Risk control rule updating method and device, computer device and readable storage medium

Abstract

This application relates to a method, apparatus, computer device, and readable storage medium for updating risk control rules. The method includes: determining an audit opinion database and a rule optimization strategy table; acquiring audit data to be reviewed, including the original risk control rules and audit opinions to be reviewed; if a target audit opinion matching the audit opinion to be reviewed exists in the audit opinion database, constructing prompt words for verifying the audit data to be reviewed; using the prompt words as verification guidelines, validating the audit data to be reviewed through a preset large model; if the audit data to be reviewed is verified as valid, determining a target association record matching the original risk control rule from the rule optimization strategy table; and updating the original risk control rule based on the rule elements in the target association record. This method can improve the accuracy of risk control rules.

Description

Technical Field

[0001] This application relates to the field of big data technology, and in particular to a risk control rule update method, apparatus, computer equipment, and readable storage medium. Background Technology

[0002] With the rapid development of fintech, in fields such as finance, internet lending, and e-commerce transactions, it is necessary to identify and intercept risks in interactive behaviors based on risk control rules to ensure fund security and reduce risks. Risk control rules can identify abnormal interactive behaviors, thereby achieving risk identification and interception.

[0003] However, with the profound development of the digital economy, risk patterns, business scenarios, and user behaviors have all undergone drastic changes, necessitating the automatic adjustment of risk control rules based on the constantly evolving business environment and risk landscape. Traditional methods of updating risk control rules primarily rely on manual experience and simple statistical analysis. Risk control personnel periodically evaluate and adjust the rules based on their own business experience and basic understanding of market risks.

[0004] Traditional methods of updating risk control rules rely on human experience, which leads to inaccurate updated rules. Therefore, a method is needed to improve the accuracy of risk control rules. Summary of the Invention

[0005] Therefore, it is necessary to provide a risk control rule update method, device, computer equipment, computer-readable storage medium, and computer program product that can improve the accuracy of risk control rules in response to the above-mentioned technical problems.

[0006] Firstly, this application provides a method for updating risk control rules, including:

[0007] Determine the audit opinion database and the rule optimization strategy table, wherein the rule optimization strategy table includes records of the relationship between each audit opinion data in the audit opinion database and its corresponding rule elements;

[0008] Obtain the data to be reviewed, which includes the original risk control rules that have been matched and the review comments to be reviewed;

[0009] If a target audit opinion matching the audit opinion to be reviewed exists in the audit opinion database, then a prompt word is constructed to verify the audit data to be reviewed;

[0010] The prompt words are used as verification guidelines. The validity of the data to be reviewed is verified by a preset large model. If the data to be reviewed is verified as valid, the target association record that matches the original risk control rule is determined from the rule optimization strategy table.

[0011] The original risk control rules are updated based on the rule elements in the target association record.

[0012] In one embodiment, the database for determining review comments includes:

[0013] Obtain expert experience data and manual review opinions of historical transactions, and perform structured processing on the expert experience data to obtain structured knowledge;

[0014] The manual review comments of the historical transactions are analyzed by a large model to obtain standard review comment data that meets the preset requirements;

[0015] Based on the structured knowledge and the standard review opinion data, a multi-level storage structure is determined for the review opinion database.

[0016] In one embodiment, the determination rule optimization strategy table includes:

[0017] The review opinion data in the review opinion database is transformed according to rules to obtain their respective rule elements; each rule element includes at least one of the following: risk control field variables, matching conditions, and execution actions corresponding to the matching conditions.

[0018] Based on the audit comments data and their corresponding rule elements, the relationship records are obtained;

[0019] The rule optimization strategy table is determined based on each of the aforementioned relationship record tables, and the data in the rule optimization strategy table is stored in a multi-level storage structure.

[0020] In one embodiment, the review opinion database contains target review opinions that match the review opinion to be reviewed, including:

[0021] The review comments to be reviewed are broken down into words to obtain the decomposed words;

[0022] Based on the word segmentation, a word segmentation matching query is performed from the first-level storage of the review opinion database. If there is a target word in the first-level storage with a similarity greater than a first preset threshold to the word segmentation, then there is a target review opinion in the first-level storage that matches the review opinion to be reviewed.

[0023] If there is no target word with a similarity greater than the first preset threshold to the decomposed word segmentation, then the semantics of the review opinion to be reviewed are identified;

[0024] Based on the audit semantics, a semantic matching query is performed from the second-level storage of the audit opinion database. If there is a target vector semantic in the second-level storage with a similarity greater than the audit semantics, then there is a target audit opinion in the second-level storage that matches the audit opinion to be reviewed.

[0025] In one embodiment, updating the original risk control rule based on the rule elements in the target association record includes:

[0026] Obtain the event identifier of the transaction event corresponding to the data to be reviewed;

[0027] If a target record exists in the risk control engine that corresponds to the attribution event identifier and the original risk control rule, then the original risk control rule is updated according to the rule elements in the target association record.

[0028] In one embodiment, the method further includes:

[0029] If there is no target record in the risk control engine that corresponds to the attribution event identifier and the original risk control rule, then a new risk control rule is added based on the rule elements in the target association record.

