risk level. The mapping group data and the behavior path atlas are combined for analysis to find out which behavior paths deviate from the conventional or expected paths. These deviated behavior paths indicate that the customer has risks such as fraud, malicious operation, data leakage, etc. Through the identification of the abnormal channel, the dynamic enterprise risk is established. The dynamic enterprise risk is more dependent on the behavior and real-time interaction feedback of the enterprise customer, and thus it changes with the change of the customer behavior, thereby providing more flexible and real-time risk control capability for the risk control system.

[0025] The risk control management strategy is generated according to the static enterprise risk and the dynamic enterprise risk.

[0026] The static enterprise risk and the dynamic enterprise risk are integrated, for example, by combining them through weighted analysis to evaluate the overall risk level of the enterprise customer. An algorithm is used to generate a risk control management strategy that adapts to different risk levels. The risk control management strategy is automatically adjusted according to the change of risk data, including risk warning, behavior restriction, further review, customer communication, etc. For example, the enterprise customer has a good behavior record in the past, but the recent behavior deviates, which increases the risk monitoring of the enterprise customer and restricts its transaction or operation.

[0027] Further, the abnormal channel identification of the mapping group data and the behavior path atlas to establish the dynamic enterprise risk includes:

[0028] The dynamic behavior sub-channel of the abnormal identification channel is activated. After the behavior path atlas is input into the dynamic behavior sub-channel, the behavior path atlas is deconstructed into a triple sequence of business activity nodes, timestamps, and operator information by using the analysis layer. The enterprise customer behavior pattern set is constructed according to the triple sequence. After the target behavior flow template is configured, the behavior deviation analysis of the enterprise customer behavior pattern set is performed according to the target behavior flow template to establish the deviation candidate identifier. The key behavior point identification is performed based on the deviation candidate identifier to establish the abnormal behavior trajectory graph with the key behavior point identification result and the deviation candidate identifier. The enterprise dynamic risk is established according to the abnormal behavior trajectory graph.

[0029] The dynamic behavior sub-channel in the anomaly recognition channel is activated, which means that the focus is on identifying the abnormal or unusual patterns of the enterprise customer in the behavior path graph next. The dynamic behavior sub-channel is specially used to process real-time behavior data to identify potential risks. The behavior path graph, which represents the historical record of the enterprise customer's behavior and presents the interrelationship between the enterprise customer's operations, is input into the dynamic behavior sub-channel.

[0030] In the dynamic behavior sub-channel, the behavior path graph is deconstructed by using an analysis layer, and the main task of the analysis layer is to extract valuable information from the behavior path graph. Specifically, the analysis layer converts the information in the behavior path graph into a sequence of triples, each triple containing a business activity node, a timestamp, and operator information. The business activity node refers to a specific behavior performed by the enterprise customer when interacting with the enterprise system, such as clicking a button, submitting an order, or initiating a customer service request. The timestamp refers to the specific time when the behavior occurred, indicating the time characteristics of the enterprise customer's behavior. The operator information refers to the identity information of the enterprise customer who performed the behavior, such as the customer ID or account information. By extracting and analyzing the sequence of triples, a set of enterprise customer behavior patterns is constructed based on the frequency and regularity of the behavior. The set of enterprise customer behavior patterns contains the regular patterns of the enterprise customer's behavior, which are used to identify the normal operations of the customer and potential abnormal behaviors.

[0031] A target behavior flow template is configured, which represents the expected and regular behavior flow of the enterprise customer when interacting with the enterprise system. The target behavior flow template is generated based on historical data and the results of behavior pattern analysis, aiming to define the behavior sequence and operations that the enterprise customer should follow under normal circumstances, and to depict the behaviors that the enterprise customer should perform within a certain period of time. Based on the target behavior flow template, the set of enterprise customer behavior patterns is analyzed to identify deviating behaviors from the target behavior flow template. For example, if the enterprise customer does not operate according to the normal behavior flow sequence, such as skipping the payment step and directly exiting, etc., this behavior is marked as a deviating behavior. Each deviating behavior may represent a potential risk. Based on the analysis results of the deviating behavior, a deviating candidate identifier is generated, which represents the significant difference between the enterprise customer's behavior and the expected behavior, and is a potential indication of abnormal behavior.

