Automatic optimization method and system for anti-fraud rule

Abstract

The invention relates to an automatic optimization method and system for an anti-fraud rule, and the method comprises the steps: carrying out the feature extraction of a real-time data stream of a case-related number, obtaining a case-related feature, carrying out the rule semantic analysis of all rules through a large language model, obtaining the rule intention of each rule, and carrying out the optimization of the anti-fraud rule. And screening out a rule matched with the case-related characteristics from the rule intentions of all rules as a to-be-optimized rule, converting an optimization problem of the to-be-optimized rule into an optimization decision problem of the intelligent agent, and calculating and executing an optimal decision of the optimization decision problem to realize automatic optimization of the to-be-optimized rule. Therefore, the anti-fraud rule can be automatically optimized without depending on manpower so as to adapt to rapidly changing fraud skills, and the overall response efficiency is improved.

Description

Technical Field

[0001] This invention relates to the field of computer technology, and in particular to an automatic optimization method and system for anti-fraud rules. Background Technology

[0002] With the widespread adoption of telecommunications services and the iterative development of communication technologies, telecommunications fraud has become increasingly sophisticated, characterized by diversified methods, rapid changes in patterns, and enhanced concealment. Current mainstream anti-fraud methods primarily rely on three models, but all of these models have inherent limitations: (a) Feature-engineered machine learning models: such as LightGBM and XGBoost, which train models using historical data to predict fraudulent numbers. However, these models have long update cycles and are difficult to adapt to rapidly changing fraud methods.

[0003] (ii) Rule engine combined with human intervention: Rules are automatically generated based on the experience of anti-fraud experts and adjusted by humans to predict fraudulent numbers. However, the construction of rules relies heavily on expert experience and requires manual review and adjustment, which is inefficient and highly subjective.

[0004] (III) Large-scale model-assisted analysis: Some solutions have begun to introduce large-scale models for feature extraction or sample generation, but the parameter adjustment of large-scale models still relies on human experience, which is highly subjective and inefficient.

[0005] Therefore, current mainstream anti-fraud methods share a common problem: over-reliance on manual intervention, which leads to shortcomings such as lagging rule updates, poor adaptability, and low overall response efficiency when facing rapidly evolving fraud threats. Summary of the Invention

[0006] The technical problem to be solved by the present invention is: the present invention provides an automatic optimization method and system for anti-fraud rules, which can automatically optimize anti-fraud rules without relying on manual intervention, so as to adapt to rapidly changing fraud methods and improve overall response efficiency.

[0007] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows: In a first aspect, the present invention provides an automatic optimization method for anti-fraud rules, comprising: The system acquires the real-time data stream of the numbers involved in the case and all rules in the rule base. It then extracts features from the real-time data stream of the numbers involved in the case to obtain the features involved in the case. At the same time, it performs semantic parsing of all rules through a large language model to obtain the rule intent of each rule. From the rule intent of all rules, rules that match the characteristics of the case are selected as rules to be optimized; The optimization problem of the rule to be optimized is transformed into an optimization decision problem of the agent. The optimal decision of the optimization decision problem is calculated and the optimal decision is executed to achieve automatic optimization of the rule to be optimized.

[0008] The beneficial effects of this invention are as follows: Unlike the traditional approach of simply treating rules as "if-then" logic, this invention performs semantic parsing of the rules, transforming each rule into an understandable rule intent. It matches the case-related characteristics of the real-time data stream of the involved numbers with the rule intent, thereby enabling accurate and rapid screening of rules to be optimized. Furthermore, it transforms the optimization problem of the rules to be optimized into an optimization decision problem for an intelligent agent. The intelligent agent can autonomously obtain the optimal decision through calculation, without relying on manual intervention, shortening the rule iteration cycle, adapting to rapidly changing fraud methods, and improving overall response efficiency.

[0009] Optionally, the features involved in the case include features of text messages involved in the case, features of calls involved in the case, and features of data traffic involved in the case. The feature extraction of the real-time involved number to obtain the features involved in the case includes: The real-time data stream of the number involved in the case is input into a pre-built feature extraction model to extract SMS features, call features, and traffic features respectively, to obtain the corresponding SMS features, call features, and traffic features involved in the case. The real-time data stream of the number involved in the case includes SMS data, call data, and traffic data.

[0010] As can be seen from the above description, diversified feature extraction ensures the completeness and comprehensiveness of the obtained case-related features.

