Rule engine data processing method and system based on artificial intelligence and electronic equipment

Abstract

The invention provides a rule engine data processing method and system based on artificial intelligence and electronic equipment, relates to the technical field of data processing and is applied to an enterprise-level automatic data circulation system, and the system comprises multiple levels of priority association rules and is used for processing cross-system multi-dimensional service data; the method comprises the following steps: when a high-priority rule cannot be matched with part of input data, inputting unsuccessfully matched data into a pre-trained artificial intelligence model, and predicting an association success rate of the data in a low-priority rule; if the prediction success rate is lower than a set threshold value, a pre-locking instruction is generated, the data is eliminated in advance before a subsequent rule is executed, and invalid processing is avoided; the artificial intelligence model is introduced to carry out pre-prediction and data screening, redundancy calculation is effectively reduced, the matching accuracy and the system resource utilization rate are improved, and the technical problems of low processing efficiency, poor accuracy and low resource utilization rate in the prior art are solved.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and more specifically, to a data processing method, system, and electronic device based on an artificial intelligence-based rule engine. Background Technology

[0002] In the field of automated data processing for enterprise-level information systems, rule engines are widely used to achieve correlation matching and flow control between multi-source heterogeneous data. These systems typically filter, compare, and bind massive amounts of data based on pre-defined business logic to support critical business processes such as financial reconciliation, order settlement, and contract fulfillment. Traditional rule engines employ a static configuration mechanism, requiring users to manually define the execution priority, matching conditions, and data locking strategies for each rule, which are then executed sequentially. While this approach can meet basic needs in simple scenarios with a small number of rules and stable data structures, it has gradually revealed its insufficient adaptability in high-concurrency and high-complexity real-world applications.

[0003] As enterprises become increasingly digitalized, data interactions across systems are becoming more frequent, leading to a trend of expanding rule systems, increasing hierarchical complexity, and intensified dynamic evolution. Existing rule engines generally rely on human experience for initial configuration, making them difficult to adjust quickly in the face of changes in data distribution or new business scenarios. Furthermore, due to a lack of ability to predict the subsequent behavior of unmatched data, all data must be matched step-by-step, resulting in a significant waste of computing resources on low-success-rate, ineffective paths. In addition, during task execution, especially with manual intervention, the system often fails to effectively identify potential operational risks, easily leading to data consistency issues. These limitations pose serious challenges to the efficiency, accuracy, and security of traditional rule enforcement models. Summary of the Invention

[0004] The purpose of this invention is to provide a data processing method, system, and electronic device based on artificial intelligence rule engine, so as to alleviate the technical problems of low processing efficiency, poor accuracy, and low resource utilization in the prior art.

[0005] In a first aspect, embodiments of the present invention provide an artificial intelligence-based rule engine data processing method applied to an enterprise-level automated data flow system. The enterprise-level automated data flow system is configured with multiple priority-based association rules for matching and processing multi-dimensional business data across systems. The method includes: responding to the completion of an operation to match input data based on a first priority association rule, and the existence of first unmatched pending data, inputting the first pending data into a pre-trained artificial intelligence model; based on the first pending data, using the artificial intelligence model to perform prediction, generating an association success rate of the first pending data in a second priority association rule; if the association success rate is lower than a preset success rate threshold, generating a pre-locking instruction for the first pending data; based on the pre-locking instruction, excluding the first pending data before executing subsequent association rules in the current execution group.

[0006] In some optional implementations, the above method further includes: when determining the actual execution order of each associated rule within the current execution group, the actual execution order is generated by a pre-deployed artificial intelligence scheduling model based on the data volume distribution of the current batch of pending data and the system resource load status.

[0007] In some optional implementations, the above method further includes: in response to receiving a manual execution request initiated by the user, detecting the association rules corresponding to the selected task, and determining whether there are association rules with higher execution order but not yet executed; if so, inputting the rule characteristics of the association rules corresponding to the selected task, the configuration information of the higher priority association rules that have not been executed, and log data of historical matching success rate into the pre-trained artificial intelligence risk assessment model.

[0008] In some optional implementations, the above method also includes: generating a risk assessment report containing the predicted data error rate and the impact level of downstream businesses based on the output of the above artificial intelligence risk assessment model; generating corresponding interactive prompt information based on the impact level of downstream businesses in the above risk assessment report; and enabling a mandatory confirmation mechanism when the impact level of downstream businesses reaches the severe level, prohibiting the continued execution of the selected task without receiving explicit confirmation from the user.

[0009] In some optional implementations, the above method further includes: during the association rule configuration phase, in response to the operation of adding or modifying association rules, invoking a pre-trained artificial intelligence recommendation model to generate priority suggestions and non-priority association locking strategy suggestions for the rule; wherein, the priority suggestions are generated based on the rule's matching field combination, filtering strictness, types of data sources involved, expected data volume to be processed, and historical matching success rate; the non-priority association locking strategy suggestions include recommending locking parent-level associated data, locking child-level associated data, or locking all associated data.

[0010] In some optional implementations, the above method further includes: periodically receiving matching execution logs and user feedback data within a preset period, and training the artificial intelligence model based on the user feedback data to update the parameters of the artificial intelligence model.

[0011] In some optional implementations, the above method further includes: recording the deviation between the recommended value output by the above artificial intelligence model and the user's actual configuration value, and inputting the above deviation as a negative sample into the next round of training of the above artificial intelligence model to correct the model deviation.

[0012] Secondly, embodiments of the present invention provide an artificial intelligence-based rule engine data processing system for multi-dimensional business data matching processing in an enterprise-level automated data flow system. The enterprise-level automated data flow system is configured with multiple priority-based association rules. The artificial intelligence-based rule engine data processing system includes: a data input module, used to input the first unprocessed data into a pre-trained artificial intelligence model in response to the completion of an operation matching input data based on a first priority association rule, and the existence of first unprocessed data that has not been successfully matched; a model prediction module, used to predict the association success rate of the first unprocessed data in a second priority association rule based on the first unprocessed data using the artificial intelligence model; a pre-locking decision module, used to generate a pre-locking instruction for the first unprocessed data if the association success rate is lower than a preset success rate threshold; and a process control module, used to exclude the first unprocessed data before executing subsequent association rules in the current execution group based on the pre-locking instruction.