[0032] The key behavior points refer to abnormal nodes appearing in the behavior path of the enterprise customer or moments of dramatic changes in customer behavior, such as the enterprise customer suddenly staying for too long while browsing the page or abnormally staying on a specific page, which may mean that the enterprise customer is hesitating or encountering difficulties or there is a fraud risk. The deviation candidate identifier is combined with the key behavior point identification result to construct an abnormal behavior trajectory graph, which shows the abnormal path and change trajectory of the enterprise customer behavior. The nodes in the graph represent the abnormal operations of the customer, and the edges represent the relationship between the operations, with special attention to the paths deviating from the normal behavior flow.

[0033] According to the information in the abnormal behavior trajectory graph, it is evaluated whether the behavior of the enterprise customer has potential risks, for example, the enterprise customer has frequent abnormal behaviors in recent time, such as skipping the payment link, frequently modifying the order, etc., which will increase the dynamic risk. According to the number, severity and frequency of the abnormal behavior trajectory, the enterprise dynamic risk of the enterprise customer is established, for example, if there are frequent abnormal behaviors or deviations in the behavior path of the enterprise customer, the risk level of the enterprise customer is increased.

[0034] Further, the establishment of the enterprise dynamic risk according to the abnormal behavior trajectory graph comprises:

[0035] The guide analysis sub-channel of the abnormal identification channel is activated, and the mapping group data is synchronized to the guide analysis sub-channel. The mapping behavior feedback in the mapping group data is extracted, and the behavior feature set is extracted according to the mapping behavior feedback. The behavior feature set includes click behavior features, stay time features, information correction behaviors, and abnormal process features. The risk trigger guide strategy is structurally deconstructed, the calibrated behavior path is established, and the target risk inducing point is identified. The deviation authentication of the behavior feature set is performed according to the calibrated behavior path and the target risk inducing point, and the deviation vector group is established. After clustering analysis of the deviation vector group, the strategy response behavior risk set is established. The enterprise dynamic risk is established according to the strategy response behavior risk set and the abnormal behavior trajectory graph.

[0036] The guide analysis sub-channel in the abnormal identification channel is activated, and all customer feedback in the mapping group data is synchronized to the guide analysis sub-channel together with the corresponding risk trigger guide strategy. The guide analysis sub-channel is used to deeply analyze the mapping behavior feedback in the mapping group data, and the mapping behavior feedback is taken as an input to further evaluate whether the customer behavior has potential risks.

[0037] The mapping behavior feedback is extracted from the mapping group data, and a set of key behavior features is extracted from the mapping behavior feedback, which are used to identify potential risks or deviations from normal behavior. The click behavior feature indicates the number of clicks or the frequency of click behavior of the enterprise customer on a certain page. If the enterprise customer repeatedly clicks on an element, it may mean that they are experiencing difficulties or have intentions on that content. The dwell time feature indicates the time the enterprise customer stays on certain pages or behavior nodes. For example, if the dwell time on a certain page is too long, it may mean that the enterprise customer is hesitating, which may indicate fraud or other risks. The information correction behavior indicates that if the enterprise customer frequently modifies or corrects the content when filling in the information, it may indicate that they are trying to fake information or evade certain risk control checks. The abnormal flow feature indicates that the enterprise customer's behavior is abnormal in the flow, such as skipping steps or not following the normal flow when filling in order information, which may be a signal of potential risk.

[0038] The risk trigger guide strategy is deconstructed to identify the key components of the strategy and analyze how these strategies affect customer behavior. By deconstructing the risk trigger guide strategy, a calibrated behavior path is created, which refers to an ideal and expected enterprise customer behavior pattern based on the requirements and design of the risk trigger guide strategy. This path describes the correct behavior path that the enterprise customer should take when facing risk prompts, such as in a high-risk transaction, the enterprise customer should first perform identity verification, then perform risk confirmation, and finally complete the transaction. In the calibrated behavior path, the target risk induction point is identified, which is a key node in the operation process that may trigger risks, such as the enterprise customer not following the expected path to perform identity verification and directly completing the transaction, which may trigger fraud risks. Focusing on these target risk induction points as the focus of subsequent risk management.

