Strategy improvement method, device, apparatus, medium and product

Abstract

This application provides a strategy improvement method, apparatus, device, medium, and product. The method includes: generating a failure vector based on operational data corresponding to a target strategy; determining the failure type of the target strategy based on the failure vector; generating a target strategy modification scheme according to the failure type, and constraining the target strategy modification scheme through consistency control rules; adjusting the target strategy using the constrained target strategy modification scheme to obtain an improved strategy. This method aims to achieve targeted and accurate adjustments to existing strategies while maintaining overall stability during strategy optimization.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence detection technology, and in particular to a strategy improvement method, apparatus, equipment, medium and product. Background Technology

[0002] Typically, when conducting security checks on content submitted by users, service providers use methods such as rule bases, threshold configurations, or continuous model learning to identify text risks and update policies.

[0003] However, the above solutions mainly involve replacing rules as a whole or adjusting parameters, with a coarse update granularity, making it difficult to accurately locate specific failure links in the detection chain. At the same time, they lack unified constraints on the relationship between each judgment link, and local modifications can easily cause fluctuations in the false positive rate and false negative rate, affecting system stability and business adaptability.

[0004] Therefore, how to accurately locate and target the failure of the detection strategy, and balance overall consistency and stability during the strategy adjustment process, has become a technical problem that needs to be solved. Summary of the Invention

[0005] This application provides a strategy improvement method, apparatus, device, medium, and product to achieve targeted and accurate adjustments to existing strategies while maintaining overall stability during strategy optimization.

[0006] In a first aspect, embodiments of this application provide a strategy improvement method, including:

[0007] Based on the operational data corresponding to the target strategy, a failure vector is generated; the failure vector consists of multiple indicator data representing the failure of the strategy.

[0008] Based on the failure vector, determine the failure type of the target strategy;

[0009] Based on the failure type, a target policy modification plan is generated, and the target policy modification plan is constrained by consistency control rules.

[0010] By adopting the constrained target strategy modification scheme, the target strategy is adjusted to obtain the improved strategy.

[0011] In one possible embodiment, the operational data includes multi-dimensional metric data;

[0012] Based on the operational data corresponding to the target strategy, a failure vector is generated, including:

[0013] Normalize the indicator data from multiple dimensions to obtain multiple normalized indicator data;

[0014] The set of multiple normalized index data is used as the failure vector.

[0015] In one possible embodiment, the multi-dimensional indicator data includes at least the route bypass rate, semantic novelty, multi-module conflict rate, manual reversal rate, appeal reversal rate, scene concentration, new identification gain change rate, and stability degradation rate corresponding to the target policy.

[0016] In one possible embodiment, the indicator data of multiple dimensions are pre-set with corresponding first weight values;

[0017] Based on the failure vector, the failure type of the target strategy is determined, including:

[0018] Based on the first weight value corresponding to the indicator data of multiple dimensions, the weighted sum of multiple normalized indicator data in the failure vector is performed to determine the failure parameters of the target strategy.

[0019] From the pre-defined mapping rules, the failure type corresponding to the parameter range that the failure parameter matches is found; the mapping rules include a one-to-one mapping relationship between multiple parameter ranges and multiple failure types.

[0020] In one possible embodiment, the multiple failure types include at least missed detection failure, misjudged failure, scenario-spreading failure, and aging failure.

[0021] In one possible embodiment, a target policy modification scheme is generated based on the failure type, and the target policy modification scheme is constrained by consistency control rules, including:

[0022] Based on the failure type, determine the set of modifiable fields corresponding to the target strategy and the evolution templates for multiple fields;

[0023] Based on multiple field evolution templates and a set of modifiable fields, multiple first-strategy modification schemes are generated;

[0024] Based on reference parameters of multiple dimensions corresponding to each first strategy modification scheme, a second strategy modification scheme is determined from multiple first strategy modification schemes;

[0025] Consistency control rules are used to perform consistency control checks on the second policy modification scheme, and the second policy modification scheme that has passed the consistency control check is taken as the target policy modification scheme.

[0026] In one possible embodiment, based on multiple field evolution templates and a set of modifiable fields, multiple first strategy modification schemes are generated, including:

[0027] Based on at least one modifiable field contained in the set of modifiable fields, at least one stopping condition is matched;

[0028] Under the constraint of at least one stopping condition, the modifiable fields in the set of modifiable fields are adjusted according to the evolution template of each field to obtain the first strategy modification scheme corresponding to multiple field evolution templates.

[0029] In one possible embodiment, the reference parameters of multiple dimensions are pre-set with corresponding second weight values;

[0030] Based on reference parameters of multiple dimensions corresponding to each first strategy modification scheme, a second strategy modification scheme is determined from multiple first strategy modification schemes, including:

[0031] For each modification scheme of the first strategy, the reference parameters of multiple dimensions are weighted according to the second weight values ​​corresponding to the reference parameters of multiple dimensions, and the optimal value corresponding to each modification scheme of the first strategy is determined based on the weighted reference parameters.

[0032] The first strategy modification scheme with the highest optimization value is selected as the second strategy modification scheme.

[0033] In one possible embodiment, the reference parameters in multiple dimensions include at least the targeted repair gain of the first strategy modification scheme for the failure type, the amount of non-targeted damage, the amount of spillover impact on related scenarios, the cost of perturbation to the target strategy, and the amount of conflict that violates the consistency control rules.

[0034] In one possible embodiment, before determining the second strategy modification scheme from multiple first strategy modification schemes based on reference parameters of multiple dimensions corresponding to each first strategy modification scheme, the method further includes:

[0035] Based on the failure type, determine the targeted repair gain of the first strategy modification scheme.