[0011] Optionally, the step of converting the optimization problem of the rule to be optimized into an optimization decision problem for the agent, and calculating the optimal decision for the optimization decision problem, includes: Obtain the current threshold and historical performance index of the rule to be optimized, and construct the state space of the optimization decision problem based on the current threshold, the historical performance index and the case-related features; The action space of the optimization decision problem is constructed based on the adjustment direction and adjustment magnitude of the current threshold; The optimal decision for the optimization problem is calculated based on the state space and the action space.

[0012] As described above, the state space includes not only the current threshold but also historical performance indicators and case-related characteristics. The multi-dimensional state space forms an objective basis for decision-making, overcoming the subjectivity of decision-making. The action space includes not only the adjustment direction of the current threshold but also the adjustment magnitude, transforming fuzzy adjustment behaviors into controllable discrete actions, avoiding drastic rule fluctuations caused by excessive adjustment of the optimal decision, and ensuring the stability of rule adjustment.

[0013] Optionally, calculating the optimal decision for the optimization problem based on the state space and the action space includes: The intelligent agent simulates the execution result of the optimal decision based on the real-time data stream of the number involved in the case, and calculates the reward value of the execution result through a multi-objective reward function; Determine whether the reward value exceeds a first threshold. If not, recalculate the optimal decision for the optimization decision problem based on the execution result, the state space, and the action space.

[0014] As described above, simulating the execution result of the optimal decision in advance avoids interference from trial-and-error optimization on actual anti-fraud operations. The reward value of the execution result is calculated through a multi-objective reward function, and the reward value triggers iterative optimization of the optimal decision, thereby improving the accuracy of the obtained optimal decision without the need for manual adjustment and improving response efficiency.

[0015] Optionally, the intelligent agent simulates the execution result of the optimal decision based on the real-time data stream of the involved number, and calculates the reward value of the execution result through a multi-objective reward function, including: The recall rate, false positive rate, and stability rate are calculated based on the execution results. The reward value of the execution results is then calculated using the recall rate, false positive rate, stability rate, and a multi-objective reward function, whereby:

[0016] Where R represents the reward value, This represents the first weighting coefficient. Indicates recall rate, This represents the second weighting coefficient. Indicates the false alarm rate. This represents the third weighting coefficient. This indicates the stability rate.

[0017] As described above, by using a multi-objective reward function to calculate the reward value, recall, false positive rate and stability rate are taken into account, avoiding the imbalance problem of single-index optimization and improving the accuracy of optimal decision-making.

[0018] Secondly, the present invention provides an automatic optimization system for anti-fraud rules, comprising: The acquisition module is used to acquire the real-time data stream of the numbers involved in the case and all the rules in the rule base, extract features from the real-time data stream of the numbers involved in the case to obtain the features involved in the case, and at the same time perform rule semantic parsing on all rules through a large language model to obtain the rule intent of each rule; The matching module is used to filter out rules that match the features involved in the case from the rule intent of all rules as rules to be optimized; An optimization module is used to transform the optimization problem of the rule to be optimized into an optimization decision problem of the agent, calculate the optimal decision of the optimization decision problem, and execute the optimal decision to achieve automatic optimization of the rule to be optimized.

[0019] The beneficial effects of this invention are as follows: Unlike the traditional approach of simply treating rules as "if-then" logic, this invention performs semantic parsing of the rules, transforming each rule into an understandable rule intent. It matches the case-related characteristics of the real-time data stream of the involved numbers with the rule intent, thereby enabling accurate and rapid screening of rules to be optimized. Furthermore, it transforms the optimization problem of the rules to be optimized into an optimization decision problem for an intelligent agent. The intelligent agent can autonomously obtain the optimal decision through calculation, without relying on manual intervention, shortening the rule iteration cycle, adapting to rapidly changing fraud methods, and improving overall response efficiency.

[0020] Optionally, the features involved in the case include features of text messages involved in the case, features of calls involved in the case, and features of data traffic involved in the case, and the acquisition module specifically includes: The real-time data stream of the number involved in the case is input into a pre-built feature extraction model to extract SMS features, call features, and traffic features respectively, to obtain the corresponding SMS features, call features, and traffic features involved in the case. The real-time data stream of the number involved in the case includes SMS data, call data, and traffic data.

[0021] As can be seen from the above description, diversified feature extraction ensures the completeness and comprehensiveness of the obtained case-related features.