[0013] Thirdly, embodiments of the present invention provide an electronic device, including a memory and a processor, wherein the memory stores a computer program that can run on the processor, and the processor executes the computer program to implement the steps of the method described in any of the first aspects above.

[0014] Fourthly, embodiments of the present invention provide a computer-readable storage medium storing computer-executable instructions, which, when invoked and executed by a processor, cause the processor to perform the method described in any of the first aspects above.

[0015] This invention provides a data processing method, system, and electronic device based on an artificial intelligence rule engine. The method is applied to an enterprise-level automated data flow system, which is configured with multiple priority-based association rules for matching and processing multi-dimensional business data across systems. The method includes: in response to the completion of an operation matching input data based on a first priority association rule, and the existence of first unmatched pending data, inputting the first pending data into a pre-trained artificial intelligence model; based on the first pending data, using the artificial intelligence model to predict and generate an association success rate for the first pending data in a second priority association rule; if the association success rate is lower than a preset success rate threshold, generating a pre-locking instruction for the first pending data; based on the pre-locking instruction, excluding the first pending data before executing subsequent association rules in the current execution group. This invention solves the technical problems of low processing efficiency, poor accuracy, and low resource utilization in existing technologies, achieving the technical effects of shortening the rule configuration cycle, reducing the error rate, and improving data reliability. Attached Figure Description

[0016] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments of the present invention will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 A flowchart illustrating a data processing method for a rule engine based on artificial intelligence, provided in an embodiment of the present invention; Figure 2 This invention provides an application example of a data processing method for a rule engine based on artificial intelligence. Figure 3 A schematic diagram of the structure of a rule engine data processing system based on artificial intelligence provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] In enterprise-level automated data flow scenarios, such as cross-system multi-dimensional automatic reconciliation, fund clearing, supply chain settlement matching, and invoice-contract-order three-way matching, there are typically high requirements for concurrency, high complexity, and high accuracy. In such scenarios, the number of rules can reach hundreds, with priority levels exceeding six. The data matching results directly affect the accuracy of financial statements and the security of funds, thus placing extremely high demands on the system's real-time decision-making capabilities and execution efficiency. However, traditional rule engines generally rely on static configurations, with key parameters such as rule priority and data locking strategies primarily set based on human experience, lacking flexibility and adaptability. Once the business environment or data distribution changes, the original configuration often becomes inapplicable, requiring frequent manual adjustments, resulting in high maintenance costs and slow response times. Furthermore, existing systems cannot intelligently predict unmatched data during execution; all data must be matched level by level, leading to a large amount of invalid computation, especially with large datasets where performance significantly degrades. In addition, due to the complex rule hierarchy and tight task arrangement logic, configuration errors or human operation mistakes are difficult to be detected in time during execution. Problems are usually only exposed after batch processing is completed, resulting in high operational risks and potential data inconsistencies, and the overall fault tolerance is weak.

[0020] Based on this, the present invention provides a data processing method, system and electronic device for a rule engine based on artificial intelligence, in order to solve the technical problems of low processing efficiency, poor accuracy and low resource utilization in the prior art.

[0021] To facilitate understanding of this embodiment, a detailed description of an artificial intelligence-based rule engine data processing method disclosed in this embodiment of the invention will be provided first. (See [link to relevant documentation]). Figure 1 The diagram illustrates a data processing method based on an artificial intelligence rule engine. This method can be applied to enterprise-level automated data flow systems, which typically have multiple priority-ordered association rules for matching and processing multi-dimensional business data across systems. The method mainly includes the following steps S102 to S108: Step S102: In response to the completion of the operation of matching the input data based on the first priority association rule, and the existence of first unmatched data to be processed, the first data to be processed is input into the pre-trained artificial intelligence model.

[0022] The first priority association rule is the highest execution priority rule preset or dynamically determined in the current execution group. Input data can be raw structured / semi-structured data records from heterogeneous business systems (such as ERP, CRM, and financial systems), including but not limited to purchase order details in ERP systems, customer contract information in CRM systems, and accounts payable vouchers in financial systems.

[0023] Before entering the rule engine, the input data has undergone field alignment, null value filling, data type normalization, and business semantic encoding through the standardized preprocessing module, forming a feature vector with fixed dimensions and uniform scale.

[0024] The first set of data to be processed can refer to the subset of data that failed to establish a valid association during the matching process of the first priority association rule because the field values ​​did not meet the filtering conditions, the logical expression evaluated as false, or the association constraints (such as uniqueness and temporal dependency) conflicted. The aforementioned artificial intelligence model can be a lightweight classification model trained through supervised learning. Its training objective is to predict the probability of successful association of a single data point under a specified subsequent rule. The model structure can adopt the XGBoost algorithm or a TinyBERT variant optimized by knowledge distillation. The deployment mode is offline training and online low-latency inference. The model input interface receives structured feature vectors, and the output interface provides standardized probability values.

[0025] As a specific example, the input features of this artificial intelligence model can be limited to the following four categories: (1) the key field combination encoding of the first data to be processed, which is encoded by jointly embedding the field name, field value hash and the relationship between fields; (2) the business domain label, which takes one of the predefined enumeration sets {financial reconciliation, supply chain settlement, order fulfillment, invoice verification}; (3) the data freshness index, which is defined as the time difference between the current system timestamp and the last update timestamp of the data in the source system, in hours, and is included in the model input after normalization; (4) the historical matching failure pattern clustering identifier of the first priority association rule, which is generated by the system based on the feature distribution of the failed samples in the most recent 30 executions through K-means clustering, and takes the value as an integer clustering number.