[0036] In one possible embodiment, determining the targeted repair gain based on the failure type includes:

[0037] In the case of a failure type that is missed, the targeted repair gain is determined based on the missed sample entry rate and recognition rate of the first strategy modification scheme.

[0038] In the case of a failure type of misjudged row failure, the targeted repair gain is determined based on the correction status of misjudged samples in the first strategy modification scheme.

[0039] In the case of scenario-spreading failure, the targeted repair gain is determined based on the convergence of the misuse scenario of the first strategy modification scheme.

[0040] In the case of aging-type failure, the targeted repair gain is determined based on the performance of the first strategy modification scheme within the degradation window.

[0041] In one possible embodiment, the consistency control rule includes multiple sub-rules corresponding to multiple modifiable fields;

[0042] The second policy modification scheme is subjected to consistency control checks using consistency control rules. The second policy modification scheme that has passed the consistency control check is taken as the target policy modification scheme, including:

[0043] Based on at least one modifiable field contained in the set of modifiable fields in the second strategy modification scheme, at least one corresponding sub-rule is matched;

[0044] The second policy modification scheme is tested using at least one matched sub-rule, and the second policy modification scheme that has passed the consistency control test is taken as the target policy modification scheme.

[0045] Secondly, embodiments of this application provide a strategy improvement apparatus, comprising:

[0046] The generation module is used to generate failure vectors based on the operational data corresponding to the target strategy; the failure vectors are composed of multiple normalized indices that characterize the failure features of the strategy.

[0047] The determination module is used to determine the failure type of the target strategy based on the failure vector;

[0048] The constraint module is used to generate target policy modification schemes based on failure types and to constrain the target policy modification schemes through consistency control rules.

[0049] The adjustment module is used to modify the scheme by adopting the constrained target strategy, adjust the target strategy, and obtain the improved strategy.

[0050] Thirdly, embodiments of this application provide a computer device, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps of the method provided above.

[0051] Fourthly, embodiments of this application provide a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the method provided above.

[0052] Fifthly, embodiments of this application provide a computer program product, including a computer program that, when executed by a processor, implements the steps of the method provided above.

[0053] The strategy improvement method, apparatus, equipment, medium, and product provided in this application can generate a failure vector composed of multiple normalized indicators characterizing the failure features of the strategy based on the operational data corresponding to the target strategy. Furthermore, it can uniformly characterize the failure performance of the target strategy and support the accurate determination of the failure type, thereby providing a clear basis for subsequent strategy modifications. By generating a target strategy modification plan based on the failure type and constraining the target strategy modification plan using consistency control rules, it can adjust the target strategy in a targeted manner while taking into account the correlation and overall coordination between various judgment stages. This reduces the fluctuations in the false positive rate and false negative rate caused by local modifications, and improves the stability and business adaptability of the improved strategy. Attached Figure Description

[0054] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0055] Figure 1 A flowchart illustrating a strategy improvement method provided in this application;

[0056] Figure 2 A schematic diagram of a strategy improvement device provided in this application;

[0057] Figure 3 This is a schematic diagram of the structure of an electronic device provided in this application.

[0058] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0059] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application.

[0060] In user-facing business scenarios such as intelligent customer service, information Q&A, and resource interaction-assisted review, service providers need to identify the text content submitted by users in real time to determine whether relevant content requires additional attention in order to avoid incidents.

[0061] However, in actual business scenarios, users often do not directly use explicit words that require attention. Instead, they express their needs in a more subtle way, such as through more natural descriptions of resource interactions, more subtle arrangements of resource interaction rhythms, or resource transfer methods that are less likely to attract attention.

[0062] Therefore, in the process of recognizing user text, the automated response strategy provided by the service provider not only needs to have the ability to quickly recognize text content, but also needs to maintain recognition accuracy under the condition that the sample expression of user text changes rapidly.

[0063] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will now be described with reference to the accompanying drawings.

[0064] In one embodiment, a strategy improvement method is provided. For example... Figure 1 As shown, the strategy improvement method includes:

[0065] Step 102: Generate a failure vector based on the operational data corresponding to the target strategy; the failure vector consists of multiple indicator data representing the failure of the strategy.

[0066] In this embodiment of the application, this step is used to transform the abnormal phenomena exhibited by the target strategy during actual operation into calculable, comparable, and traceable structured data, so as to facilitate subsequent failure type identification.

[0067] The target strategy can be a single detection strategy for identifying intent from user input text, or it can be a sub-strategy in a set of rules, such as a text detection strategy for asking questions about specific content.

[0068] It should be noted that in this embodiment, the target strategy can be represented by multiple fields, for example, it can be represented as S=<E,P,J,X,A,T> .

[0069] The meanings of the various fields in the target strategy can be shown in Table 1.

[0070] Table 1: Meanings of multiple fields E, P, J, X, A, T in target strategy S:

[0071]

[0072] Operational data can come from log data, monitoring data, or statistical data generated during policy execution by the text access layer, risk detection layer, manual review layer, and policy configuration layer. Specifically, it can include the number of policy triggers, the proportion of releases after a hit, the proportion of transfers to manual review, the proportion of manual re-judgments, the number of missed samples of synonym expressions, the concentration of risk samples in specific scenarios, the correlation results of users' subsequent risk behaviors, and the performance differences before and after policy version changes.

[0073] The failure vector is used to summarize multiple normalized indicators that characterize failure features related to the operation of the target strategy into a vector. Each component in the vector comes from the monitoring indicators extracted from the operation data, which are used to calculate failure parameters and locate the failure type of the strategy.