[0022] Optionally, the optimization module specifically comprises: Obtain the current threshold and historical performance index of the rule to be optimized, and construct the state space of the optimization decision problem based on the current threshold, the historical performance index and the case-related features; The action space of the optimization decision problem is constructed based on the adjustment direction and adjustment magnitude of the current threshold; The optimal decision for the optimization problem is calculated based on the state space and the action space.

[0023] As described above, the state space includes not only the current threshold but also historical performance indicators and case-related characteristics. The multi-dimensional state space forms an objective basis for decision-making, overcoming the subjectivity of decision-making. The action space includes not only the adjustment direction of the current threshold but also the adjustment magnitude, transforming fuzzy adjustment behaviors into controllable discrete actions, avoiding drastic rule fluctuations caused by excessive adjustment of the optimal decision, and ensuring the stability of rule adjustment.

[0024] Optionally, the optimization module includes: The reward iteration module is used by the agent to simulate the execution result of the optimal decision based on the real-time data stream of the number involved in the case, and to calculate the reward value of the execution result through a multi-objective reward function; Determine whether the reward value exceeds a first threshold. If not, recalculate the optimal decision for the optimization decision problem based on the execution result, the state space, and the action space.

[0025] As described above, simulating the execution result of the optimal decision in advance avoids interference from trial-and-error optimization on actual anti-fraud operations. The reward value of the execution result is calculated through a multi-objective reward function, and the reward value triggers iterative optimization of the optimal decision, thereby improving the accuracy of the obtained optimal decision without the need for manual adjustment and improving response efficiency.

[0026] Optionally, the reward iteration module specifically comprises: The recall rate, false positive rate, and stability rate are calculated based on the execution results. The reward value of the execution results is then calculated using the recall rate, false positive rate, stability rate, and a multi-objective reward function, whereby:

[0027] Where R represents the reward value, This represents the first weighting coefficient. Indicates recall rate, This represents the second weighting coefficient. Indicates the false alarm rate. This represents the third weighting coefficient. This indicates the stability rate.

[0028] As described above, by using a multi-objective reward function to calculate the reward value, recall, false positive rate and stability rate are taken into account, avoiding the imbalance problem of single-index optimization and improving the accuracy of optimal decision-making. Attached Figure Description

[0029] Figure 1 A flowchart illustrating an automatic optimization method for anti-fraud rules provided in this embodiment; Figure 2 This is a schematic diagram of the overall process of an automatic optimization method for anti-fraud rules provided in this embodiment; Figure 3 This is a schematic diagram of the structure of an automatic optimization system for anti-fraud rules provided in this embodiment.

[0030] [Explanation of Labels in the Attached Image] 1. An automatic optimization system for anti-fraud rules; 2. Acquisition module; 3. Matching module; 4. Optimization module; 41. Reward iteration module. Detailed Implementation

[0031] To better understand the above technical solutions, exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present invention are shown in the drawings, it should be understood that the present invention can be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that the present invention can be understood more clearly and thoroughly, and that the scope of the present invention can be fully conveyed to those skilled in the art.

[0032] Example 1 Please refer to Figures 1 to 2 This invention provides an automatic optimization method for anti-fraud rules, comprising the following steps: S1. Obtain the real-time data stream of the number involved in the case and all rules in the rule base, extract features from the real-time data stream of the number involved in the case to obtain the features involved in the case, and at the same time perform rule semantic parsing on all rules through a large language model to obtain the rule intent of each rule; In this embodiment, as Figure 2 As shown, the system acquires the real-time data stream of the numbers involved in the case and all rules in the rule base. The real-time data stream of the numbers involved in the case includes, but is not limited to: blacklisted numbers, case time, fraud type, SMS data, call data, and traffic data. Feature extraction is performed on the real-time data stream of the numbers involved in the case to obtain the case features. At the same time, a large language model is used to perform rule semantic parsing on all rules to obtain the rule intent of each rule. For example: Rule 1: The average monthly traffic in the past three months is greater than 100GB. After the large language model performs rule semantic parsing on Rule 1, the rule intent obtained is: to capture abnormally high traffic behavior. Rule 2: The duration of a single call is less than 30s. After the large language model performs rule semantic parsing on Rule 2, the rule intent obtained is: to capture abnormally short call behavior.