[0039] Based on the calibrated behavior path and the target risk induction point, it is analyzed whether the actual behavior feature set of the customer deviates significantly from the expected behavior pattern. If the customer behavior does not follow the ideal path, such as bypassing certain security steps or taking unreasonable behavior, it is considered as deviating behavior. Through deviation authentication, it can be identified whether the enterprise customer has abnormal behavior, which is an indication of potential risks such as fraud, evasion of risk control strategies or other malicious behavior. After deviation authentication, all deviating behaviors are converted into deviation vectors, which represent the difference between customer behavior and the calibrated path, and form a deviation vector group.

[0040] The deviation vector group is subjected to clustering analysis, and the clustering algorithm includes K-means, DBSCAN, hierarchical clustering, etc. The purpose is to combine behaviors with similar deviation characteristics together, so as to identify the pattern of deviation behaviors and identify which type of deviation behavior is more likely to trigger risks. Through clustering analysis, a strategy response behavior risk set is created according to different categories of deviation behaviors. This risk set contains the matching of various deviation behaviors and corresponding risk management strategies. For example, for some deviation behaviors, additional verification is required, or the behaviorally abnormal customers are limited.

[0041] The abnormal behavior trajectory graph displays the abnormal path of customer behavior. The strategy response behavior risk set is based on the risk categories after clustering analysis and the corresponding risk control strategies. The strategy response behavior risk set and the abnormal behavior trajectory graph are combined to finally generate the enterprise dynamic risk of the enterprise customer. For example, if the behavior of an enterprise customer frequently deviates from the expected path and matches a high-risk category, the enterprise dynamic risk rating of the enterprise customer will be high, and corresponding risk control measures will be taken.

[0042] Further, the establishment of the enterprise dynamic risk according to the strategy response behavior risk set and the abnormal behavior trajectory graph includes:

[0043] The strategy response behavior risk set and the abnormal behavior trajectory graph are synchronized to the fusion analysis sub-channel. The risk superposition authentication is performed by using the fusion analysis sub-channel to establish the enterprise dynamic risk.

[0044] The strategy response behavior risk set and the abnormal behavior trajectory graph are synchronized to the fusion analysis sub-channel. The fusion analysis sub-channel is a module for comprehensive analysis of multiple data sources. The purpose is to more comprehensively evaluate the risk of customers by combining risk data from different sources. For example, by combining the strategy response information in the strategy response behavior risk set and the deviation information in the abnormal behavior trajectory graph, a more accurate dynamic risk analysis result is generated.

[0045] Risk superposition authentication is performed using the fusion analysis sub-channel. Risk superposition authentication refers to the combination of risk information in the strategy response behavior risk set and the abnormal behavior trajectory graph to evaluate the overall dynamic risk of the customer. For example, the fusion analysis sub-channel integrates risk data from different sources by weighting. For example, for some high-risk behaviors, a higher weight is given according to the historical behavior trajectory graph. For the matching of behavior and risk control strategy, the risk rating is based on the information in the strategy response behavior risk set. Through risk superposition authentication, various risk signals are integrated to finally calculate the enterprise dynamic risk of the enterprise customer. This risk assessment is based on the comprehensive analysis of real-time behavior and past data of the customer, which can provide dynamic and real-time risk control data for enterprises.

[0046] Further, the generating the risk management strategy according to the static enterprise risk and the dynamic enterprise risk comprises:

[0047] configuring a calibrated risk level database; after joint risk analysis of the static enterprise risk and the dynamic enterprise risk, performing trigger matching of the calibrated risk level database to establish a matching result; identifying the risk level of the enterprise client according to the matching result, and establishing a pre-stored action database.

[0048] The calibrated risk level database is a database for storing standards and rules corresponding to different risk levels. The calibrated risk level database includes the risk level division standards of the enterprise client when facing static and dynamic risks, and how to assess the overall risk of the client based on these risks. The calibrated risk level database includes multiple fields, such as risk category, risk threshold, risk level (e.g. low, medium, high), and corresponding countermeasures. The database needs to develop corresponding risk assessment standards according to the specific needs, industry characteristics and historical data of the enterprise client.

[0049] The static enterprise risk is assessed according to the fixed characteristics of the enterprise client, and the dynamic enterprise risk is assessed based on the real-time behavior changes of the enterprise client. Through joint analysis, the overall risk of the enterprise client can be comprehensively evaluated. After joint risk analysis, trigger matching is performed in the calibrated risk level database. The process of trigger matching is to compare the overall risk characteristics of the client with the standards defined in the database to determine the risk level of the client.