[0030] Secondly, this application also provides a risk control rule updating device, including:

[0031] The data processing module is used to determine the review opinion database and the rule optimization strategy table. The rule optimization strategy table includes the relationship records between each review opinion data in the review opinion database and their corresponding rule elements.

[0032] The data acquisition module is used to acquire the data to be reviewed, which includes the original risk control rules that have been hit and the review opinions to be reviewed.

[0033] The prompt word construction module is used to construct prompt words for verifying the audit data to be reviewed if there is a target audit opinion in the audit opinion database that matches the audit opinion to be reviewed.

[0034] The data verification module is used to use the prompt words as verification guidelines and to verify the validity of the data to be reviewed through a preset large model. If the data to be reviewed is verified as valid, the target association record that matches the original risk control rule is determined from the rule optimization strategy table.

[0035] The rule update module is used to update the original risk control rules based on the rule elements in the target association record.

[0036] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the following steps:

[0037] Determine the audit opinion database and the rule optimization strategy table, wherein the rule optimization strategy table includes records of the relationship between each audit opinion data in the audit opinion database and its corresponding rule elements;

[0038] Obtain the data to be reviewed, which includes the original risk control rules that have been matched and the review comments to be reviewed;

[0039] If a target audit opinion matching the audit opinion to be reviewed exists in the audit opinion database, then a prompt word is constructed to verify the audit data to be reviewed;

[0040] The prompt words are used as verification guidelines. The validity of the data to be reviewed is verified by a preset large model. If the data to be reviewed is verified as valid, the target association record that matches the original risk control rule is determined from the rule optimization strategy table.

[0041] The original risk control rules are updated based on the rule elements in the target association record.

[0042] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, performs the following steps:

[0043] Determine the audit opinion database and the rule optimization strategy table, wherein the rule optimization strategy table includes records of the relationship between each audit opinion data in the audit opinion database and its corresponding rule elements;

[0044] Obtain the data to be reviewed, which includes the original risk control rules that have been matched and the review comments to be reviewed;

[0045] If a target audit opinion matching the audit opinion to be reviewed exists in the audit opinion database, then a prompt word is constructed to verify the audit data to be reviewed;

[0046] The prompt words are used as verification guidelines. The validity of the data to be reviewed is verified by a preset large model. If the data to be reviewed is verified as valid, the target association record that matches the original risk control rule is determined from the rule optimization strategy table.

[0047] The original risk control rules are updated based on the rule elements in the target association record.

[0048] Fifthly, this application also provides a computer program product, including a computer program that, when executed by a processor, performs the following steps:

[0049] Determine the audit opinion database and the rule optimization strategy table, wherein the rule optimization strategy table includes records of the relationship between each audit opinion data in the audit opinion database and its corresponding rule elements;

[0050] Obtain the data to be reviewed, which includes the original risk control rules that have been matched and the review comments to be reviewed;

[0051] If a target audit opinion matching the audit opinion to be reviewed exists in the audit opinion database, then a prompt word is constructed to verify the audit data to be reviewed;

[0052] The prompt words are used as verification guidelines. The validity of the data to be reviewed is verified by a preset large model. If the data to be reviewed is verified as valid, the target association record that matches the original risk control rule is determined from the rule optimization strategy table.

[0053] The original risk control rules are updated based on the rule elements in the target association record.

[0054] The aforementioned risk control rule update method, apparatus, computer equipment, computer-readable storage medium, and computer program product achieve zero-code automatic addition / modification of rule engines by constructing an audit opinion database and establishing an association mapping between audit opinions and rule conditions. For audit data to be reviewed, the validity of the data is verified based on the audit opinion database and the rule optimization strategy table. If the review is valid, a target association record matching the original risk control rule is determined from the rule optimization strategy table based on AI judgment using prompt words. This achieves the update of risk control rules, eliminates the subjectivity of manual validity judgment, and improves the accuracy of risk control rules. Attached Figure Description

[0055] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0056] Figure 1 This is an application environment diagram of the risk control rule update method in one embodiment;

[0057] Figure 2This is a flowchart illustrating a risk control rule update method in one embodiment;

[0058] Figure 3 This is a flowchart illustrating how the review opinion database is determined in one embodiment;

[0059] Figure 4 A flowchart illustrating a method for updating risk control rules in another embodiment;

[0060] Figure 5 This is a flowchart illustrating the risk control rule update method in another embodiment;

[0061] Figure 6 This is a structural block diagram of a risk control rule update device in one embodiment;

[0062] Figure 7 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0063] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0064] In finance, internet lending, and e-commerce, risk control rule engines are core tools for mitigating risk. They typically use pre-defined logical rules to identify and block risks in real-time across scenarios such as transactions, credit, and user behavior. Examples include rules like "triggering review if a single transaction exceeds a specified limit" or "rejecting credit if the number of overdue payments in the past N months is greater than or equal to a preset number." However, with the rapid development of the digital economy, risk patterns, business scenarios, and user behavior have all undergone dramatic changes. Due to market shifts, the limitations of traditional risk control rule engines are becoming increasingly apparent. Pre-defined logical rules no longer meet actual needs. Traditional risk control requires manual intervention to detect market changes. When pre-defined logical rules fail to meet actual needs, and strategy experts manually adjust risk strategies to adapt to the market, a method is needed to improve the accuracy of risk control rules, address dynamic risk environments, and meet real-time requirements.