[0074] In one embodiment, the multi-dimensional indicator data includes at least the route bypass rate, semantic novelty, multi-module conflict rate, manual reversal rate, appeal reversal rate, scene concentration, new identification gain change rate, and stability degradation rate corresponding to the target policy.

[0075] As an example, step 102 can first extract several basic indicator data from the running data, including the route bypass rate, semantic novelty, multi-module conflict rate, manual reversal rate, appeal reversal rate, scene concentration, new identification gain change rate, and stability degradation rate corresponding to the target strategy. Then, missing value imputation, outlier pruning, and time window alignment are performed on each indicator to make the monitoring data from different sources comparable within the same statistical period. Subsequently, each basic indicator is mapped to a unified dimension interval, such as using range normalization, quantile normalization, or a standardized mapping method based on historical stable sample distribution, converting each indicator into a normalized value between 0 and 1, and combining multiple vectors according to a predefined vector order to obtain the failure vector.

[0076] Step 104: Determine the failure type of the target strategy based on the failure vector.

[0077] Failure type can be used to characterize the main abnormal patterns of the target strategy in the current running cycle.

[0078] In one embodiment, as shown in Table 2, the various failure types include at least missed detection failure, misjudged failure, scene-spreading failure, and aging failure.

[0079] Table 2: Meaning and Typical Manifestations of Various Failure Types

[0080]

[0081] In one possible embodiment, a weighted calculation can be performed based on each normalized index in the failure vector, and the calculation result can be compared with a preset failure type rule to determine the failure type of the target strategy.

[0082] Specifically, a corresponding weight wi can be assigned to each dimension of the normalized index in the failure vector, and a scoring function Fk = Σwi × vi can be calculated for a certain type of failure, where Fk represents the comprehensive score of the k-th type of failure, wi represents the contribution of a certain normalized index to the type of failure, and vi represents the normalized value of the corresponding normalized index.

[0083] When the overall score of a certain type of normalized index is higher than the trigger threshold of that type of failure and has the highest priority compared with other failure types, the target strategy can be determined as the corresponding failure type.

[0084] Step 106: Generate a target policy modification plan based on the failure type, and constrain the target policy modification plan through consistency control rules.

[0085] The target strategy modification plan can be reflected in the adjustment results of the entry conditions, feature extraction logic, judgment rules, risk score mapping relationship, context association conditions, whitelist exemption conditions, manual triggering conditions, or multi-field combination relationships in the target strategy.

[0086] In one possible implementation, a modification rule corresponding to the failure type can be invoked based on the failure type, thereby generating one or more target policy modification schemes.

[0087] For example, when the failure type indicator target strategy has a missed detection problem, modification schemes related to the entry conditions can be generated, specifically including expanding the risk expression word set, adding semantically similar expression matching conditions, supplementing context triggering paths, relaxing the original overly strict filtering conditions, and introducing new risk scenario identifiers; when the failure type indicator target strategy has a misjudgment problem, modification schemes related to the judgment logic can be generated, specifically including raising the judgment threshold, adding compliance consultation identification conditions, supplementing training and testing context exemption rules, restricting certain highly ambiguous words to trigger only in specific contexts, and optimizing the risk score synthesis weight.

[0088] In another possible embodiment, multiple target strategy modification schemes can be generated based on failure type combinations within a preset range of modifiable fields. These schemes can then be filtered based on replay sample performance, historical version differences, and conflict constraints, retaining candidate schemes that meet the target indicator improvement requirements for subsequent consistency verification.

[0089] The function of consistency control rules is to constrain the target policy modification scheme to ensure the matching of each field of the adjusted policy.

[0090] Specifically, the steps to constrain the target policy modification scheme through consistency control rules can be as follows: after generating multiple policy modification schemes based on the failure type, use consistency control rules to detect and constrain the policy modification schemes, check whether the relevant fields are coordinated, such as whether the judgment logic needs to be adjusted synchronously after modifying the entry point, so as to avoid global inconsistency caused by local modification.

[0091] In one possible embodiment, consistency control rules may include field linkage rules, boundary control rules, priority conflict rules, version compatibility rules, and business exception protection rules. Field linkage rules are used to check whether related fields are updated synchronously after a field changes. For example, when entry conditions are expanded to include new transaction description patterns, the corresponding semantic tag mapping, risk score range, and manual transfer conditions in the judgment logic also need to be adapted synchronously. Boundary control rules are used to limit the scope of modifications to prevent excessive relaxation or tightening at once, which could cause system instability. Priority conflict rules are used to detect whether new rules conflict with existing whitelists, blacklists, or upper-level overall control strategies. Version compatibility rules are used to confirm that the modification scheme is consistent with the current operating environment, configuration format, and calling interface. Business exception protection rules are used to prevent normal research, training, testing, or compliance consultation texts from being misled into risky paths.

[0092] Step 108: Adopt the modified target strategy with constraints, adjust the target strategy, and obtain the improved strategy.

[0093] Step 108 is used to apply the executable modification result obtained after being constrained by the consistency control rules to the target strategy, thereby forming a new improved strategy and fixing the original failure problem while meeting the matching requirements.

[0094] In one possible embodiment, the constrained target policy modification scheme can be directly applied to the target policy to replace the corresponding field configuration in the target policy, thereby obtaining an improved policy.