[0050] According to the matching result, the specific risk level of the enterprise client is assigned. The risk level identification is the evaluation result of the comprehensive risk of the enterprise client, which is usually divided into multiple levels, such as low, medium, high, and extremely high. According to the risk level of the enterprise client, a pre-stored action database is created to store risk control measures corresponding to different risk levels, such as requiring additional identity verification for high-risk clients, limiting part of the function or performing real-time monitoring.

[0051] Further, the capturing enterprise client behavior based on the collected data set, and establishing a behavior path atlas based on the behavior capture result comprises:

[0052] The collected data set is sorted by time to establish a time-sequenced enterprise client behavior event stream. Based on the pre-set identification rules, the enterprise client behavior event stream is captured to extract behavior nodes, behavior time and behavior context. The behavior nodes, behavior time and behavior context are connected in a directed manner to establish a behavior path atlas.

[0053] The collected data set is sorted by time because customer behavior has time dependence, and subsequent analysis will be based on the time sequence of customer behavior. The sorted data set generates a linear event stream according to the timestamp, and each interaction behavior of the customer is arranged in chronological order. The enterprise customer behavior event stream refers to the interaction and behavior sequence between the enterprise customer and the enterprise system. In the time-ordered data, each behavior is associated with a timestamp to form an event stream. For example, the customer visited the website, viewed the product, added the product to the shopping cart, completed the payment, and so on. These behaviors are arranged in sequence to form a time-sequenced behavior event stream.

[0054] The preset identification rule refers to a set of rules for capturing key behaviors from the enterprise customer behavior event stream. These rules are based on the specific needs and goals of the enterprise and are used to identify specific high-risk behaviors, key user behaviors, etc. For example, the preset identification rule includes "customer modifies order multiple times within 30 minutes" and the like. According to the behavior capture based on the preset identification rule, the behavior node refers to each behavior event, such as the customer clicking on a certain page, adding a product to the shopping cart, etc. The behavior time is the timestamp corresponding to each behavior node, indicating the specific time when the behavior occurred. The behavior context includes additional information related to the behavior, such as the page where the user is located, the transaction amount, the operating device, etc.

[0055] After capturing each behavior node, these behavior nodes are connected in the order of their behavior time. The behavior node and the behavior time form a directed connection, i.e., the behavior occurs in sequence and has time dependence. In addition to the behavior node and the behavior time, the behavior context is also combined in the behavior path graph. In this way, not only can we view which behaviors the customer has performed, but also can we grasp the environmental information at the time of these behaviors, such as the page where the user is located, the operating device, etc. The obtained behavior path graph reveals the customer's behavior pattern within a certain time period, helping to analyze how the customer interacts with the system and the evolution of the behavior path.

[0056] Further, the static enterprise risk is used as an attention factor to configure a risk trigger guidance strategy based on internal interaction data, including:

[0057] The static enterprise risk is analyzed to obtain a static risk feature set. After normalization processing of the static risk feature set, a feature vector is established. For each risk feature in the feature vector, a shallow network is used for high-dimensional embedding to construct a feature embedding matrix. An attention score network layer is used to perform attention weighting analysis of the feature embedding matrix to calculate the attention score of each embedding vector to establish an attention factor.

[0058] The static enterprise risk is analyzed, and a series of static risk features are extracted, which represent the risk level of the enterprise, including: financial features such as balance sheet, revenue, profit rate and other financial indicators; industry risk, indicating market volatility, competition, policy changes and other factors in the industry where the enterprise is located; historical credit record, including credit score, past loan record, debt paying ability and other factors; compliance risk, including whether there has been a violation of rules, audit results and other factors. Static risk features are extracted from various data sources, including financial statements, industry reports, credit ratings, and other sources, and are aggregated into a static risk feature set as the basis for subsequent analysis.

[0059] The static risk feature set usually includes data of different dimensions and units, such as total assets, annual income, credit score, etc. The range of these data varies greatly. In order to balance the influence of different static risk features, normalization processing is performed on these static risk features, including min-max normalization method, Z-score standardization method, etc. The normalized static risk feature set is converted into a feature vector for subsequent processing and analysis.