[0065] The risk control rule update method provided in this application embodiment can be applied to, for example, Figure 1In the application environment shown, terminal 102 communicates with server 104 via a network. A data storage system can store the data that server 104 needs to process. The data storage system can be integrated onto server 104 or located in the cloud or on another network server. The server obtains review comment data from the terminal and determines the review comment database and the rule optimization strategy table associated with the database based on the obtained review comment data. The obtained review comment data includes real-time review comment data and historical review comment data; furthermore, it may include expert experience data and manual review comment data.

[0066] The server retrieves the data to be reviewed from the terminal. The data includes the original risk control rules that have been matched and the review comments to be reviewed. If a target review comment that matches the review comments exists in the review comment database, a prompt word is constructed to verify the data to be reviewed. The prompt word is used as a verification guide to verify the validity of the data to be reviewed through a preset large model. If the data to be reviewed is verified as valid, the target association record that matches the original risk control rule is determined from the rule optimization strategy table. The original risk control rule is updated according to the rule elements in the target association record.

[0067] The terminal 102 can be, but is not limited to, various personal computers, laptops, smartphones, tablets, IoT devices, and portable wearable devices. The server 104 can be a standalone physical server, a server cluster or distributed system consisting of multiple physical servers, or a cloud server providing cloud computing services.

[0068] In one exemplary embodiment, such as Figure 2 As shown, a risk control rule update method is provided, which is applied to... Figure 1 Taking the server in the example, the explanation includes the following steps 202 to 210. Wherein:

[0069] Step 202: Determine the audit opinion database and the rule optimization strategy table. The rule optimization strategy table includes records of the relationships between each audit opinion data in the audit opinion database and their corresponding rule elements.

[0070] The review opinion database is determined based on expert experience data and manual review opinion data, which may include historical risk control data. It's important to note that in the risk control review field, expert experience data relies on experts' extensive industry knowledge and experience in handling past cases to make professional judgments and decisions regarding various risk situations. However, this data is usually in unstructured form. To better detect abnormal interactive behaviors in risk control scenarios, expert experience data undergoes vector structuring processing, ensuring it is stored according to a preset structure. For example, expert experience data might suggest that a transaction is suspected of being a cash-out scheme and recommend rejection due to reasons such as multiple devices and accounts, and abnormal IP addresses from different locations. Alternatively, it might state that although the customer has made multiple large withdrawals, their historical credit is good, and the transaction can be approved.

[0071] Each record in the review opinion database may include review opinion data, a corresponding business scenario identifier, and a timestamp. Rule elements must include at least one of the following: risk control field variables, matching conditions, and the execution action corresponding to the matching conditions.

[0072] For example, expert experience data and manual review opinion data are collected from the risk control operation platform. After data cleaning and semantic deduplication, the processed expert experience data undergoes vector structuring, and the processed manual review opinion data is standardized using a pre-defined large model. Based on the processed expert experience data and standardized manual review opinion data, a review opinion database is determined. For each review opinion data in the database, the relationship between the pre-defined review opinions and rules is optimized, transforming the review opinion data to obtain executable rule conditions (i.e., rule elements), thus obtaining a rule optimization strategy table. For example, the columns of the rule optimization strategy table include standard review opinions, risk control field variable columns, and matching conditions & execution action columns. Corresponding examples include the actual stored data in the table: Common Devices -> deviceId -> deviceIdin user's common device pool; Major Promotion Activities -> tradeAmount > tradeAmount > 60000.

[0073] Furthermore, based on this, real-time data streams can be collected from at least one business data source (including but not limited to transaction systems, third-party data interfaces for user behavior logs, etc.), written into a streaming data processing engine, preprocessed using FlinkSQL, and feature processing logic defined using FlinkSQL on the preprocessed real-time stream data. Multi-source and multi-dimensional data are spliced ​​and aggregated to construct a set of model input features, generating a feature data stream. The deployed AI model is called through the model service interface to obtain multiple features output by the AI ​​model. Based on these multiple features, an input event stream is constructed. Based on pre-configured business rules or risk control rules, the input event stream is matched in the rule engine to generate corresponding review opinion data, which is then updated in the review opinion database.

[0074] Step 204: Obtain the data to be reviewed. The data to be reviewed includes the original risk control rules that were hit and the review comments to be reviewed.

[0075] The purpose of obtaining the data awaiting review is to analyze the reasons for manual approval and whether it constitutes a necessary condition for rule adjustment and optimization, thereby reducing the likelihood of manual review being required when the same conditions are triggered again. This data can be obtained in real-time by monitoring the risk control engine's intercept callback interface, or by periodically scanning records in the review task table for batch retrieval.

[0076] The data awaiting review can be determined in real-time through Flink tasks. Taking remote login as an example, for a transaction review, the risk control rule engine identifies the transaction risk. If the remote login rule is triggered, the transaction is sent to a human reviewer. Review experts analyze the transaction to determine the reason for the human review and whether it should be approved, thus obtaining the expert's review opinion. For example, the review opinion might be that the user switched to a different, commonly used device, therefore the transaction can be approved. After this transaction review is completed, the transaction review data is obtained through a Flink task deployed in the server backend. This data may include the triggered risk control rules, the review expert's review results, and the review opinion—in other words, the data awaiting review.