[0026] The purpose of step S102 is to provide the first data to be processed, which is not matched by the first priority association rule, as a structured input to the pre-trained artificial intelligence model, so as to initiate the quantitative prediction process of its matching potential in subsequent rules, thereby providing a deterministic data basis for the generation of association success rate in the subsequent step S104.

[0027] Step S104: Based on the first data to be processed, use an artificial intelligence model to make a prediction and generate the association success rate of the first data to be processed in the second priority association rule. The second priority association rule is the next priority rule that follows the first priority association rule and belongs to the same current execution group (i.e., execution order number + 1); the association success rate is the probability value output by the artificial intelligence model, which represents the expected confidence level of the first data to be processed under the constraints of the second priority association rule, which satisfies all matching conditions and successfully establishes a business semantic association.

[0028] The prediction process of the above model integrates a dual context: rule context and data context. Preferably, the rule context may include the field coverage, filtering strictness, and locking strategy type of the second priority association rule; the data context may include the failure attribution distribution and cross-source consistency score of the first data to be processed in historical similar rules.

[0029] Furthermore, the association success rate obtained through step S104 can be compared with a preset success rate threshold, and the comparison result can be used to determine whether to perform a pre-locking operation on the first data to be processed.

[0030] Preferably, the preset success threshold is not a fixed constant, but can be dynamically calibrated based on the current system resource level (automatically reduced to 1.5% when CPU utilization is >85%), business SLA level (such as a default threshold of 3% for financial reconciliation scenarios and 8% for supply chain matching scenarios) and rule grouping risk coefficients, so as to ensure that the pre-locking strategy has both robustness and scenario adaptability.

[0031] Step S106: If the association success rate is lower than the preset success rate threshold, a pre-locking instruction is generated for the first data to be processed. The preset success rate threshold can be pre-set based on the actual business needs of the enterprise-level automated data flow system and stored in the system's configuration center. These business needs include, but are not limited to, one or more of the following: the fault tolerance requirements of the business scenario being processed (for example, financial reconciliation scenarios require high accuracy, so the corresponding threshold is set lower; while preliminary clue matching scenarios allow for a certain number of missed matches, so the corresponding threshold is set higher), the overall size of the current batch of data to be processed, and the system resource load status.

[0032] As a concrete example, system resource load status can include CPU utilization, memory usage, I / O latency, etc.

[0033] Furthermore, the preset success rate threshold supports two modes: static configuration and dynamic adjustment. In static configuration mode, the system administrator inputs a fixed value through the configuration interface. In dynamic adjustment mode, the process control module adjusts the threshold proportionally based on the current system resource load status. The direction of the fluctuation is positively correlated with the load status, that is, the higher the system resource load, the lower the threshold, so as to enhance the pre-locking strength and prioritize system stability.

[0034] Preferably, the preset success rate threshold ranges from 0.1% to 15%; the setting and adjustment process of this threshold is recorded in the system operation log to ensure traceability.

[0035] The pre-locking instruction in step S106 above can be a structured control signal. Preferably, its contents include: a unique identifier of the first data to be processed, the applicable execution scope (limited to all association rules in the current execution group whose order is after the first priority association rule), the instruction effective time point (marked by the system clock synchronization timestamp), and the basis for instruction generation (i.e. the association success rate value and the corresponding preset success rate threshold).

[0036] Once generated, this pre-lock instruction triggers the system to exclude the first piece of data to be processed in subsequent rule matching processes. Specifically, before the data is sent to any subsequent association rule in the current execution group for matching, its further flow within that group is terminated, preventing it from participating in the input processing, condition judgment, and association establishment processes of any subsequent association rules within that group. This exclusion operation is irreversible and does not affect the normal execution path of other data to be processed in the same group. By generating and executing this pre-lock instruction, redundant matching attempts for data with low association probability are effectively avoided, thereby reducing the system's ineffective computational load and improving overall matching efficiency and resource utilization.

[0037] Step S108: Based on the pre-locking instruction, exclude the first data to be processed before executing subsequent association rules in the current execution group.

[0038] The current execution group can be a set of association rules with internal execution constraints, which are aggregated according to business semantics or data source consistency in an enterprise-level automated data flow system. This set is used to limit multi-dimensional business data across systems to a logical closed loop for matching processing.

[0039] Preferably, the current execution group is divided according to business type, such as "financial three-document matching group" (covering invoice-contract-order association rules) and "supplier settlement group" (covering payment application-bank receipt-accounts payable write-off rules).

[0040] Subsequent association rules refer to all association rules following the first priority association rule. They can be used to make progressive association attempts on the same data under different matching strictness or different combinations of data sources. For example, in "financial three-invoice matching grouping", if the first priority association rule is "exact match of invoice code + contract number", then the subsequent association rules can include "invoice amount ±5% tolerance + fuzzy match of contract number" and "matching only the contract number and verifying the time window", etc.

[0041] Excluding the first pending data means terminating the entry of the first pending data into the input queue of subsequent association rules in the data flow path, so that it does not participate in any condition judgment, field comparison and association relationship establishment process, thereby realizing the early termination of the matching process of the data in the current execution group; preferably, the exclusion operation can be achieved by adding a pre-locking mark to the data message header, and the rule scheduler will skip the subsequent rule distribution steps after recognizing the mark.

[0042] In one embodiment, the above method may further include: when determining the actual execution order of each associated rule within the current execution group, generating the actual execution order by a pre-deployed artificial intelligence scheduling model based on the data volume distribution of the current batch of data to be processed and the system resource load status.

[0043] The data volume distribution of the current batch of data to be processed is a statistical description reflecting the expected input scale of various types of data in the batch under different rule matching conditions, and can be used to estimate the data reduction effect after the execution of each rule. The system resource load status is a set of quantitative indicators reflecting the real-time operating pressure of an enterprise-level automated data flow system, and can be used to dynamically constrain the resource consumption of rule execution.