[0095] The aforementioned strategy improvement method can accurately locate strategy failures by structuring abnormal behaviors during operation into failure vectors and establishing a correspondence between failure vectors and failure types. Furthermore, by generating modification schemes based on failure types and introducing consistency control rules for constraints, strategy updates can be limited to a range with controllable matching and coordination, avoiding local modifications from disrupting the overall balance. Finally, by applying the constrained modification schemes to the target strategy to form an improved strategy, content detection can balance problem-solving efficiency, strategy update accuracy, and system operation continuity when facing environments with rapid evolution of text expression, continuous changes in business scenarios, and high stability requirements.

[0096] In some optional embodiments, the runtime data includes multi-dimensional metric data;

[0097] Step 102 includes:

[0098] Normalize the indicator data from multiple dimensions to obtain multiple normalized indicator data;

[0099] The set of multiple normalized index data is used as the failure vector.

[0100] When generating failure vectors based on the operational data corresponding to the target strategy, the indicator data of multiple dimensions are first normalized to eliminate the incomparability between different indicators caused by differences in units, value ranges and fluctuation amplitudes.

[0101] Normalization can be achieved by using max-min normalization, interval mapping normalization, or standardization based on historical distribution, to map each indicator to the same numerical scale, so that each indicator can participate in subsequent failure characterization with the same weight.

[0102] After normalization, multiple normalized indicator data are combined in a preset order to form a failure vector corresponding to the target strategy.

[0103] This failure vector can express the anomalous features of the strategy in different dimensions in a vectorized form, which facilitates subsequent similarity calculation, cluster analysis, or classification identification, thus providing a unified input for failure type determination. Furthermore, since the scales of the indicators in each dimension have been standardized, each component in the failure vector can directly reflect the contribution of different indicators to the degree of strategy failure, thereby improving the accuracy of failure localization.

[0104] As an example, the failure vector for a target policy can be represented by the following formula:

[0105] Z(S) =<z1,z2,z3,z4,z5,z6,z7,z8>

[0106] Where Z(S) represents the failure vector, and z1, z2, z3, z4, z5, z6, z7, z8 are used to represent the route bypass rate, semantic novelty, multi-module conflict rate, manual reversal rate, appeal reversal rate, scene concentration, new identification gain change rate, and stability degradation rate corresponding to the target policy, respectively. z1, z2, z3, z4, z5, z6, z7, z8 are all normalized to the interval [0,1].

[0107] It should be noted that the larger any value from z1 to z8, the stronger the corresponding indicator data.

[0108] The definitions of the indicator data z1-z8 are shown in Table 2.

[0109] Table 3: Labels and Definitions for the Eight Indicator Data

[0110]

[0111] The aforementioned strategy improvement method can avoid analytical biases caused by differences in the dimensions of different indicators, making the expression of failure characteristics more stable and consistent, and improving the accuracy and interpretability of subsequent failure type identification. At the same time, this approach facilitates the transformation of strategy operational anomalies from macroscopic statistical results into multidimensional structured representations, providing a reliable basis for targeted modification of the target strategy.

[0112] In some optional embodiments, the indicator data of multiple dimensions are pre-set with corresponding first weight values;

[0113] Step 104 includes:

[0114] Based on the first weight value corresponding to the indicator data of multiple dimensions, the weighted sum of multiple normalized indicator data in the failure vector is performed to determine the failure parameters of the target strategy.

[0115] From the pre-defined mapping rules, the failure type corresponding to the parameter range that the failure parameter matches is found; the mapping rules include a one-to-one mapping relationship between multiple parameter ranges and multiple failure types.

[0116] The first weight value is used to characterize the importance of each dimension indicator in the failure determination. Different weights can be set for the indicator data of multiple dimensions to reflect the differences in the impact of indicators such as route bypass rate, semantic novelty, multi-module conflict rate, manual reversal rate, appeal reversal rate, scene concentration, new identification gain change rate and stability degradation rate on strategy failure.

[0117] In this embodiment, each normalized index can be multiplied with its corresponding first weight value, and the results of the multiplication can be accumulated to obtain the failure parameters, so that the failure parameters can comprehensively reflect the overall state of the target strategy in terms of multidimensional failure characteristics.

[0118] The mapping rules can be pre-established as a correspondence table between parameter ranges and failure types. For example, lower parameter ranges can be associated with entry-triggered failures, middle parameter ranges with distorted judgment criteria, and higher parameter ranges with conflicting modification schemes or overall stability degradation.

[0119] This embodiment can retrieve the parameter range in which the failure parameter is located after obtaining the failure parameter, and output the failure type corresponding to the range, thereby completing the determination of the failure category of the target strategy.

[0120] As an example, the process of calculating the failure parameters of the target strategy in this embodiment can be performed using the following formula:

[0121]

[0122] Where T represents the failure parameter of the target strategy, , , , , , , , These are the first weight values ​​corresponding to eight indicators: route bypass rate, semantic novelty, multi-module conflict rate, manual reversal rate, appeal reversal rate, scene concentration, new recognition gain change rate, and stability degradation rate.

[0123] The aforementioned strategy improvement method eliminates reliance on single indicators or empirical judgments for strategy failures. Instead, it uses a weighted approach based on multi-dimensional indicators for targeted identification, enabling different failure characteristics to be uniformly mapped to a clear type. This method improves the stability and interpretability of failure type determination and provides a basis for generating targeted strategy modification schemes, thereby enhancing the positioning accuracy and adjustment efficiency in content security detection scenarios.

[0124] In some alternative embodiments, step 106 includes:

[0125] Based on the failure type, determine the set of modifiable fields corresponding to the target strategy and the evolution templates for multiple fields;

[0126] Based on multiple field evolution templates and a set of modifiable fields, multiple first-strategy modification schemes are generated;

[0127] Based on reference parameters of multiple dimensions corresponding to each first strategy modification scheme, a second strategy modification scheme is determined from multiple first strategy modification schemes;

[0128] Consistency control rules are used to perform consistency control checks on the second policy modification scheme, and the second policy modification scheme that has passed the consistency control check is taken as the target policy modification scheme.