[0077] Step 206: If a target audit opinion matching the audit opinion to be reviewed exists in the audit opinion database, then construct a prompt word to verify the audit data to be reviewed.

[0078] One way to determine whether there is a target review opinion in the review opinion database that matches the review opinion to be reviewed is to calculate the semantic embedding cosine similarity between the review opinion to be reviewed and each item in the review opinion database, and determine the item with the highest similarity as the target review opinion. Alternatively, the review opinion to be reviewed can be segmented into words, and the segmented words can be matched with each item in the review opinion database for similarity.

[0079] The prompt words can be structured instruction text designed for a pre-defined large model, containing task instructions, hypotheses to be verified, fragments of contextual evidence, verification criteria, and output format constraints. They can be used to transform abstract audit validity verification tasks into inference instructions that the large model can understand and execute. For example, in an exemplary embodiment, the prompt words could be:

[0080] "As an expert in determining the validity of manual review of transaction risk control, you must strictly adhere to the following rules:"

[0081] 1. Judgment Rules Explanation: If the manual review comments can reasonably explain the human review rules hit by the risk control rule engine, the review is judged as "valid"; otherwise, it is judged as "invalid".

[0082] 2. Input information:

[0083] - Automatic risk control review rules: These are the risk control rules that this transaction has triggered;

[0084] - Manual review result: The result of the manual review of the transaction (e.g., Approved).

[0085] - Manual review comments: These are the comments from a human reviewer regarding the transaction.

[0086] 3. Output requirements: Output only "[Valid after review]" or "[Invalid after review]", without any extra text, punctuation, or explanation.

[0087] Please give the judgment result directly:

[0088] Step 208: Using the prompt words as verification guidelines, the validity of the data to be reviewed is verified through a preset large model. If the data to be reviewed is verified as valid, the target correlation record matching the original risk control rule is determined from the rule optimization strategy table.

[0089] Among them, the preset large model can be a large language model that has been injected with knowledge from the field of financial risk control and fine-tuned with instructions. It has the ability to understand risk control terminology, identify contradictions in behavioral patterns, and assess the logical consistency of audit conclusions. It can be used to replace manual judgment of the validity of context-sensitive audits and perform multi-step logical verification on the audit data to be reviewed under the guidance of prompt words.

[0090] For example, prompt words are used as verification guidelines. A pre-set large model is used to verify the validity of the data to be reviewed. If the data to be reviewed is verified as valid, the target correlation record matching the original risk control rule is determined from the rule optimization strategy table. It can be understood that each correlation record in the rule optimization strategy table has its own corresponding risk control rule.

[0091] Step 210: Update the original risk control rules based on the rule elements in the target association record.

[0092] One method for updating the rules is to parse the rule element fields in the target relationship record and call the rule engine API to perform atomic update operations, such as modifying threshold parameters, adding conditional clauses, or replacing feature variables, thereby updating the original risk control rules.

[0093] The aforementioned risk control rule update method combines expert opinions and AI analysis to ensure the accuracy and reliability of the review opinion database and constructs a mapping between review opinions and rule conditions, enabling zero-code automatic addition / modification of the rule engine. For the review data to be reviewed, the validity of the data is verified based on the review opinion database and the rule optimization strategy table. If the review is valid, the system constructs prompt words and uses AI judgment based on these prompt words to determine the target association record matching the original risk control rule from the rule optimization strategy table. This update of the risk control rule eliminates the subjectivity of manual validity judgment and improves the accuracy of the risk control rule.

[0094] In practical applications, expert experience data on risk control is scattered and not uniformly stored. Furthermore, this expert experience data is often verbal and not reusable, resulting in low data utilization. In an exemplary embodiment, a method for determining the audit opinion database is provided, such as... Figure 3 As shown, it includes the following steps:

[0095] Step 302: Obtain expert experience data and manual review opinions of historical transactions, and perform structured processing on the expert experience data to obtain structured knowledge.

[0096] Step 304: Analyze the manual review opinions of historical transactions using a large model to obtain standard review opinion data that meets the preset requirements.

[0097] Step 306: Based on structured knowledge and standard audit comment data, determine the audit comment database with a multi-level storage structure.

[0098] The multi-level storage structure can include two levels. The first level can be a word segmentation storage, where structured knowledge and standard review opinion data are segmented by ik_smart and then stored in an inverted index for later search matching. The second level is a vector library, where structured knowledge and standard review opinion data are vectorized to obtain their corresponding vectors and then stored in the vector library. The review opinion database is determined based on these two levels.

[0099] Understandably, the first-level architecture's word segmentation can accurately segment Chinese words and achieve millisecond-level keyword matching based on inverted indexes; the second-level architecture's vector query has semantic query capabilities, and can be used as a fallback if the first-level architecture cannot match the words. Normally, the first-level architecture is used first for querying because word segmentation is lightweight and has a fast response time; the vector library has high response latency, low throughput, and high hardware costs.