[0044] For example, the data distribution can be represented by a histogram, with the horizontal axis representing the key field combinations that can be covered by the rules (such as "invoice code + contract number" or "contract number only"), and the vertical axis representing the number of data records that satisfy the combination. For instance, in a certain financial reconciliation task, data containing all three invoice fields accounted for 12%, data containing only contract number and invoice code accounted for 67%, and the remainder was data with severely missing fields. The system's resource load status includes average CPU utilization, available memory percentage, and disk I / O wait queue length, with a collection cycle of no more than 5 seconds, updated to the scheduling decision context.

[0045] In one example, the pre-deployed AI scheduling model can be a lightweight ranking model trained on historical execution logs, used to select the sequence that optimizes overall computational cost from multiple candidate execution orders. For example, this model could use ListNet or PointwiseRankingSVM, with the input being the feature vectors of each rule (including expected input data volume, average matching time, and resource utilization coefficient) and the current resource load status, and the output being a ranked sequence of rule IDs.

[0046] In a specific example, when rule R1 (high strictness, low data volume) and rule R2 (low strictness, high data volume) belong to the same group, if the model predicts that executing R1 first can reduce the input data volume of R2 by 83%, then R1 is placed before R2 as the actual execution order; thus, while ensuring matching coverage, the total computation and response latency within the group are significantly reduced.

[0047] The actual execution order in this embodiment differs from the static execution order preset by the user, and a strong locking principle is adhered to during execution, meaning that data that has been successfully matched does not participate in the matching attempts of subsequent rules within the group; thus, while dynamically optimizing the execution path, the semantic integrity of the dependencies between rules is maintained, avoiding duplicate data matching or association conflicts caused by order adjustments.

[0048] In one embodiment, the method may further include: in response to receiving a manual execution request initiated by a user, detecting the association rules corresponding to the selected task, and determining whether there are association rules with a higher execution order but not yet executed; if so, inputting the rule features of the association rules corresponding to the selected task, the configuration information of the higher priority association rules that have not been executed, and log data of historical matching success rates into a pre-trained artificial intelligence risk assessment model.

[0049] Manual execution requests are immediate execution commands for specific orchestrated tasks, proactively triggered by users through the graphical user interface provided by the enterprise-level automated data workflow system. This interface can be integrated into the system management console, providing core operational functions such as task list display, status filtering, one-click execution, and risk confirmation to authorized users such as business operations personnel and system administrators.

[0050] Preferably, the interface adopts a responsive web architecture, supporting access on PCs and mobile devices; for example, in the order fulfillment monitoring dashboard, operations personnel can manually trigger the "T001-Cross-border Order Invoice Verification" task.

[0051] The association rules corresponding to the selected task are usually one or more specific association rules bound to the task in the rule engine, which are used to complete the matching action of specific business semantics; for example, the task "T001-Cross-border Order Invoice Verification" corresponds to rule R3, whose function is to compare the invoice details exported from the ERP system with the amount and product code of the outbound order from the WMS system.

[0052] Rule features can be a set of structured attributes that characterize the essence of the matching behavior of the association rule, used to represent its business impact scope and technical execution characteristics; preferably, the rule features include: filtering conditions (such as "amount difference ≤ 0.5%"), field combinations (such as "order number + SKU + delivery date"), and the business module to which it belongs (such as "supply chain - fulfillment").

[0053] The configuration information for higher-priority association rules that have not been executed can be a set of configurable parameters defined in the system for that rule, which reflects its control intent over the data flow; for example, the configuration information for rule R1 (which has a higher priority than R3) includes its locking policy ("lock all associated data"), matching timeout threshold (300ms), and cross-source consistency check switch (enabled).

[0054] The historical matching success rate log data is the percentage of times the rule has successfully established a valid association in the past N executions, used to characterize its stability and reliability; preferably, N is 30, the statistical window is updated on a rolling basis according to natural days, and the data comes from the execution audit logs automatically collected by the system.

[0055] Artificial intelligence risk assessment models can be binary classification models trained based on historical human intervention events and result feedback, used to quantify the data inconsistency risks that may be caused by skipping high-priority rules.

[0056] In a specific example, the input parameters of this AI risk assessment model include: the filtering conditions, field combinations, and business module to which the current rule R3 to be executed belongs (i.e., rule characteristics); the number of higher priority unexecuted rules R1 (1 in this case), their locking strategy and timeout threshold (i.e. configuration information); and the number of times that downstream financial vouchers were incorrect due to the past 30 similar manual operations (skipping R1 and executing R3 directly) (i.e., the number of failed records in the historical matching success rate log data).

[0057] Furthermore, the output of the AI ​​risk assessment model is a risk score, ranging from 0 to 100, corresponding to four risk levels: "low," "medium," "high," and "severe." This score can be used to drive subsequent risk alerts and mandatory confirmation mechanisms, thereby keeping business losses caused by misoperation within an acceptable range while retaining the flexibility of human intervention.

[0058] In one embodiment, the method may further include: generating a risk assessment report containing a predicted data error rate and a downstream business impact level based on the output of an artificial intelligence risk assessment model; generating corresponding interactive prompts based on the downstream business impact level in the risk assessment report; and activating a mandatory confirmation mechanism when the downstream business impact level reaches a severe level, prohibiting the continued execution of the selected task without receiving explicit confirmation from the user.

[0059] Among them, the predicted data error ratio is a quantitative indicator obtained by calibrating and mapping the risk score output by the artificial intelligence risk assessment model. It is used to characterize the expected proportion of related erroneous data caused by skipping high-priority rules and directly executing the current task in the total processed data.

[0060] Preferably, this value is presented as a percentage, ranging from 0.01% to 45%, and is based on the statistical distribution of the proportion of errors that actually occurred in historical human intervention events.

[0061] The downstream business impact level is a classification indicator determined based on the predicted proportion of data errors and the importance of the business links involved in the erroneous data. It is used to reflect the degree of transmissible harm of the operational error to the terminal business results.

[0062] Preferably, the level is divided into four levels: low (error rate <1%, and does not affect core accounting), medium (1% ≤ error rate <5%, affecting single business line reports), high (5% ≤ error rate <15%, affecting cross-system reconciliation balance), and severe (error rate ≥15%, or erroneous data involves key nodes such as financial closing and fund payment).