[0060] The feature vector usually has a high dimension, and there is no explicit linear relationship between each feature. In order to improve the calculation efficiency and capture more complex feature relationships, high-dimensional embedding of features is performed through a shallow network. The shallow network is used to convert each static risk feature into a low-dimensional embedding representation. This embedding layer maps each feature value to a new space, reducing the dimensionality of the features and learning the relationship between the features as an embedding representation. By training the shallow network, a low-dimensional embedding vector is generated for each static risk feature, and a feature embedding matrix is finally formed. Each row of the matrix corresponds to an embedding representation of a static risk feature, which can more effectively express the mutual relationship between the features. Through this embedding method, high-dimensional and complex static risk features are converted into low-dimensional vectors with more compact and informative information, thereby better preparing for subsequent analysis.

[0061] The attention mechanism is applied to the feature embedding matrix for attention weighting analysis. The attention mechanism enables the model to focus on the most important features for risk assessment and gives lower weights to less important features. By calculating the attention score, the influence of different features on the final risk assessment can be quantified. Through the attention score network layer, the features in the feature embedding matrix are weighted, and the weighted feature representation is finally calculated to obtain the attention score of each embedding vector. This attention score reflects the importance of the feature in calculating the final risk. Features with high scores are given greater weights, and vice versa. Finally, attention factors are established according to the attention scores, which determine the influence of each feature in the risk assessment process.

[0062] Further, the static enterprise risk is taken as an attention factor, and the risk trigger guide strategy is configured based on internal interaction data, which further includes:

[0063] A general guide trigger database is established, the internal interaction data is matched with the guide trigger database to establish an adaptive matching result, and the risk trigger guide strategy is established after the adaptive matching result is enhanced by the attention factor.

[0064] The general guide trigger database is a database that stores preset and general risk trigger strategies, which are a set of rules for guiding the behavior of enterprise customers. These rules can be used to identify potential risks in customer behavior and take appropriate risk control measures if necessary, such as triggering a warning, limiting transactions, or conducting further review if the customer's behavior matches the risk rules. The database includes information such as trigger conditions, risk types, corresponding trigger actions, and effective time.

[0065] The internal interaction data includes the interaction records between the enterprise and the customer. The internal interaction data is matched with the rules in the guide trigger database, and the customer's internal interaction data is compared with the risk trigger rules in the database through matching algorithms such as pattern matching and rule engine matching to identify which behaviors meet the risk trigger rules defined in the database. Through adaptive matching, the customer's behavior is classified, and an adaptive matching result is generated, for example, if a customer makes multiple large transactions in a short period of time and these behaviors match the frequent transaction rules in the guide trigger database, the customer's behavior is marked as high risk, and the corresponding adaptive matching result is generated.

[0066] The adaptive matching result is enhanced by the attention factor, which represents the importance of each feature or behavior to the final decision. By using the attention factor, the matching result is analyzed by weighting different features, and it is identified which factors have a greater impact on the customer's risk assessment, so as to optimize the risk trigger strategy. For example, if some behaviors such as frequent account changes are given a higher attention score, the importance of the behavior in the risk strategy will increase, and the trigger strategy will be more stringent. Through weighted analysis and strategy enhancement, an updated risk trigger guide strategy is finally generated to help the risk control team develop appropriate response strategies for different customers and different risk levels.

[0067] Further, the data collection of the enterprise customer is performed to establish a collection data set, which includes:

[0068] After data cleaning of the data collection result, the data trust degree is configured, and the data trust degree is identified after cleaning the data collection result as the collection data set output.

[0069] Data cleaning is performed on the data collection results, the purpose being to remove redundant, invalid or erroneous data, to ensure that subsequent analysis is based on accurate and high-quality data, and the data cleaning process includes removing noisy data, format standardization, data completion, etc. Data trustworthiness is configured, which refers to the reliability and accuracy of the data, and is evaluated based on factors such as data source, data integrity and historical consistency, and data with high trustworthiness is given a higher weight, and conversely, data with low trustworthiness is considered unreliable and affects the decision-making process.

[0070] After data cleaning and trustworthiness evaluation, the data trustworthiness is identified on each data record, which means that each collected data will have a trustworthiness score indicating the reliability and validity of the data. The data set after the trustworthiness is identified is used as the final data output for subsequent analysis and risk control strategy.