[0100] Based on a two-level storage structure, this paper provides a method for determining whether a target audit opinion matching the audit opinion to be reviewed exists in the audit opinion database, including the following steps:

[0101] The review comments to be reviewed are segmented into words to obtain segmented words; based on the segmented words, a word matching query is performed from the first-level storage of the review comments database. If there is a target word in the first-level storage with a similarity greater than the first preset threshold with the segmented words, then there is a target review comment in the first-level storage that matches the review comments to be reviewed.

[0102] If no target word with a similarity greater than the first preset threshold exists, the semantics of the review opinion to be reviewed are identified; based on the review semantics, a semantic matching query is performed from the second-level storage of the review opinion database. If a target vector semantic with a similarity greater than the second preset threshold exists in the second-level storage, then a target review opinion matching the review opinion to be reviewed exists in the second-level storage.

[0103] The word segmentation can be the set of smallest semantic units obtained after fine-grained segmentation of the text to be reviewed. Based on the word segmentation, word matching queries are performed from the first-level storage of the review opinion database. This can be achieved by submitting each word segment to Elasticsearch's multi_match query, enabling the ngram segmenter and synonym expansion, and aggregating the weighted sum of the matching scores of each word. For example, this operation can be implemented by dynamically weighting the scores of each word using Elasticsearch's script_score, or by building a RedisSearch index to achieve low-latency, memory-level queries, thereby completing literal feature retrieval in milliseconds and quickly filtering out obviously irrelevant historical opinions.

[0104] The target word segment can be the existing word segment in the first-level storage that has the highest similarity to the split word segment. This word segment belongs to a complete review comment data, and the review comment data to which it belongs has a valid association record.

[0105] The semantics to be reviewed can be a dense vector representation obtained by deep semantic encoding of the review comments text. The second-level storage can be used to achieve robust semantic matching, compensating for the expression coverage blind spots of the first-level storage. The second-level storage can be obtained by batch processing all review comment data using the same fine-tuning encoder as the semantics to be reviewed during the review comment database construction phase, generating vectors and writing them into a vector library, while simultaneously maintaining a mapping table between vector IDs and original review comment data IDs. The second-level storage can form a query-index relationship with the semantics to be reviewed; its output target vector semantics are the key input for triggering subsequent prompt word construction; it forms a hierarchical complementary architecture with the first-level storage, jointly constituting the dual-modal retrieval capability of the review comment database. The first preset threshold is greater than the second preset threshold.

[0106] For example, based on the review opinion database, a word segmentation matching query is performed on the review opinions to be reviewed. If a review opinion with a similarity greater than 0.9 is found, then a target review opinion matching the review opinion to be reviewed exists in the first-level storage. If the similarity is less than 0.9, a vector library matching query is performed, and a review opinion with a similarity matching value greater than 0.8 is obtained. Then, a target review opinion matching the review opinion to be reviewed exists in the second-level storage. The step of constructing prompt words to verify the review data to be reviewed is then executed to validate the validity of the review data.

[0107] In the above embodiments, structured knowledge is obtained by structuring expert experience data; and standard review opinion data that meets preset requirements is obtained by analyzing the manual review opinions of the historical transactions through a large model. The combination of expert opinions and AI analysis ensures the accuracy and reliability of the review opinion database. Furthermore, the review opinion database is stored in the form of a multi-level storage structure to meet the query needs in different scenarios.

[0108] In an exemplary embodiment, determining the rule optimization strategy table includes: performing rule transformation on each audit opinion data in the audit opinion database to obtain corresponding rule elements; the rule elements include at least one of risk control field variables, matching conditions, and execution actions corresponding to the matching conditions; obtaining association records based on each audit opinion data and its corresponding rule elements; determining the rule optimization strategy table based on each association record table, wherein the data of the rule optimization strategy table is stored in a multi-level storage structure.

[0109] Among them, risk control field variables can be structured business indicators or feature identifiers identified from review opinion data that can be directly referenced by the risk control rule engine, with clear semantic boundaries and data source definitions. Risk control field variables can be determined by entity recognition and relationship extraction from review opinion text through domain knowledge graphs. Matching conditions can be logical judgment expressions applied to risk control field variables, consisting of comparison operators, threshold parameters, and optional connectors, used to define the quantitative or qualitative boundaries of risk triggering, and can be used to transform abstract risk perception into executable Boolean judgments to determine whether the original risk control rule is triggered. Execution actions can be deterministic handling instructions that the rule engine should execute when the matching conditions are met.

[0110] Rule transformation of the audit opinion data in the audit opinion database can be achieved by taking the audit opinion data as input and sequentially performing risk control field variable identification, matching condition extraction, action normalization, and combining the three to generate structured rule elements. For example, rule transformation of the audit opinion data in the audit opinion database can be achieved by jointly identifying fields, conditions, and actions based on a pre-trained sequence labeling model (such as BERT-CRF), or by constructing a domain grammar template library and combining regular expressions and dependency parsing for matching and filling.

[0111] A multi-level storage structure can be a two-level storage structure, with the first level being an in-memory architecture and the second level being MySQL. The data stored in the in-memory architecture and the data stored in MySQL are exactly the same. The in-memory architecture is used to accelerate the processing speed of the entire process, and the data stored in the in-memory architecture and the data stored in MySQL complement each other to improve processing efficiency. When performing a lookup based on the rule-based optimization strategy table, the corresponding target relationship record can be retrieved from the in-memory architecture and / or MySQL according to actual needs.