[0129] Among them, the set of modifiable fields is used to limit the range of policy fields that can be adjusted under the current failure type, and the field evolution template is used to provide modification rules and transformation methods for the corresponding fields so as to generate candidate modification results within the limited range.

[0130] The first strategy modification scheme represents the initial candidate scheme formed after adjusting the modifiable fields based on the field evolution template.

[0131] In one embodiment, the reference parameters in multiple dimensions include at least the targeted repair gain of the first strategy modification scheme for the failure type, the amount of non-targeted damage, the amount of spillover impact on related scenarios, the cost of perturbation to the target strategy, and the amount of conflict that violates the consistency control rules.

[0132] The second strategy modification scheme refers to the preferred candidate scheme obtained by summing up multiple reference parameters and evaluating and screening multiple first strategy modification schemes.

[0133] Consistency control rules are used to verify whether the modification scheme of the second strategy meets the requirements of field matching, logical consistency, and rule constraints.

[0134] As an example, the content of a multi-field evolution template can be shown in the following table:

[0135] Table 4: Meaning and Function of Multiple Field Evolution Templates

[0136]

[0137]

[0138]

[0139]

[0140]

[0141]

[0142]

[0143]

[0144]

[0145]

[0146] Furthermore, the j-th modification scheme of the first strategy can be expressed as:

[0147]

[0148] in, This represents the j-th modification scheme of the first strategy. This represents the set of fields that need to be modified in the j-th first strategy modification scheme. This indicates the specific structural modifications required for the fields involved in the j-th first strategy modification scheme.

[0149] Furthermore, in the step of using consistency control rules to perform consistency control checks on the second strategy modification scheme, the specific consistency control rules used are determined according to the fields that need to be modified in the second strategy modification scheme.

[0150] In one embodiment, the consistency control rule includes multiple sub-rules corresponding to multiple modifiable fields;

[0151] The steps of performing consistency control checks on the second policy modification scheme using consistency control rules, and then using the second policy modification scheme that has passed the consistency control check as the target policy modification scheme, include:

[0152] Based on at least one modifiable field contained in the set of modifiable fields in the second strategy modification scheme, at least one corresponding sub-rule is matched;

[0153] The second policy modification scheme is tested using at least one matched sub-rule, and the second policy modification scheme that has passed the consistency control test is taken as the target policy modification scheme.

[0154] As an example, in this embodiment, the sub-rules corresponding to at least one modifiable field in the set of modifiable fields in the second strategy modification scheme can be matched according to the following table, and used as the consistency control rules adopted by the second strategy modification scheme.

[0155] Table 5: Sub-rules corresponding to multiple fields that need to be modified

[0156]

[0157] The aforementioned strategy improvement method enables the strategy modification process to generate and filter strategies based on failure types, reducing the randomness caused by empirical parameter tuning. Furthermore, it suppresses the damage of local optimal solutions to the overall strategy system through consistency control rules, thereby improving the accuracy, adaptability, and operational stability of strategy updates in content security detection scenarios.

[0158] In some optional embodiments, the step of generating multiple first strategy modification schemes based on multiple field evolution templates and a set of modifiable fields includes:

[0159] Based on at least one modifiable field contained in the set of modifiable fields, at least one stopping condition is matched;

[0160] Under the constraint of at least one stopping condition, the modifiable fields in the set of modifiable fields are adjusted according to the evolution template of each field to obtain the first strategy modification scheme corresponding to multiple field evolution templates.

[0161] In this embodiment, the stop condition is used to limit the termination boundary of field adjustment, which can be composed of conditions such as field change range threshold, template adaptation limit, field conflict trigger threshold, or meeting preset consistency requirements.

[0162] Table 5: Stop conditions for multiple modifiable fields

[0163]

[0164] In this embodiment, when generating multiple first strategy modification schemes, the current value of a field in the set of modifiable fields can be compared with the target value range. Conditions satisfying preset change ranges, dependencies, or conflicts are identified as stopping conditions. Then, under these constraints, the field parameters are adjusted sequentially according to different field evolution templates. Each field evolution template corresponds to a set of adjustment rules, which can generate different combinations of field values ​​while maintaining logical consistency between fields, thereby forming multiple first strategy modification schemes.

[0165] For the same modifiable field, when different field evolution templates give different adjustment directions, the candidate results under each field evolution template can be retained to ensure that the modification scheme of the first strategy has diversity and comparability.

[0166] The aforementioned strategy improvement method constrains the field evolution process through stopping conditions, ensuring that adjustments to each field are completed within controllable boundaries. It also combines multiple field evolution templates to generate multiple candidate first-stage strategy modification schemes, thus providing sufficient input for subsequent scheme selection and consistency control. This approach not only increases the coverage of generated modification schemes but also reduces the risk of uncontrolled field propagation, enabling strategy updates to achieve targeted corrections based on failure types while maintaining structural stability.

[0167] In some optional embodiments, the reference parameters for multiple dimensions are pre-set with corresponding second weight values;

[0168] The steps for determining a second strategy modification scheme from multiple first strategy modification schemes based on reference parameters of multiple dimensions corresponding to each first strategy modification scheme include:

[0169] For each modification scheme of the first strategy, the reference parameters of multiple dimensions are weighted according to the second weight values ​​corresponding to the reference parameters of multiple dimensions, and the optimal value corresponding to each modification scheme of the first strategy is determined based on the weighted reference parameters.