[0063] The corresponding level of interactive prompts are structured prompts automatically generated based on the impact level of downstream business, clearly conveying the nature of the risk and response requirements to the operator. Furthermore, the mandatory confirmation mechanism requires users to complete identity verification and risk acknowledgment through explicit interactive actions before the system allows continued execution of the selected task. This mechanism is automatically activated at severity levels, and its execution status is written to the operation audit log in real time, ensuring traceability and accountability throughout the entire process.

[0064] In one embodiment, the above method may further include: during the association rule configuration phase, in response to the operation of adding or modifying an association rule, invoking a pre-trained artificial intelligence recommendation model to generate priority suggestions and non-priority association locking strategy suggestions for the rule; The operation of adding or modifying association rules can be performed by a user in the rule configuration interface by editing, saving, and submitting the rule conditions. This operation can trigger the system to start the AI-assisted decision-making process. Preferably, this operation can be initiated through a web-based configuration wizard or API call. For example, a financial system administrator can create a new invoice fuzzy matching rule based on OCR recognition results in the "Three-invoice Matching Rule Library" and click "Save and Get AI Suggestions".

[0065] The AI ​​recommendation model can be a multi-task learning model trained based on historical rule configuration records and subsequent execution effect feedback. It provides data-driven configuration optimization suggestions at the initial stage of the rule lifecycle. This AI recommendation model can simultaneously output two independent but related suggestions: a priority suggestion and a non-priority related locking strategy suggestion.

[0066] Preferably, the inputs to the aforementioned AI recommendation model include: the matching field combination of the rule to be configured, the filtering strictness, the types of data sources involved, the expected amount of data to be processed, and the historical matching success rate; its output is a structured suggestion pair, where the priority suggestion is an integer number (e.g., P1 to P6), and the priority-external association locking strategy suggestion is an enumeration value ("lock parent-level associated data", "lock child-level associated data", "lock all associated data"). In another example, the priority suggestion can be generated based on the rule's matching field combination, filtering strictness, types of data sources involved, expected amount of data to be processed, and historical matching success rate; the priority-external association locking strategy suggestion can be a constraint recommendation on the eligibility of the associated data to participate in matching in subsequent priority rules after the rule is successfully matched, used to balance matching coverage and computational efficiency.

[0067] Preferably, the priority recommendation includes at least one of the following three types: 1. Lock the parent-level associated data, which is suitable for scenarios where once the parent-level data (such as a contract) is successfully matched, its child-level data (such as an order) is unlikely to participate in the matching of low-priority rules; 2. Lock the child-level associated data, which is suitable for scenarios where, after the child-level data (such as an invoice) is successfully matched, it is still necessary to backtrack to the parent-level data (such as a contract) for supplementary verification; 3. Lock all associated data, which is suitable for scenarios where the matching result has strong finality and subsequent rule retry has no business benefit.

[0068] In one embodiment, the above method may further include: periodically receiving matching execution logs and user feedback data within a preset period, and training the artificial intelligence model based on the user feedback data to update the parameters of the artificial intelligence model.

[0069] The matching execution log can be a structured runtime record automatically recorded after each rule matching task is completed, used to reflect the correspondence between the AI ​​model's predicted behavior and the actual business results. Preferably, the matching execution log includes at least: the first input data feature to be processed, the model's output association success rate, the number of the subsequent association rule actually executed, the matching result (success / failure) of the rule for that data, and the matching time.

[0070] User feedback data can be records of explicit confirmation or correction actions taken by users after the model's recommendation results are presented, used to characterize the degree of consistency between the model's suggestions and human experience judgment. Preferably, this user feedback data includes: the recommended values ​​output by the artificial intelligence model (such as priority suggestions, locking strategy suggestions, risk levels, etc.), the user's adoption status of the recommended values ​​(accept / reject / modify), and the actual configuration values ​​modified by the user.

[0071] Furthermore, training an AI model based on user feedback data is a process of incrementally optimizing model parameters using user feedback data as a supervisory signal. This is used to continuously improve the model's prediction accuracy and decision adaptability in real business scenarios. This process converts user feedback data into labeled training samples, inputs them into the model retraining process, and updates the model weight parameters, thereby achieving a closed-loop evolution of model capabilities.

[0072] In one embodiment, the deviation between the recommended value output by the artificial intelligence model and the user's actual configuration value is recorded, and the deviation is used as a negative sample input into the next round of training of the artificial intelligence model to correct the model bias.

[0073] The user's actual configuration value can be the rule configuration parameters that the user sets and saves after receiving recommendations from the artificial intelligence model and making a manual judgment. These parameters include, but are not limited to: the priority number adopted or modified by the user, the selected priority-related locking strategy type, and the risk assessment level for confirming execution or rejection. This value is automatically captured and persistently stored by the system when the configuration is submitted.

[0074] The aforementioned bias refers to the inconsistency between the recommended value output by the artificial intelligence model and the user's actual configuration value. It is represented by classification difference (e.g., the model recommends P3 while the user configures P1) or enumeration mismatch (e.g., the model recommends "lock parent" while the user selects "lock all"). After the bias is extracted in a structured manner, it together with the corresponding original input features constitutes a negative sample.

[0075] By using biases as negative samples in the next round of training, the model is guided to learn and correct its prediction tendency under similar input conditions by applying higher weights to negative samples in the loss function, thereby reducing the probability of the same biases recurring. This mechanism enables the model to not only learn statistical patterns from historical logs, but also absorb domain expert knowledge from human intervention, significantly enhancing its robustness and credibility in dynamic business environments.

[0076] In another embodiment, during the execution of multiple association rules within the same priority, in response to the configuration instruction of the relationship restriction mode within the priority, a temporary exclusion operation is performed on the parent and / or child data that have successfully established a parent-child relationship, so that they no longer participate in the matching process of subsequent association rules within the current priority, thereby preventing duplicate matching.