[0112] In the above embodiments, by performing rule transformation on each review opinion data in the review opinion database, corresponding rule elements are obtained. Relationship records are obtained based on each review opinion data and its corresponding rule elements. A rule optimization strategy table is determined based on each relationship record, and the data of the rule optimization strategy table is stored in a multi-level storage structure. This solves the pain point that expert experience data cannot be transformed into executable rule conditions, and provides a clear input-output mapping relationship for subsequent automatic rule optimization. In other words, by constructing the association mapping between review opinions and rule conditions, the rule engine can achieve zero-code automatic addition and modification, and the optimization cycle is shortened from days to seconds.

[0113] In one exemplary embodiment, a method for updating risk control rules is provided, such as... Figure 4 As shown, it includes the following steps:

[0114] Step 402: Obtain the event identifier of the transaction event corresponding to the data to be reviewed.

[0115] Step 404: If there is a target record in the risk control engine that corresponds to the attribution event identifier and the original risk control rule, then update the original risk control rule according to the rule elements in the target association record.

[0116] The target record can be understood as the hit record of the original risk control rule corresponding to the transaction. For example, it can be the atomic execution trace generated by the risk control engine for a certain original risk control rule when processing a specific transaction event. It includes the event identifier, rule ID, matching result, trigger timestamp, input feature snapshot, and decision path summary. It can be used as tamper-proof evidence of the actual triggering behavior of the rule to verify whether the review opinion is derived from the valid rule judgment in the actual business flow.

[0117] For example, if there is a target record in the risk control engine that corresponds to the attribution event identifier and the original risk control rule, the original risk control rule is updated according to the risk control field variables, matching conditions and execution actions corresponding to the matching conditions in the target association record, and the updated risk control rule is prompted to the terminal.

[0118] Step 406: If there is no target record in the risk control engine that corresponds to the attribution event identifier and the original risk control rule, then add a new risk control rule based on the rule elements in the target association record.

[0119] Understandably, if there is no target record corresponding to the event identifier and the original risk control rule, it means that a new risk control rule needs to be added.

[0120] For example, if there is no target record in the risk control engine that corresponds to the attribution event identifier and the original risk control rule, a new risk control rule is added based on the risk control field variables, matching conditions and execution actions corresponding to the matching conditions in the target association record, and the terminal is prompted to add the new risk control rule.

[0121] Understandably, for the transactional operations in steps 404 and 406, it is necessary to record the historical rules and conditions beforehand, as well as the currently optimized rules and conditions, and to support post-audit verification.

[0122] In this embodiment, based on the aforementioned determination of the review opinion database and rule optimization strategy table, and the construction of prompt words for verifying the review data to be reviewed, the risk control rules are updated based on whether there are target records in the risk control engine that correspond to the attribution event identifier and the original risk control rules, and prompts are given for the updated risk control rules. This achieves full-process traceability from data collection to rule optimization, has an early warning mechanism for strategy experts, and takes into account both automation and compliance controllability.

[0123] In one exemplary embodiment, a risk control rule update method is provided, such as... Figure 5 As shown, it includes the following steps:

[0124] Step 502: Determine the audit opinion database and the rule optimization strategy table. The rule optimization strategy table includes records of the relationships between each audit opinion data in the audit opinion database and their corresponding rule elements.

[0125] Step 504: Obtain the data to be reviewed. The data to be reviewed includes the original risk control rules that were hit and the review comments to be reviewed.

[0126] Step 506: If a target audit opinion matching the audit opinion to be reviewed exists in the audit opinion database, then construct a prompt word to verify the audit data to be reviewed.

[0127] Step 508: Using the prompt words as verification guidelines, the validity of the data to be reviewed is verified through a preset large model. If the data to be reviewed is verified as valid, the target correlation record matching the original risk control rule is determined from the rule optimization strategy table.

[0128] Step 510: Obtain the event identifier of the transaction event corresponding to the data to be reviewed.

[0129] Step 512: If there is a target record in the risk control engine that corresponds to the attribution event identifier and the original risk control rule, then update the original risk control rule according to the rule elements in the target association record.

[0130] Step 514: If there is no target record in the risk control engine that corresponds to the attribution event identifier and the original risk control rule, then add a new risk control rule based on the rule elements in the target association record.

[0131] Taking login from a different location as an example, a transaction is reviewed. The risk control rule engine identifies the transaction risk. If the transaction violates the rule against login from a different location, it is sent to a human reviewer. The reviewer analyzes the reason for the transaction being sent to a human reviewer and whether it can be approved. The reviewer's opinion is then obtained. For example, the reviewer's opinion may be that the user has switched to another commonly used device, so the transaction can be approved.

[0132] After the transaction review is completed, the transaction review data is obtained through a Flink task deployed on the server backend. This data may include the risk control rules that were hit, as well as the review results and opinions of the review experts, which is the review data that needs to be reviewed.