[0170] The first strategy modification scheme with the highest optimization value is selected as the second strategy modification scheme.

[0171] Among them, the targeted repair gain is used to characterize the repair strength of the first policy modification scheme on the target policy.

[0172] As an example, reference parameters for multiple dimensions include the targeted repair gain of the first strategy modification scheme for the failure type, the amount of non-targeted destruction, the amount of spillover impact on related scenarios, the cost of perturbation to the target strategy, and the amount of conflict that violates the consistency control rules.

[0173] Accordingly, the step of weighting the reference parameters of multiple dimensions based on the second weight values ​​corresponding to the reference parameters of multiple dimensions, and determining the optimal value corresponding to each first strategy modification scheme based on the weighted reference parameters, can be expressed by the following formula:

[0174]

[0175] in, Indicating targeting The preferred value of the j-th first strategy modification scheme for this failure type. This represents the targeted repair gain of the j-th first strategy modification scheme for the current failure problem. This represents the non-target destruction amount of the j-th first strategy modification scheme. The second weight value represents the amount of non-target damage. This represents the spillover impact of the j-th first strategy modification on the related scenario. Let represent the perturbation cost of the target policy by the j-th first policy modification scheme. The number of conflicts in the j-th first strategy modification scheme that violates the consistency control rules.

[0176] Among them, non-target damage amount This indicates the degree to which the first strategy modification plan disrupts the already stable and effective capabilities of the target strategy, and is used to prevent the system from ruining what was originally done correctly when fixing the current problem.

[0177] Spillover effect This indicates the degree of disturbance that the first strategy modification scheme causes to the related scenarios, and is used to measure whether the first strategy modification scheme inadvertently affects other business scenarios adjacent to the boundary of the current target strategy.

[0178] Structural disturbance cost This indicates the extent to which the first strategy modification plan alters the overall structure of the target strategy, and is used to prioritize the option that can fix the problem with minimal changes.

[0179] Consistency Conflicts This indicates whether there are still unresolved structural conflicts between the entry layer, process layer, output layer, exception layer, scope layer, and exit layer in the first strategy modification plan.

[0180] In this embodiment, by using the preferred value The largest first strategy modification scheme is used as the second strategy modification scheme. This can map multiple candidate first strategy modification schemes from multi-dimensional reference parameters to a single evaluation quantity, and then highlight the impact of key dimensions based on the second weight value, making it easier to select the first strategy modification scheme with better repair effect and smaller overall disturbance.

[0181] Since the preferred value is formed based on multiple weighted reference parameters, it can avoid the local optimum problem caused by making a choice based on a single indicator, thereby improving the matching degree between the second strategy modification scheme and the current failure type.

[0182] The aforementioned strategy update method can reliably select the best-performing solution from multiple first-strategy modification options, reducing the risk of introducing new misjudgments or missed detections after strategy updates, and improving the consistency and controllability of the strategy correction process. Since the second weight value can reflect the importance of different dimensions, the system can flexibly balance repair benefits and business disturbances in different business scenarios, thereby improving the accuracy and operational stability of strategy adjustments in financial risk control content security detection.

[0183] In some optional embodiments, the step of determining the targeted repair gain of the first strategy modification scheme for the failure type includes:

[0184] In the case of a failure type that is missed, the targeted repair gain is determined based on the missed sample entry rate and recognition rate of the first strategy modification scheme.

[0185] In the case of a failure type of misjudged row failure, the targeted repair gain is determined based on the correction status of misjudged samples in the first strategy modification scheme.

[0186] In the case of scenario-spreading failure, the targeted repair gain is determined based on the convergence of the misuse scenario of the first strategy modification scheme.

[0187] In the case of aging-type failure, the targeted repair gain is determined based on the performance of the first strategy modification scheme within the degradation window.

[0188] The aforementioned strategy update method employs a unified yet distinguishable gain representation for different failure types, enabling the effectiveness evaluation of candidate modification schemes to directly correspond to failure characteristics. This improves the relevance of modification schemes to the target problem and reduces disturbances to irrelevant scenarios. Since targeted repair gain can simultaneously reflect hit recovery, misjudgment correction, scenario convergence, and temporal stability, it can further enhance the controllability, stability, and business continuity of the updated strategy in content security detection.

[0189] It should be understood that although the steps in the flowcharts of the above embodiments are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the above embodiments may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0190] Based on the same inventive concept, this application also provides a strategy improvement apparatus for implementing the strategy improvement method described above. The solution provided by this strategy improvement apparatus is similar to the solution described in the strategy improvement method above. Therefore, the specific limitations in one or more apparatus embodiments provided below can be found in the limitations of the strategy improvement method described above, and will not be repeated here.

[0191] In one embodiment, such as Figure 2 As shown, a strategy improvement device 200 is provided, comprising:

[0192] The generation module 202 is used to generate a failure vector based on the running data corresponding to the target strategy; the failure vector consists of multiple normalized indices that characterize the failure features of the strategy.

[0193] The determination module 204 is used to determine the failure type of the target strategy based on the failure vector;

[0194] The constraint module 206 is used to generate a target policy modification scheme based on the failure type and to constrain the target policy modification scheme through consistency control rules.

[0195] The adjustment module 208 is used to modify the scheme by adopting the constrained target strategy, adjust the target strategy, and obtain the improved strategy.

[0196] In some optional embodiments, the runtime data includes multi-dimensional metric data;

[0197] Generation module 202 is also configured as follows:

[0198] Normalize the indicator data from multiple dimensions to obtain multiple normalized indicator data;

[0199] The set of multiple normalized index data is used as the failure vector.