[0077] The priority-based relation constraint mode can be a constraint strategy set for the semantic dependency relationship between multiple association rules within the same priority, used to control the reuse boundary of parent and child data within this priority; this mode can be specified by the user when configuring rule groups, or generated by the artificial intelligence recommendation model and then enabled by the user.

[0078] Parent-child relationships can refer to the matching results established between data entities that have a hierarchical dependency relationship in business semantics. The parent data represents the upper-level business object (such as contract, master order), and the child data represents its subordinate detailed objects (such as order line items, invoice details). This relationship is established through field references, primary and foreign key mappings, or business rule logic determination.

[0079] Temporary exclusion is a phased isolation of successfully associated parent or child data within the rule execution sequence of the current priority, making it invisible, uninputtable, and preventing it from triggering new associations during the matching process of the remaining rules in that priority. This operation does not change the state of the data itself, nor does it affect its availability in other priorities or groups.

[0080] This embodiment prevents the same parent data from being repeatedly matched by multiple child rules, or the same child data from being repeatedly bound by multiple parent rules, thereby avoiding business logic conflicts and data redundancy caused by relationship ambiguity. Furthermore, it ensures that parent-child relationships meet preset business consistency constraints within the same priority level, improving the semantic rigor of single group matching results and the consumability of downstream systems.

[0081] The AI-based rule engine data processing method provided in this invention is applied to an enterprise-level automated data flow system. This system includes multi-level priority association rules for processing multi-dimensional business data across systems. When a high-priority rule cannot match some input data, the unmatched data is input into a pre-trained AI model to predict its association success rate in low-priority rules. If the predicted success rate is lower than a set threshold, a pre-lock instruction is generated to exclude the data before executing subsequent rules, avoiding invalid processing. By introducing an AI model for pre-prediction and data filtering, redundant calculations are effectively reduced, and matching accuracy and system resource utilization are improved. This invention solves the problems of low processing efficiency, high error rate, and unreasonable resource allocation in traditional rule engines, achieving the technical effects of shortening the rule configuration cycle and improving data processing reliability and automation.

[0082] For ease of understanding, this embodiment of the invention also provides an application example of an artificial intelligence-based rule engine data processing method (see [link]). Figure 2 This illustration is not for reference only. Figure 1 The data processing method shown is not a replacement or parallel solution for the basic process in the AI-based rule engine, but rather a... Figure 1 The core methods (steps S102 to S108) include a systematic capability extension and structured implementation in complex enterprise-level business scenarios. All its technical features are based on... Figure 1The illustrated embodiment establishes "AI prediction, pre-locking, and dynamic exclusion" as its core operational process. Following the technical logic and execution sequence of this process, it is extended, refined, and systematically implemented across four dimensions: rule configuration layer, task orchestration layer, execution control layer, and human-machine collaboration layer. Together, these constitute an enhanced technical system where each stage supports the others, creates a feedback loop, and complements each other's capabilities. Specifically, it includes the following five interrelated and progressively layered implementation stages: Step S201, AI model training and deployment; As Figure 1 The embodiments shown illustrate the foundation for building and the mechanism for the continuous evolution of pre-trained artificial intelligence models.

[0083] This step corresponds to Figure 1 The source and maintenance path of the pre-trained artificial intelligence model relied upon in steps S102 and S104. A training dataset is constructed based on historical correlation data, matching success rate, business value, and other indicators. , where X i This includes rule characteristics (matching field combinations, data source, data volume, and the business module to which they belong) and historical execution results (success rate). Average time T avg ), Y i Let P be the optimal priority of the rule, L be the optimal locking strategy, and R be the relational restriction pattern within the priority.

[0084] This allows for the training and deployment of a multi-task AI model M. AI ={M P M L M R These components respectively provide priority suggestions, locking strategy recommendations, and relationship constraint pattern predictions. This model can be encapsulated as an API service. Figure 1 In the illustrated embodiment, each AI-driven module (such as the modules corresponding to data input in S102, model prediction in S104, and pre-locking decision in S106) provides underlying capability support and supports retraining based on the latest matching execution logs and user feedback data at a preset cycle (such as weekly). Figure 1 The embodiments shown provide the engineering basis for the online model update and deviation correction mechanism.

[0085] Step S202: AI-assisted rule configuration and strategy recommendation; At the rule configuration layer, when a user initiates an operation to add or modify associated rules, the system immediately calls M. P The model generates priority suggestions (e.g., "AI recommendation priority: P3"), which are then called by M. L The model generates priority-based external association locking strategy suggestions (such as "lock parent-related data") and calls M. RThe model generates priority-based relation constraint pattern suggestions (e.g., "parent-child relationships are all unique"). All recommendations are based on... Figure 1 The input dimensions described in the illustrated embodiment are: matching field combination, filtering strictness, types of data sources involved, expected data volume to be processed, and historical matching success rate.

[0086] The system synchronously records the user's adoption status (accept / reject / modify) of the AI ​​recommendation value and the actual configuration value, constituting the "biased negative sample" defined in the above embodiment, which is used to feed back to M. AI The next round of training will then achieve a complete closed loop of recommendation, configuration, feedback, and correction.

[0087] Step S203: Dynamic task orchestration and AI-optimized execution; This further expands the operational mechanism for generating the actual execution order based on an artificial intelligence scheduling model. Its core lies in determining the actual execution order of each associated rule within the current execution group. It not only relies on the static setting of the first priority associated rule in the aforementioned embodiments but also introduces real-time context, namely, the data volume distribution of the current batch of data to be processed (such as a histogram of the number of data records covered by various field combinations) and the system resource load status (CPU utilization, memory usage, I / O wait queue length), determined by M. AI The scheduling sub-model (such as ListNet) generates a dynamic execution order P. dynamic .

[0088] For example: If the amount of data to be matched corresponding to a high-priority rule R1 is extremely small, then N1 N total The amount of data to be matched for the low-priority rule R2 is extremely large, N2≈N total The AI ​​might suggest executing R1 first (even if R1 has a priority of P1). static R2 corresponds to priority P2 static The goal is to quickly lock in a small amount of data, reduce the data cardinality of R2, and complete the intra-group matching in the shortest possible time.