[0133] If a target audit opinion matching the audit opinion to be reviewed exists in the audit opinion database, then a prompt word is constructed to verify the audit data to be reviewed. The prompt word is used as a verification guide, and the validity of the audit data to be reviewed is verified through a preset large model. If the audit data to be reviewed is verified as valid, then the target association record matching the original risk control rule is determined from the rule optimization strategy table.

[0134] Next, locate the risk control event attribution for this original transaction, find the risk control rule in the rule engine. If the rule engine already has this rule, the rule conditions need to be updated according to the target relationship record. If the rule engine does not have this rule, it needs to be added according to the target relationship record. It is understandable that the specific implementation method of this example can be achieved through the methods described above, and will not be elaborated upon here.

[0135] The system combines expert opinions and AI analysis to ensure the accuracy and reliability of the review opinion database and to build a mapping between review opinions and rule conditions, enabling zero-code automatic addition / modification of the rule engine. For review data to be reviewed, the system verifies the validity of the data based on the review opinion database and the rule optimization strategy table. If the review is valid, AI-based judgment using prompt words identifies target association records matching the original risk control rules from the rule optimization strategy table. Based on the existence of target records in the risk control engine that correspond to the attribution event identifier and the original risk control rules, the risk control rules are updated, and a prompt is given for the updated risk control rules. This achieves full-process traceability from data collection to rule optimization, has an early warning mechanism for strategy experts, balances automation and compliance controllability, eliminates the subjectivity of manual review validity judgment, and improves the accuracy of risk control rules.

[0136] It should be understood that although the steps in the flowcharts of the embodiments described above 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 embodiments described above 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.

[0137] Based on the same inventive concept, this application also provides a risk control rule updating device for implementing the risk control rule updating method described above. The solution provided by this device is similar to the solution described in the above method; therefore, the specific limitations in one or more risk control rule updating device embodiments provided below can be found in the limitations of the risk control rule updating method described above, and will not be repeated here.

[0138] In one exemplary embodiment, such as Figure 6 As shown, a risk control rule update device is provided, including: a data processing module 602, a data acquisition module 604, a prompt word construction module 606, a data verification module 608, and a rule update module 610, wherein:

[0139] Data processing module 602 is used to determine the audit opinion database and the rule optimization strategy table. The rule optimization strategy table includes the relationship records between each audit opinion data in the audit opinion database and their corresponding rule elements.

[0140] The data acquisition module 604 is used to acquire the data to be reviewed, which includes the original risk control rules that have been hit and the review opinions to be reviewed.

[0141] The prompt word construction module 606 is used to construct prompt words for verifying the audit data to be reviewed if there is a target audit opinion in the audit opinion database that matches the audit opinion to be reviewed.

[0142] The data verification module 608 is used to use prompt words as verification guidelines and to verify the validity of the data to be reviewed through a preset large model. If the data to be reviewed is verified as valid, the target correlation record matching the original risk control rule is determined from the rule optimization strategy table.

[0143] The rule update module 610 is used to update the original risk control rules based on the rule elements in the target association record.

[0144] The aforementioned risk control rule update device achieves zero-code automatic addition / modification of the rule engine by constructing an audit opinion database and establishing an association mapping between audit opinions and rule conditions. For audit data to be reviewed, the device verifies the validity of the audit data based on the audit opinion database and the rule optimization strategy table. If the audit is valid, the device constructs prompt words and uses AI judgment based on the prompt words to determine the target association record that matches the original risk control rule from the rule optimization strategy table. This update of the risk control rule eliminates the subjectivity of manual validity judgment and improves the accuracy of the risk control rule.

[0145] In an exemplary embodiment, the data processing module 602 is further configured to acquire expert experience data and manual review opinion data of historical transactions, and perform structured processing on the expert experience data to obtain structured knowledge.

[0146] By analyzing the manual review comments of historical transactions through a large model, standard review comment data that meets the preset requirements is obtained;

[0147] Based on structured knowledge and standard audit comment data, a multi-level storage structure audit comment database is determined.

[0148] In an exemplary embodiment, the data processing module 602 is further configured to perform rule transformation on each audit opinion data in the audit opinion database to obtain corresponding rule elements; the rule elements include at least one of risk control field variables, matching conditions, and execution actions corresponding to the matching conditions.

[0149] Based on the data from each audit comment and their corresponding rule elements, the relationship records are obtained;

[0150] The rule optimization strategy table is determined based on the record tables of each relationship, and the data in the rule optimization strategy table is stored in a multi-level storage structure.

[0151] In an exemplary embodiment, the above-described apparatus further includes a data query module, which is used to perform word segmentation on the review comments to be reviewed, and obtain the segmented words.

[0152] Based on the word segmentation, a word segmentation matching query is performed from the first-level storage of the review opinion database. If there is a target word in the first-level storage with a similarity greater than the first preset threshold with the word segmentation, then there is a target review opinion in the first-level storage that matches the review opinion to be reviewed.

[0153] If there is no target word with a similarity greater than the first preset threshold to the decomposed word segmentation, then identify the semantics of the review comments to be reviewed.

[0154] Based on the review semantics, a semantic matching query is performed from the second-level storage of the review opinion database. If there is a target vector semantic in the second-level storage with a similarity greater than the second preset threshold to the review semantics, then there is a target review opinion in the second-level storage that matches the review opinion to be reviewed.