[0200] In some optional embodiments, the multi-dimensional indicator data includes at least the route bypass rate, semantic novelty, multi-module conflict rate, manual reversal rate, appeal reversal rate, scene concentration, new identification gain change rate, and stability degradation rate corresponding to the target policy.

[0201] In some optional embodiments, the indicator data of multiple dimensions are pre-set with corresponding first weight values;

[0202] Module 204 is also configured as follows:

[0203] Based on the first weight value corresponding to the indicator data of multiple dimensions, the weighted sum of multiple normalized indicator data in the failure vector is performed to determine the failure parameters of the target strategy.

[0204] From the pre-defined mapping rules, the failure type corresponding to the parameter range that the failure parameter matches is found; the mapping rules include a one-to-one mapping relationship between multiple parameter ranges and multiple failure types.

[0205] In some alternative embodiments, the various failure types include at least missed detection failure, false alarm failure, scene-spreading failure, and aging failure.

[0206] In some optional embodiments, the constraint module 206 is further configured to:

[0207] Based on the failure type, determine the set of modifiable fields corresponding to the target strategy and the evolution templates for multiple fields;

[0208] Based on multiple field evolution templates and a set of modifiable fields, multiple first-strategy modification schemes are generated;

[0209] Based on reference parameters of multiple dimensions corresponding to each first strategy modification scheme, a second strategy modification scheme is determined from multiple first strategy modification schemes;

[0210] Consistency control rules are used to perform consistency control checks on the second policy modification scheme, and the second policy modification scheme that has passed the consistency control check is taken as the target policy modification scheme.

[0211] In some optional embodiments, the constraint module 206 is further configured to:

[0212] Based on at least one modifiable field contained in the set of modifiable fields, at least one stopping condition is matched;

[0213] Under the constraint of at least one stopping condition, the modifiable fields in the set of modifiable fields are adjusted according to the evolution template of each field to obtain the first strategy modification scheme corresponding to multiple field evolution templates.

[0214] In some optional embodiments, the reference parameters for multiple dimensions are pre-set with corresponding second weight values;

[0215] Constraint module 206 is also configured as follows:

[0216] For each modification scheme of the first strategy, the reference parameters of multiple dimensions are weighted according to the second weight values ​​corresponding to the reference parameters of multiple dimensions, and the optimal value corresponding to each modification scheme of the first strategy is determined based on the weighted reference parameters.

[0217] The first strategy modification scheme with the highest optimization value is selected as the second strategy modification scheme.

[0218] In some optional embodiments, the reference parameters in multiple dimensions include at least the targeted repair gain of the first strategy modification scheme for the failure type, the amount of non-targeted damage, the amount of spillover impact on related scenarios, the cost of perturbation to the target strategy, and the amount of conflict that violates the consistency control rules.

[0219] In some optional embodiments, the constraint module 206 is further configured to:

[0220] In the case of a failure type that is missed, the targeted repair gain is determined based on the missed sample entry rate and recognition rate of the first strategy modification scheme.

[0221] In the case of a failure type of misjudged row failure, the targeted repair gain is determined based on the correction status of misjudged samples in the first strategy modification scheme.

[0222] In the case of scenario-spreading failure, the targeted repair gain is determined based on the convergence of the misuse scenario of the first strategy modification scheme.

[0223] In the case of aging-type failure, the targeted repair gain is determined based on the performance of the first strategy modification scheme within the degradation window.

[0224] In some optional embodiments, consistency control rules include multiple sub-rules corresponding to multiple modifiable fields;

[0225] Constraint module 206 is also configured as follows:

[0226] Based on at least one modifiable field contained in the set of modifiable fields in the second strategy modification scheme, at least one corresponding sub-rule is matched;

[0227] The second policy modification scheme is tested using at least one matched sub-rule, and the second policy modification scheme that has passed the consistency control test is taken as the target policy modification scheme.

[0228] Each module in the above-mentioned device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in the processor of a computer device in hardware form or independent of it, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0229] Figure 3 A schematic diagram of the structure of the electronic device provided in this application. Figure 3 As shown, the electronic device 30 provided in this embodiment includes at least one processor 301 and a memory 302. Optionally, the electronic device 30 further includes a communication component 303. The processor 301, memory 302, and communication component 303 are connected via a bus 304.

[0230] In the specific implementation process, at least one processor 301 executes computer execution instructions stored in memory 302, causing at least one processor 301 to execute the above-mentioned strategy improvement method.

[0231] The specific implementation process of processor 301 can be found in the above method embodiments, and its implementation principle and technical effect are similar. It will not be repeated here.

[0232] In the above embodiments, it should be understood that the processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in this invention can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules within the processor.

[0233] The memory may include random access memory (RAM) and may also include non-volatile memory (NVM), such as at least one disk storage device.

[0234] The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of illustration, the buses shown in the accompanying drawings are not limited to a single bus or a single type of bus.

[0235] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method.

[0236] This application also provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the above-described method.

[0237] The aforementioned readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The readable storage medium can be any available medium accessible to a general-purpose or special-purpose computer.

[0238] An exemplary readable storage medium is coupled to a processor, enabling the processor to read information from and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can reside in an Application Specific Integrated Circuit (ASIC). Alternatively, the processor and the readable storage medium can exist as discrete components in the device.

[0239] The division of units is merely a logical functional division; in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, devices, or units, and may be electrical, mechanical, or other forms.

[0240] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0241] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0242] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0243] Those skilled in the art will understand that all or part of the steps of the above-described method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0244] Finally, it should be noted that other embodiments of the invention will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This invention is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein, and is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of the invention is limited only by the appended claims.