[0089] Regardless of how the dynamic order is adjusted, the execution logic within a group still strictly follows the strong locking principle shown in step S108 above, that is, data that has been successfully matched does not participate in subsequent rule matching within the same group, ensuring semantic integrity and consistency with the main logic.

[0090] Although there is no locking between groups, AI can optimize its execution timing. AI can predict which group G will execute based on historical execution data. A The execution result of G has the highest dependency on subsequent processes (such as accounting and settlement), so G is scheduled first. A run.

[0091] This AI-driven dynamic scheduling Pdynamic This enables the rule engine to achieve adaptive optimal performance in different operating environments.

[0092] At the task orchestration layer, a new "task orchestration" function has been added, allowing rule-based task grouping. The system incorporates AI to optimize execution logic, dynamically adjusting the order of priority batches within a group based on real-time conditions such as the current volume of data to be matched and peak business periods, ensuring maximum resource utilization. A strong locking mechanism is still used within each group.

[0093] Step S204, real-time prediction and pre-locking mechanism; During rule execution, the system uses AI models to predict unmatched data in real time, assessing its potential success rate and business value in relation to subsequent rules. For data with extremely low success rates, after excluding it from high-priority rule matching, pre-locking can be performed based on AI suggestions, causing it to exit subsequent matching early and further improving efficiency.

[0094] Specifically, during the data flow phase of rule execution, the system introduces an AI real-time prediction module M. pred .

[0095] 1) Real-time prediction of association success rate: When data D unmatch In high-priority rule R high If a match fails, the process is prepared to proceed to the next priority rule R. low When the system calls M pred Model. M pred According to D unmatch Features and R low Features, real-time calculation of D unmatch In R low The probability P of a successful association success .

[0096] 2) Pre-locking strategy: The system is configured with a pre-locking threshold P. threshold (e.g., P) threshold =1%). If P success <P threshold If the AI ​​determines that the data is almost impossible to associate successfully with any of the subsequent low-priority rules, the system can perform pre-locking based on the configuration (the AI ​​suggests exiting the matching process). The data will be marked and prematurely removed from the entire matching process.

[0097] 3) Effect: This AI-driven pre-locking mechanism is a further efficiency optimization than "priority external locking". It not only locks successfully associated data, but also preemptively removes data that is judged by AI as invalid matching attempts, thereby achieving a geometric reduction in computational load in scenarios with huge amounts of data.

[0098] Step S205: Manually performed AI risk assessment and guidance; When a user manually performs a task, the system first prioritizes it. Based on this, an AI model intervenes to assess the potential impact of the manual operation on overall data consistency and historical success rates (AI risk assessment). If the risk is high, the AI ​​will provide more mandatory guidance (such as a high-risk warning) to ensure the stability and accuracy of the execution.

[0099] Specifically, when a user manually triggers the execution of a orchestrated task, the system first performs a mandatory priority check (step four). Based on this, AI-powered decision support is then incorporated. 1) AI Risk Assessment: If the task Tselected selected by the user depends on a higher-priority task Thigh that has not yet been executed, the AI ​​model Mrisk intervenes. Mrisk assesses the proportion of data association errors or duplications that would result if the user skipped Thigh and directly executed Tselected, and the potential impact on downstream operations (such as financial closing). downstream .

[0100] 2) Dynamic interactive prompts: The system provides prompts based on M. risk The evaluation results δ and I downstream Generate dynamic interactive prompts: Low risk / highest priority: A simplified prompt will appear, such as: "Confirm to execute the selected task?" High risk (δ>text10% or I) downstream (Severe): A red warning box pops up, with the message enhanced to: "Risk Warning! AI assessment has detected a higher-priority dependent task for the selected task. Skipping execution may result in data matching errors for delta. Please execute the higher-priority task first. Are you sure you want to continue?"

[0101] 3) Forced guidance: The system only allows further execution after the user explicitly confirms the high-risk warning. This ensures that even manual intervention must be carried out under the premise of AI intelligent assessment and explicit user knowledge, avoiding accidental operation.

[0102] This solution upgrades the static configuration of the rules engine to AI-assisted dynamic optimization, realizing the transformation from human experience-driven to data intelligence-driven.

[0103] Based on the same inventive concept, this embodiment of the invention also provides an artificial intelligence-based rule engine data processing system, which can be used for multi-dimensional business data matching and processing in enterprise-level automated data flow systems. This enterprise-level automated data flow system is configured with multiple association rules divided by priority. See also... Figure 3 As shown, this AI-based rule engine data processing system mainly includes the following components: The data input module 310 is used to input the first data to be processed into the pre-trained artificial intelligence model in response to the completion of the operation of matching the input data based on the first priority association rule and the existence of the first data to be processed that has not been successfully matched. The model prediction module 320 is used to predict the association success rate of the first data to be processed in the second priority association rule based on the first data to be processed using an artificial intelligence model. The pre-locking decision module 330 is used to generate a pre-locking instruction for the first data to be processed if the association success rate is lower than a preset success rate threshold. The process control module 340 is used to exclude the first data to be processed before executing subsequent association rules in the current execution group, based on the pre-locking instruction.

[0104] The AI-based rule engine data processing system provided in this embodiment of the invention can be specific hardware on a device or software or firmware installed on the device. The system provided in this embodiment of the invention has the same implementation principle and technical effects as the foregoing method embodiments. For the sake of brevity, any parts not mentioned in the system embodiments can be referred to the corresponding content in the foregoing method embodiments. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can all be referred to the corresponding processes in the above method embodiments, and will not be repeated here.

[0105] Based on the same inventive concept, embodiments of the present invention also provide an electronic device, specifically, the electronic device includes a processor and a storage device; the storage device stores a computer program, and the computer program, when run by the processor, executes the method described in any of the above embodiments.