[0033] At this point, the case-related features mentioned in step S1 include case-related SMS features, case-related call features, and case-related data traffic features. The step of extracting features from the real-time case-related number to obtain the case-related features includes: S11. Input the real-time data stream of the number involved in the case into the pre-constructed feature extraction model to extract SMS features, call features and traffic features respectively, to obtain the corresponding SMS features, call features and traffic features involved in the case. The real-time data stream of the number involved in the case includes SMS data, call data and traffic data.

[0034] In this embodiment, as Figure 2As shown, the real-time data stream of the numbers involved in the case is input into a pre-built feature extraction model. SMS features are extracted from the SMS data, call features from the call data, and traffic features from the data traffic, thus obtaining the relevant SMS features, call features, and traffic features. The obtained features are then filtered, removing those with variance below a first variance threshold. The obtained features are output in a unified format, such as JSON.

[0035] S2. Select rules that match the characteristics of the case from all the rules' intents as the rules to be optimized; In this embodiment, as Figure 2 As shown, rules that match the characteristics of the case are selected from the rule intent of all rules as rules to be optimized. That is, the goal is to locate the rules to be optimized, rather than directly generating new rules. Instead, the rules that best match the characteristics of the case are located and optimized, thereby avoiding the problem of rule base expansion.

[0036] S3. The optimization problem of the rule to be optimized is transformed into an optimization decision problem of the agent, the optimal decision of the optimization decision problem is calculated, and the optimal decision is executed to achieve automatic optimization of the rule to be optimized.

[0037] At this point, step S3, which involves transforming the optimization problem of the rule to be optimized into an optimization decision problem for the agent, and calculating the optimal decision for the optimization decision problem, includes: S31. Obtain the current threshold and historical performance index of the rule to be optimized, and construct the state space of the optimization decision problem based on the current threshold, the historical performance index and the case-related features; S32. Construct the action space of the optimization decision problem based on the adjustment direction and adjustment magnitude of the current threshold; S33. Calculate the optimal decision for the optimization decision problem based on the state space and the action space.

[0038] In this embodiment, as Figure 2 As shown, the optimization problem of the rule to be optimized is transformed into an optimization decision problem of the agent. The state space and action space of the agent's optimization decision problem are constructed respectively. The state space is constructed based on the current threshold, historical performance indicators and case features of the rule to be optimized. The action space is constructed based on the adjustment direction and adjustment magnitude of the current threshold. The optimal decision of the optimization decision problem is calculated based on the state space and action space.

[0039] At this point, step S33 includes: S331. The intelligent agent simulates the execution result of the optimal decision based on the real-time data stream of the number involved in the case, and calculates the reward value of the execution result through a multi-objective reward function; At this point, step S331 specifically involves: The recall rate, false positive rate, and stability rate are calculated based on the execution results. The reward value of the execution results is then calculated using the recall rate, false positive rate, stability rate, and a multi-objective reward function, whereby:

[0040] Where R represents the reward value, This represents the first weighting coefficient. Indicates recall rate, This represents the second weighting coefficient. Indicates the false alarm rate. This represents the third weighting coefficient. This indicates the stability rate.

[0041] S332. Determine whether the reward value exceeds the first threshold. If not, recalculate the optimal decision for the optimization decision problem based on the execution result, the state space, and the action space.

[0042] In this embodiment, as Figure 2 As shown, the intelligent agent simulates the execution result of the optimal decision based on the real-time data stream of the involved number. That is, it simulates and deduces the optimal decision to obtain the execution result, and calculates the recall rate, false positive rate, and stability rate based on the execution result. The recall rate is obtained by judging whether the latest threshold obtained by the optimal decision covers the involved number and similar numbers. The recall rate, false positive rate, and stability rate are input into the multi-objective reward function to calculate the reward value. If the reward value exceeds the first threshold, it means that the optimal decision can be adopted and executed. Otherwise, if the reward value does not exceed the first threshold, it means that other optimal decisions can be further explored. The optimal decision is iteratively optimized, that is, the optimal decision of the optimization decision problem is recalculated based on the execution result, state space, and action space until the new reward value of the new optimal decision exceeds the first threshold, thereby realizing the automatic optimization of the rule to be optimized.

[0043] Example 2 Please refer to Figure 3 The present invention provides an automatic optimization system 1 for anti-fraud rules, comprising: an acquisition module 2, a matching module 3, an optimization module 4, and a reward iteration module 41.