[0071] In summary, the enterprise customer risk control management method based on AI intelligence provided by the embodiments has the following technical effects:

[0072] By collecting internal interaction data and external public data, various information of enterprise customers can be comprehensively integrated, which provides multi-dimensional data support for subsequent risk control management and behavior analysis; by capturing the behavior of the collected data set, the behavior pattern of the enterprise customer can be mastered in detail, and the establishment of the behavior path atlas can show the interaction behavior sequence of the customer, reflecting the regular and abnormal behavior of the enterprise customer when interacting with the enterprise system, helping the risk control team to intuitively understand the pattern and changes of customer behavior, thereby effectively identifying potential risk behavior; by evaluating the static enterprise risk of the enterprise customer, the potential risk of the enterprise customer in the long term is identified, and after the static enterprise risk is taken as the attention factor, attention is focused on the features that have the greatest impact on risk control decisions, thereby configuring precise risk trigger guidance strategies to help the risk control system respond appropriately to different risk scenarios; by mapping the interaction between the risk trigger guidance strategy and the enterprise customer to establish a mapping behavior feedback, the risk control strategy can be adjusted according to the actual behavior of the customer, and this feedback mechanism helps the system to update the risk assessment in real time to respond to changes in customer behavior, and the establishment of the mapping group data enables the risk control system to dynamically adjust the risk management strategy according to different customer behavior feedback, thereby improving the flexibility and adaptability of risk control management; by identifying abnormal channels in the mapping group data and the behavior path atlas, abnormal paths in the customer behavior can be identified, and after identifying the abnormal channels, the dynamic enterprise risk of the customer is evaluated in real time to provide timely information for risk control decisions; by combining the static enterprise risk and the dynamic enterprise risk, the overall risk of the enterprise customer is comprehensively evaluated, thereby generating more accurate risk control management strategies, making the risk control management more comprehensive and accurate.

[0073] Embodiment two, based on the same inventive concept as the AI intelligent-based enterprise customer risk control management method in the preceding embodiment, as Figure 2 As shown in the embodiment of the present application, an AI intelligent-based enterprise customer risk control management system is provided, which comprises:

[0074] A data collection module 10 is configured to perform data collection of an enterprise customer, and establish a collection data set, wherein the collection data set comprises internal interaction data and external public data; a behavior capture module 20 is configured to capture enterprise customer behavior based on the collection data set, and establish a behavior path atlas based on the behavior capture result; a strategy triggering module 30 is configured to obtain static enterprise risks of an enterprise customer, and configure a risk triggering guide strategy based on the internal interaction data, taking the static enterprise risks as attention factors; a mapping group data establishment module 40 is configured to, after the enterprise customer interacts based on the risk triggering guide strategy, establish mapping behavior feedback, and establish the mapping behavior feedback and the corresponding risk triggering guide strategy as mapping group data; an abnormal channel identification module 50 is configured to identify abnormal channels based on the mapping group data and the behavior path atlas, and establish dynamic enterprise risks; and a management strategy generation module 60 is configured to generate a risk control management strategy based on the static enterprise risks and the dynamic enterprise risks.

[0075] Further, the abnormal channel identification module 50 is configured to perform the following operation steps:

[0076] The dynamic behavior sub-channel of the activated abnormal identification channel is inputted into the dynamic behavior sub-channel, and the behavior path atlas is deconstructed into a triple sequence of business activity nodes, timestamps and operator information by using an analysis layer, and an enterprise customer behavior mode set is constructed according to the triple sequence; after a target behavior flow template is configured, behavior deviation analysis is performed on the enterprise customer behavior mode set according to the target behavior flow template, and a deviation candidate identifier is established; key behavior point identification is performed based on the deviation candidate identifier, and an abnormal behavior trajectory graph is established based on the key behavior point identification result and the deviation candidate identifier; and an enterprise dynamic risk is established according to the abnormal behavior trajectory graph.

[0077] Further, the abnormal channel identification module 50 is configured to perform the following operation steps:

[0078] The guide analysis sub-channel of the activated abnormality recognition channel is guided, and the mapping group data is synchronized to the guide analysis sub-channel; a mapping behavior feedback in the mapping group data is extracted, a behavior feature set is extracted according to the mapping behavior feedback, the behavior feature set includes a click behavior feature, a stay time feature, an information correction behavior and an abnormal flow feature; a risk trigger guide strategy is structurally deconstructed, a calibrated behavior path is established, and a target risk inducing point is identified; a deviation authentication of the behavior feature set is performed according to the calibrated behavior path and the target risk inducing point, a deviation vector group is established; after clustering analysis of the deviation vector group, a strategy response behavior risk set is established; and an enterprise dynamic risk is established according to the strategy response behavior risk set and the abnormal behavior trajectory graph.