[0155] In an exemplary embodiment, the rule update module 610 is used to obtain the attribution event identifier of the transaction event corresponding to the data to be reviewed;

[0156] If a target record exists in the risk control engine that corresponds to the attribution event identifier and the original risk control rule, the original risk control rule is updated based on the rule elements in the target association record.

[0157] In an exemplary embodiment, the rule update module 610 is used to add a new risk control rule based on the rule elements in the target association record if there is no target record in the risk control engine corresponding to the attribution event identifier and the original risk control rule.

[0158] Each module in the aforementioned risk control rule update device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0159] In one exemplary embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 7 As shown, this computer device includes a processor, memory, input / output interfaces (I / O), and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores data related to risk control rule updates. The I / O interfaces are used for information exchange between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a risk control rule update method.

[0160] Those skilled in the art will understand that Figure 7 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0161] In one embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above method embodiments.

[0162] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon that, when executed by a processor, implements the steps in the above method embodiments.

[0163] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0164] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0165] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence (AI) processors, etc., and are not limited to these.

[0166] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0167] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for updating risk control rules, characterized in that, The method includes: Determine the audit opinion database and the rule optimization strategy table, wherein the rule optimization strategy table includes records of the relationship between each audit opinion data in the audit opinion database and its corresponding rule elements; Obtain the data to be reviewed, which includes the original risk control rules that have been matched and the review comments to be reviewed; If a target audit opinion matching the audit opinion to be reviewed exists in the audit opinion database, then a prompt word is constructed to verify the audit data to be reviewed; The prompt words are used as verification guidelines. The validity of the data to be reviewed is verified by a preset large model. If the data to be reviewed is verified as valid, the target association record that matches the original risk control rule is determined from the rule optimization strategy table. The original risk control rules are updated based on the rule elements in the target association record.

2. The method according to claim 1, characterized in that, The database for determining review comments includes: Obtain expert experience data and manual review opinions of historical transactions, and perform structured processing on the expert experience data to obtain structured knowledge; The manual review comments of the historical transactions are analyzed by a large model to obtain standard review comment data that meets the preset requirements; Based on the structured knowledge and the standard review opinion data, a multi-level storage structure for the review opinion database is determined.

3. The method according to claim 1, characterized in that, The determination rule optimization strategy table includes: The review opinion data in the review opinion database is transformed according to rules to obtain their respective rule elements; each rule element includes at least one of the following: risk control field variables, matching conditions, and execution actions corresponding to the matching conditions. Based on the audit comments data and their corresponding rule elements, the relationship records are obtained; The rule optimization strategy table is determined based on each of the aforementioned relationship record tables, and the data in the rule optimization strategy table is stored in a multi-level storage structure.

4. The method according to any one of claims 1 to 3, characterized in that, The review opinion database contains target review opinions that match the review opinion to be reviewed, including: The review comments to be reviewed are broken down into words to obtain the decomposed words; Based on the word segmentation, a word segmentation matching query is performed from the first-level storage of the review opinion database. If there is a target word in the first-level storage with a similarity greater than a first preset threshold to the word segmentation, then there is a target review opinion in the first-level storage that matches the review opinion to be reviewed. If there is no target word with a similarity greater than the first preset threshold to the decomposed word, then the semantics of the review opinion to be reviewed are identified; Based on the audit semantics, a semantic matching query is performed from the second-level storage of the audit opinion database. If there is a target vector semantic in the second-level storage with a similarity greater than a second preset threshold to the audit semantics, then there is a target audit opinion in the second-level storage that matches the audit opinion to be reviewed.

5. The method according to claim 1, characterized in that, The step of updating the original risk control rule based on the rule elements in the target association record includes: Obtain the event identifier of the transaction event corresponding to the data to be reviewed; If a target record exists in the risk control engine that corresponds to the attribution event identifier and the original risk control rule, then the original risk control rule is updated according to the rule elements in the target association record.

6. The method according to claim 5, characterized in that, The method further includes: If there is no target record in the risk control engine that corresponds to the attribution event identifier and the original risk control rule, then a new risk control rule is added based on the rule elements in the target association record.

7. A risk control rule update device, characterized in that, The device includes: The data processing module is used to determine the review opinion database and the rule optimization strategy table. The rule optimization strategy table includes the relationship records between each review opinion data in the review opinion database and their corresponding rule elements. The data acquisition module is used to acquire the data to be reviewed, which includes the original risk control rules that have been hit and the review opinions to be reviewed. The prompt word construction module is used to construct prompt words for verifying the audit data to be reviewed if there is a target audit opinion in the audit opinion database that matches the audit opinion to be reviewed. The data verification module is used to use the prompt words as verification guidelines and to verify the validity of the data to be reviewed through a preset large model. If the data to be reviewed is verified as valid, the target association record that matches the original risk control rule is determined from the rule optimization strategy table. The rule update module is used to update the original risk control rules based on the rule elements in the target association record.

8. A computer device comprising a memory and a processor, wherein the memory stores 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 to 6.

9. A computer-readable storage medium having a computer program stored thereon, 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 to 6.

10. 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 to 6.

Images