[0044] Among them, the acquisition module 2 is used to acquire the real-time data stream of the number involved in the case and all the rules in the rule base, extract features from the real-time data stream of the number involved in the case to obtain the features involved in the case, and at the same time perform rule semantic parsing on all rules through a large language model to obtain the rule intent of each rule; Matching module 3 is used to filter out rules that match the features involved in the case from the rule intent of all rules as rules to be optimized; The optimization module 4 is used to convert the optimization problem of the rule to be optimized into an optimization decision problem of the agent, calculate the optimal decision of the optimization decision problem, and execute the optimal decision to achieve automatic optimization of the rule to be optimized.

[0045] Specifically, the features involved in the case include features of text messages involved in the case, features of calls involved in the case, and features of data traffic involved in the case, and the acquisition module 2 specifically includes: The real-time data stream of the number involved in the case is input into a pre-built feature extraction model to extract SMS features, call features, and traffic features respectively, to obtain the corresponding SMS features, call features, and traffic features involved in the case. The real-time data stream of the number involved in the case includes SMS data, call data, and traffic data.

[0046] Specifically, the optimization module 4 is as follows: Obtain the current threshold and historical performance index of the rule to be optimized, and construct the state space of the optimization decision problem based on the current threshold, the historical performance index and the case-related features; The action space of the optimization decision problem is constructed based on the adjustment direction and adjustment magnitude of the current threshold; The optimal decision for the optimization problem is calculated based on the state space and the action space.

[0047] Specifically, the optimization module 4 includes: The reward iteration module 41 is used by the agent to simulate the execution result of the optimal decision based on the real-time data stream of the number involved in the case, and to calculate the reward value of the execution result through a multi-objective reward function; Determine whether the reward value exceeds a first threshold. If not, recalculate the optimal decision for the optimization decision problem based on the execution result, the state space, and the action space.

[0048] Specifically, the reward iteration module 41 is as follows: The recall rate, false positive rate, and stability rate are calculated based on the execution results. The reward value of the execution results is then calculated using the recall rate, false positive rate, stability rate, and a multi-objective reward function, whereby:

[0049] Where R represents the reward value, This represents the first weighting coefficient. Indicates recall rate, This represents the second weighting coefficient. Indicates the false alarm rate. This represents the third weighting coefficient. This indicates the stability rate.

[0050] Since the systems / devices described in the above embodiments of the present invention are systems / devices used to implement the methods of the above embodiments of the present invention, those skilled in the art can understand the specific structure and modifications of the systems / devices based on the methods described in the above embodiments of the present invention, and therefore will not be repeated here. All systems / devices used in the methods of the above embodiments of the present invention fall within the scope of protection of the present invention.

[0051] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0052] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, as well as combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions.

[0053] It should be noted that any reference numerals placed between parentheses in the claims should not be construed as limiting the claims. The word "comprising" does not exclude the presence of components or steps not listed in the claims. The word "a" or "an" preceding a component does not exclude the presence of a plurality of such components. The invention can be implemented by means of hardware comprising several different components and by means of a suitably programmed computer. In claims that enumerate several means, several of these means may be embodied by the same hardware. The use of the terms first, second, third, etc., is merely for convenience of expression and does not indicate any order. These terms can be understood as part of the component names.

[0054] Furthermore, it should be noted that in the description of this specification, the terms "one embodiment," "some embodiments," "embodiment," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Furthermore, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0055] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the claims should be interpreted to include both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0056] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, then this invention should also include these modifications and variations.

Claims

1. An automatic optimization method for anti-fraud rules, characterized in that, include: The system acquires the real-time data stream of the numbers involved in the case and all rules in the rule base. It then extracts features from the real-time data stream of the numbers involved in the case to obtain the features involved in the case. At the same time, it performs semantic parsing of all rules through a large language model to obtain the rule intent of each rule. From the rule intent of all rules, rules that match the characteristics of the case are selected as rules to be optimized; The optimization problem of the rule to be optimized is transformed into an optimization decision problem of the agent. The optimal decision of the optimization decision problem is calculated and the optimal decision is executed to achieve automatic optimization of the rule to be optimized.

2. The automatic optimization method for anti-fraud rules as described in claim 1, characterized in that, The features involved in the case include features of text messages involved in the case, features of calls involved in the case, and features of data traffic involved in the case. The feature extraction from the real-time involved phone number to obtain the features involved in the case includes: The real-time data stream of the number involved in the case is input into a pre-built feature extraction model to extract SMS features, call features, and traffic features respectively, to obtain the corresponding SMS features, call features, and traffic features involved in the case. The real-time data stream of the number involved in the case includes SMS data, call data, and traffic data.