[0079] Further, the abnormal channel recognition module 50 is configured to perform the following operation steps:

[0080] The strategy response behavior risk set and the abnormal behavior trajectory graph are synchronized to the fusion analysis sub-channel; risk superposition authentication is performed by using the fusion analysis sub-channel, and an enterprise dynamic risk is established.

[0081] Further, the management strategy generation module 60 is configured to perform the following operation steps:

[0082] A calibrated risk level database is configured; after joint risk analysis of the static enterprise risk and the dynamic enterprise risk, trigger matching of the calibrated risk level database is performed, and a matching result is established; risk level identification of an enterprise customer is performed according to the matching result, and a pre-stored action database is established.

[0083] Further, the behavior capturing module 20 is configured to perform the following operation steps:

[0084] The collected data set is sorted according to time, and a time-sequenced enterprise customer behavior event stream is established; behavior capturing is performed on the enterprise customer behavior event stream based on a preset identification rule, and a behavior node, a behavior time and a behavior context are extracted; the behavior node, the behavior time and the behavior context are directionally connected, and a behavior path graph is established.

[0085] Further, the strategy trigger module 30 is configured to perform the following operation steps:

[0086] The static enterprise risk is parsed, and a static risk feature set is obtained; after normalization processing of the static risk feature set, a feature vector is established; each risk feature in the feature vector is embedded in a high dimension by using a shallow network, and a feature embedding matrix is constructed; attention weighting analysis of the feature embedding matrix is performed by using an attention score network layer, attention scores of each embedding vector are calculated, and an attention factor is established.

[0087] Further, the strategy triggering module 30 is configured to perform the following steps:

[0088] A general guidance triggering database is established; the guidance triggering database is adaptively matched according to the internal interaction data, and an adaptive matching result is established; after the adaptive matching result is enhanced by the attention factor, a risk triggering guidance strategy is established.

[0089] Further, the data collection module 10 is configured to perform the following steps:

[0090] After the data collection result is cleaned, the data trust degree is configured; after the data trust degree is marked on the cleaned data collection result, the data collection result is output as a collection data set.

[0091] Through the foregoing detailed description of the enterprise customer risk control management method based on AI intelligence, those skilled in the art can clearly understand the enterprise customer risk control management system based on AI intelligence in the embodiments. Since the system corresponds to the method disclosed in the embodiments, the system is described relatively simply, and the relevant part can be referred to the method part.

[0092] The above description of the disclosed embodiments enables a person skilled in the art to implement or use the present application. Various modifications to the embodiments will be apparent to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. An AI-based enterprise customer risk control management method, characterized in that, The method includes: Perform data collection from enterprise clients and establish a collection dataset, which includes internal interactive data and external publicly available data; Enterprise customer behavior is captured from the collected dataset, and a behavior path map is established based on the behavior capture results; Obtain the static enterprise risk of enterprise customers, use the static enterprise risk as an attention factor, and configure risk triggering guidance strategies based on internal interaction data; After interacting with enterprise customers based on the risk triggering guidance strategy, a mapping behavior feedback is established, and the mapping behavior feedback and the corresponding risk triggering guidance strategy are established as a mapping group data. Anomaly channel identification is performed on the mapping group data and the behavior path map to establish dynamic enterprise risk; Risk control management strategies are generated based on the static enterprise risks and the dynamic enterprise risks. Using the aforementioned static enterprise risk as an attention factor, a risk-triggered guidance strategy is configured based on internal interaction data, including: Analyze static enterprise risks to obtain a set of static risk characteristics; After normalizing the static risk feature set, a feature vector is established; For each risk feature in the feature vector, a shallow network is used to perform high-dimensional embedding to construct a feature embedding matrix; Attention-weighted analysis of the feature embedding matrix is ​​performed using an attention-scoring network layer to calculate the attention score of each embedding vector in order to establish an attention factor. Establish a universal boot trigger database; Based on the internal interaction data, guide and trigger the database adaptation and matching to establish the adaptation and matching results; After enhancing the strategy by adapting the matching results through the attention factor, a risk-triggered guidance strategy is established.