[0106] Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. The electronic device 400 includes: a processor 410, a memory 420, a communication interface 430, and a bus 440. The memory 420 stores machine-readable instructions that can be executed by the processor 410. When the electronic device is running, the processor 410 communicates with the memory 420 through the bus 440. The processor 410 executes the machine-readable instructions to perform the steps of the method described above.

[0107] Specifically, the memory 420 and processor 410 can be general-purpose memory and processor, without any specific limitations. When the processor 410 runs the computer program stored in the memory 420, it can execute the above method.

[0108] Processor 410 may be an integrated circuit chip with signal processing capabilities. In implementation, each step of the above method can be completed by the integrated logic circuitry in the hardware of processor 410 or by instructions in software form. The processor 410 may be a general-purpose processor, including a Central Processing Unit (CPU), a Network Processor (NP), etc.; it may also be a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field-Programmable Gate Array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this invention. The general-purpose processor may be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this invention can be directly manifested as execution by a hardware decoding processor, or execution by a combination of hardware and software modules in the decoding processor. The software module can reside in a mature storage medium in the art, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, or registers. This storage medium is located in memory 420, and processor 410 reads the information from memory 420 and, in conjunction with its hardware, completes the steps of the above method.

[0109] Corresponding to the above method, this embodiment of the invention also provides a computer-readable storage medium storing machine-executable instructions. When the computer-executable instructions are called and run by a processor, the computer-executable instructions cause the processor to perform the steps of the above method.

[0110] In the embodiments provided by this invention, it should be understood that the disclosed apparatus and method can be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. Furthermore, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Additionally, the displayed or discussed mutual couplings, direct couplings, or communication connections may be through some communication interfaces; indirect couplings or communication connections between devices or units may be electrical, mechanical, or other forms.

[0111] Furthermore, 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.

[0112] Furthermore, the functional modules in the various embodiments of the present invention can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0113] It should be noted that if the functionality is implemented as a software module 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 described in 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.

[0114] In this document, relational terms such as first and second are used only to distinguish one entity or operation from another entity or operation, without necessarily requiring or implying any such actual relationship or order between these entities or operations.

[0115] The above description is merely an embodiment of the present invention and is not intended to limit the scope of protection of the present invention. For those skilled in the art, the present invention can have various modifications and variations. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A data processing method for a rule engine based on artificial intelligence, characterized in that, The method is applied to an enterprise-level automated data flow system, which is configured with multiple priority-based association rules for matching and processing multi-dimensional business data across systems; the method includes: In response to the completion of the operation of matching the input data based on the first priority association rule, and the existence of a first unmatched pending data, the first pending data is input into the pre-trained artificial intelligence model; Based on the first data to be processed, the artificial intelligence model is used to make a prediction and generate the association success rate of the first data to be processed in the second priority association rule. If the association success rate is lower than a preset success rate threshold, a pre-locking instruction is generated for the first data to be processed; Based on the pre-locking instruction, the first data to be processed is excluded before the subsequent association rules in the current execution group are to be executed.

2. The method according to claim 1, characterized in that, The method further includes: When determining the actual execution order of each associated rule within the current execution group, the actual execution order is generated by a pre-deployed artificial intelligence scheduling model based on the data volume distribution of the current batch of pending data and the system resource load status.

3. The method according to claim 1, characterized in that, The method further includes: In response to receiving a manual execution request from a user, the associated rules corresponding to the selected task are checked, and it is determined whether there are any associated rules with a higher execution order but which have not yet been executed. If present, the rule characteristics of the association rule corresponding to the selected task, the configuration information of the higher priority association rule that has not been executed, and the log data of the historical matching success rate will be input into the pre-trained artificial intelligence risk assessment model.

4. The method according to claim 3, characterized in that, The method further includes: Based on the output of the AI ​​risk assessment model, a risk assessment report is generated that includes the predicted data error rate and the impact level on downstream business. Based on the downstream business impact level in the risk assessment report, generate corresponding interactive prompt information. When the impact level of the downstream business reaches the severe level, a mandatory confirmation mechanism is activated to prohibit the continued execution of the selected task without receiving explicit confirmation from the user.

5. The method according to claim 1, characterized in that, The method further includes: During the association rule configuration phase, in response to the operation of adding or modifying association rules, a pre-trained artificial intelligence recommendation model is invoked to generate priority suggestions and non-priority association locking strategy suggestions for the rule. The priority recommendation is generated based on the combination of matching fields of the rules, the strictness of filtering, the types of data sources involved, the expected amount of data to be processed, and the historical matching success rate; the priority-external association locking strategy recommendation includes recommending locking the parent-level associated data, locking the child-level associated data, or locking all associated data, at least one of these options.

6. The method according to claim 1, characterized in that, The method further includes: The system periodically receives matching execution logs and user feedback data within a preset period, and trains the artificial intelligence model based on the user feedback data to update the parameters of the artificial intelligence model.

7. The method according to claim 6, characterized in that, The method further includes: The deviation between the recommended value output by the artificial intelligence model and the user's actual configuration value is recorded, and the deviation is used as a negative sample to be input into the next round of training of the artificial intelligence model in order to correct the model deviation.

8. A rule engine data processing system based on artificial intelligence, characterized in that, This system is used for multi-dimensional business data matching and processing in an enterprise-level automated data flow system, which is configured with multiple priority-based association rules. The AI-based rule engine data processing system includes: The data input module is used to respond to the completion of the operation of matching input data based on the first priority association rule, and the existence of first unmatched data to be processed, and input the first data to be processed into the pre-trained artificial intelligence model. The model prediction module is used to make predictions based on the first data to be processed using the artificial intelligence model, and generate the association success rate of the first data to be processed in the second priority association rule. The pre-locking decision module is used to generate a pre-locking instruction for the first data to be processed if the association success rate is lower than a preset success rate threshold. The process control module is used to exclude the first data to be processed before executing subsequent association rules in the current execution group, based on the pre-locking instruction.

9. An electronic device comprising a memory and a processor, wherein the memory stores a computer program executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions that, when invoked and executed by a processor, cause the processor to perform the method according to any one of claims 1 to 7.

Images