3. The automatic optimization method for anti-fraud rules as described in claim 1, characterized in that, The step of transforming the optimization problem of the rule to be optimized into an optimization decision problem of the agent, and calculating the optimal decision of the optimization decision problem includes: Obtain the current threshold and historical performance index of the rule to be optimized, and construct the state space of the optimization decision problem based on the current threshold, the historical performance index and the case-related features; The action space of the optimization decision problem is constructed based on the adjustment direction and adjustment magnitude of the current threshold; The optimal decision for the optimization problem is calculated based on the state space and the action space.

4. The automatic optimization method for anti-fraud rules as described in claim 3, characterized in that, The calculation of the optimal decision for the optimization problem based on the state space and the action space includes: The intelligent agent simulates the execution result of the optimal decision based on the real-time data stream of the number involved in the case, and calculates the reward value of the execution result through a multi-objective reward function; Determine whether the reward value exceeds a first threshold. If not, recalculate the optimal decision for the optimization decision problem based on the execution result, the state space, and the action space.

5. The automatic optimization method for anti-fraud rules as described in claim 4, characterized in that, The intelligent agent simulates the execution result of the optimal decision based on the real-time data stream of the involved number, and calculates the reward value of the execution result through a multi-objective reward function, including: The recall rate, false positive rate, and stability rate are calculated based on the execution results. The reward value of the execution results is then calculated using the recall rate, false positive rate, stability rate, and a multi-objective reward function, whereby: Where R represents the reward value, This represents the first weighting coefficient. Indicates recall rate, This represents the second weighting coefficient. Indicates the false alarm rate. This represents the third weighting coefficient. This indicates the stability rate.

6. An automatic optimization system for anti-fraud rules, characterized in that, include: The acquisition module is used to acquire the real-time data stream of the numbers involved in the case and all the rules in the rule base, extract features from the real-time data stream of the numbers involved in the case to obtain the features involved in the case, and at the same time perform rule semantic parsing on all rules through a large language model to obtain the rule intent of each rule; The matching module is used to filter out rules that match the features involved in the case from the rule intent of all rules as rules to be optimized; An optimization module is used to transform the optimization problem of the rule to be optimized into an optimization decision problem of the agent, calculate the optimal decision of the optimization decision problem, and execute the optimal decision to achieve automatic optimization of the rule to be optimized.

7. The automatic optimization system for anti-fraud rules as described in claim 6, characterized in that, The characteristics involved in the case include characteristics of text messages involved in the case, characteristics of calls involved in the case, and characteristics of data traffic involved in the case. The acquisition module specifically includes: The real-time data stream of the number involved in the case is input into a pre-built feature extraction model to extract SMS features, call features, and traffic features respectively, to obtain the corresponding SMS features, call features, and traffic features involved in the case. The real-time data stream of the number involved in the case includes SMS data, call data, and traffic data.

8. The automatic optimization system for anti-fraud rules as described in claim 6, characterized in that, The optimization module specifically includes: Obtain the current threshold and historical performance index of the rule to be optimized, and construct the state space of the optimization decision problem based on the current threshold, the historical performance index and the case-related features; The action space of the optimization decision problem is constructed based on the adjustment direction and adjustment magnitude of the current threshold; The optimal decision for the optimization problem is calculated based on the state space and the action space.

9. The automatic optimization system for anti-fraud rules as described in claim 8, characterized in that, The optimization module includes: The reward iteration module is used by the agent to simulate the execution result of the optimal decision based on the real-time data stream of the number involved in the case, and to calculate the reward value of the execution result through a multi-objective reward function; Determine whether the reward value exceeds a first threshold. If not, recalculate the optimal decision for the optimization decision problem based on the execution result, the state space, and the action space.

10. The automatic optimization system for anti-fraud rules as described in claim 9, characterized in that, The reward iteration module specifically includes: The recall rate, false positive rate, and stability rate are calculated based on the execution results. The reward value of the execution results is then calculated using the recall rate, false positive rate, stability rate, and a multi-objective reward function, whereby: Where R represents the reward value, This represents the first weighting coefficient. Indicates recall rate, This represents the second weighting coefficient. Indicates the false alarm rate. This represents the third weighting coefficient. This indicates the stability rate.

Images