2. The AI-based enterprise customer risk control management method as described in claim 1, characterized in that, The process of identifying abnormal channels in the mapping group data and the behavior path map to establish dynamic enterprise risk includes: Activate the dynamic behavior sub-channel of the anomaly identification channel, input the behavior path graph into the dynamic behavior sub-channel, and use the parsing layer to deconstruct the behavior path graph into a triplet sequence of business activity nodes, timestamps, and operator information, and construct an enterprise customer behavior pattern set based on the triplet sequence. After configuring the target behavior flow template, perform behavior deviation analysis on the enterprise customer behavior pattern set based on the target behavior flow template, and establish deviation candidate identifiers; Based on the deviation candidate identifiers, key behavior points are identified, and an abnormal behavior trajectory map is established using the key behavior point identification results and deviation candidate identifiers. Establish dynamic enterprise risk based on the aforementioned abnormal behavior trajectory map.

3. The AI-based enterprise customer risk control management method as described in claim 2, characterized in that, The establishment of dynamic enterprise risk based on the abnormal behavior trajectory map includes: Activate the guidance analysis sub-channel of the anomaly identification channel and synchronize the mapping group data to the guidance analysis sub-channel; Extract mapping behavior feedback from the mapping group data, and extract a behavior feature set based on the mapping behavior feedback. The behavior feature set includes click behavior features, dwell time features, information correction behavior, and abnormal process features. The risk triggering guidance strategy is structurally deconstructed to establish a calibrated behavioral path and identify the target risk triggering point; Based on the calibrated behavioral path and target risk triggering point, deviation authentication of the behavioral feature set is performed, and a deviation vector group is established; After performing cluster analysis on the deviation vector group, a risk set of strategy response behavior is established; Enterprise dynamic risk is established based on the risk set of strategy response behavior and the trajectory map of abnormal behavior.

4. The AI-based enterprise customer risk control management method as described in claim 3, characterized in that, The establishment of enterprise dynamic risk based on the strategy response behavior risk set and the abnormal behavior trajectory map includes: Synchronize the risk set of the strategy response behavior and the trajectory map of the abnormal behavior to the fusion analysis sub-channel; The aforementioned fusion analysis sub-channel is used for risk overlay authentication to establish dynamic enterprise risk.

5. The AI-based enterprise customer risk control management method as described in claim 1, characterized in that, The generation of risk control management strategies based on the static enterprise risk and the dynamic enterprise risk includes: Configure a risk level database; After performing joint risk analysis on the static enterprise risk and the dynamic enterprise risk, trigger matching is performed on the risk level database to establish matching results; Based on the matching results, the risk level of enterprise customers is identified, and a pre-stored action database is established.

6. The AI-based enterprise customer risk control management method as described in claim 1, characterized in that, The step of capturing enterprise customer behavior from the collected dataset and establishing a behavior path map based on the behavior capture results includes: The collected dataset is sorted by time to establish a time-series enterprise customer behavior event stream. Based on preset recognition rules, the system captures enterprise customer behavior event streams and extracts behavior nodes, behavior time, and behavior context. Directed connections are made between the behavior nodes, behavior times, and behavior contexts to establish a behavior path graph.

7. The AI-based enterprise customer risk control management method as described in claim 1, characterized in that, The process of collecting data from enterprise customers and establishing a collection dataset includes: After cleaning the collected data, configure the data trust level; The data trust level is used to label the cleaned data collection results and then output as the collected dataset.

8. An AI-based enterprise customer risk control management system, characterized in that: The system is used to implement the AI-based enterprise customer risk control management method according to any one of claims 1-7, the system comprising: The data acquisition module is used to perform data acquisition from enterprise customers and establish a collection dataset, which includes internal interactive data and external public data. The behavior capture module is used to capture enterprise customer behavior from the collected dataset and build a behavior path map based on the behavior capture results. The strategy triggering module is used to acquire the static enterprise risk of enterprise customers, use the static enterprise risk as an attention factor, and configure a risk triggering guidance strategy based on internal interaction data. The mapping group data establishment module is used to establish mapping behavior feedback after enterprise customer interaction based on the risk triggering guidance strategy, and establish mapping behavior feedback and corresponding risk triggering guidance strategy as mapping group data. An abnormal channel identification module is used to identify abnormal channels in the mapping group data and the behavior path map to establish dynamic enterprise risk. The management strategy generation module is used to generate risk control management strategies based on the static enterprise risk and the dynamic enterprise risk.

Images