Claims

1. A strategy improvement method, characterized by, include: Based on the operational data corresponding to the target strategy, a failure vector is generated; The failure vector consists of multiple index data representing the failure of the strategy; Based on the failure vector, the failure type of the target strategy is determined; A target policy modification scheme is generated based on the failure type, and the target policy modification scheme is constrained by consistency control rules. By adopting the constrained target strategy modification scheme, the target strategy is adjusted to obtain the improved strategy.

2. The method of claim 1, wherein, The operational data includes indicator data from multiple dimensions; The generation of failure vectors based on the operational data corresponding to the target strategy includes: Normalize the indicator data from multiple dimensions to obtain multiple normalized indicator data; The set of multiple normalized index data is used as the failure vector.

3. The method of claim 2, wherein, The multi-dimensional indicator data includes at least the route bypass rate, semantic novelty, multi-module conflict rate, manual reversal rate, appeal reversal rate, scene concentration, new identification gain change rate, and stability degradation rate corresponding to the target strategy.

4. The method of claim 2, wherein, The indicator data of multiple dimensions are pre-set with corresponding first weight values; Determining the failure type of the target strategy based on the failure vector includes: Based on the first weight values ​​corresponding to the indicator data of multiple dimensions, the multiple normalized indicator data in the failure vector are weighted and summed to determine the failure parameters of the target strategy. From the pre-defined mapping rules, the failure type corresponding to the parameter range that the failure parameter conforms to is matched; the mapping rules include a one-to-one mapping relationship between multiple parameter ranges and multiple failure types.

5. The method of claim 4, wherein, The various failure types mentioned include at least missed detection failure, misjudged failure, scenario-spreading failure, and aging failure.

6. The method of claim 1, wherein, The step of generating a target policy modification scheme based on the failure type and constraining the target policy modification scheme through consistency control rules includes: Based on the failure type, determine the set of modifiable fields and multiple field evolution templates corresponding to the target strategy; Based on the multiple field evolution templates and the set of modifiable fields, multiple first strategy modification schemes are generated; Based on reference parameters of multiple dimensions corresponding to each of the first strategy modification schemes, a second strategy modification scheme is determined from multiple first strategy modification schemes. The second policy modification scheme is subjected to consistency control detection using consistency control rules, and the second policy modification scheme that has passed the consistency control detection is taken as the target policy modification scheme.

7. The method of claim 6, wherein, The process generates multiple first strategy modification schemes based on multiple field evolution templates and the set of modifiable fields, including: Based on at least one modifiable field included in the set of modifiable fields, at least one stopping condition is matched; Under the constraint of at least one of the stopping conditions, the modifiable fields in the set of modifiable fields are adjusted according to each of the field evolution templates to obtain a first strategy modification scheme corresponding to multiple field evolution templates.

8. The method of claim 6, wherein, The reference parameters for multiple dimensions are pre-set with corresponding second weight values; The step of determining a second strategy modification scheme from multiple first strategy modification schemes based on reference parameters of multiple dimensions corresponding to each of the first strategy modification schemes includes: For each of the first strategy modification schemes, the reference parameters of the multiple dimensions are weighted based on the second weight values ​​corresponding to the reference parameters of the multiple dimensions, and the preferred value corresponding to each of the first strategy modification schemes is determined based on the weighted reference parameters. The first strategy modification scheme with the largest preferred value is used as the second strategy modification scheme.

9. The method of claim 6, wherein, The reference parameters for the multiple dimensions include at least the targeted repair gain of the first strategy modification scheme for the failure type, the amount of non-target damage, the amount of spillover impact on the associated scenario, the cost of perturbation to the target strategy, and the amount of conflict that violates the consistency control rules.

10. The method of claim 9, wherein, The step of determining the targeted repair gain of the first strategy modification scheme for the failure type includes: In the case where the failure type is a missed failure, the targeted repair gain is determined based on the missed sample entry rate and recognition rate of the first strategy modification scheme. In the case where the failure type is a misjudged row failure, the targeted repair gain is determined based on the misjudged sample correction status of the first strategy modification scheme; In the case where the failure type is a scenario-spreading failure, the targeted repair gain is determined based on the convergence of the misuse scenario of the first strategy modification scheme. In the case where the failure type is aging failure, the targeted repair gain is determined based on the performance of the first strategy modification scheme within the degradation window.

11. The method of claim 6, wherein, The consistency control rule includes multiple sub-rules corresponding to multiple modifiable fields; The step of performing consistency control checks on the second policy modification scheme using consistency control rules, and using the second policy modification scheme that has passed the consistency control check as the target policy modification scheme, includes: Based on at least one modifiable field contained in the set of modifiable fields in the second strategy modification scheme, at least one corresponding sub-rule is matched; The second policy modification scheme is tested using at least one of the matched sub-rules, and the second policy modification scheme that has passed the consistency control test is taken as the target policy modification scheme.

12. A policy improvement apparatus characterized by comprising: include: The generation module is used to generate failure vectors based on the runtime data corresponding to the target strategy. The failure vector is composed of normalized indices representing the failure characteristics of multiple strategies; The determination module is used to determine the failure type of the target strategy based on the failure vector; The constraint module is used to generate a target policy modification scheme based on the failure type, and to constrain the target policy modification scheme through consistency control rules; The adjustment module is used to modify the target strategy by adopting the constrained target strategy, and to adjust the target strategy to obtain the improved strategy.

13. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1-11.

14. A computer readable storage medium having stored thereon a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1-11.

15. A computer program product comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1-11.

Images