Intelligent anti-repeated submission method and system based on multi-feature analysis and decision feedback

Abstract

The invention discloses an intelligent anti-repeated submission method and system based on multi-feature analysis and decision feedback, and relates to the technical field of file processing, and the method comprises the steps: generating a real-time multi-dimensional feature vector according to the context feature of an accessed service request; performing accurate matching and fuzzy matching on the real-time multi-dimensional feature vector through a decision engine to obtain a matching result; dynamically selecting a service submission strategy according to a matching result and a system real-time state; executing the dynamically selected service submission strategy to obtain an execution result and a final service state; and performing feedback optimization on rule parameters of the multi-level rule decision engine according to the execution result and the final business state. By introducing spatio-temporal context feature extraction, a multi-level rule engine and a strategy feedback closed loop, normal form conversion from passive prevention to active cooperative processing is realized, the processing efficiency, the operation stability and the automation level of the system are fundamentally improved, and the problems of blind areas, static stiffness and incapability of predicting repeated request flood peaks are solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of file processing, in particular to an intelligent anti-duplicate submission method and system based on multi-feature analysis and decision feedback. BACKGROUND

[0002] In an online collaborative research and development platform, preventing duplicate submission is a key technical point to ensure data integrity. The existing technology has experienced evolution from single machine lock to distributed Token, and then to configurable business rules. However, the existing technology generally has a "technical bias": it is believed that anti-duplicate submission is a purely and business-independent technical problem, and the solution should focus on the idempotency of the request itself, and avoid involving business logic, because the introduction of business logic will lead to complex, bloated and difficult-to-universal solutions.

[0003] In a distributed system, especially in an online collaboration platform, the existing anti-duplicate submission technology (such as Token mechanism, serial number mechanism) mainly adopts a static and passive response strategy, which has obvious limitations and is difficult to effectively deal with different users submitting requests with highly similar business intentions, or the same user submitting requests with the same intention in different ways. The existing technology is difficult to deal with the macro-predictable duplicate request flood caused by the periodicity of the business (such as concentrated declaration at the end of the quarter). It is also unable to make dynamic decision adjustments according to real-time system load, business risk level and other factors, and lacks flexibility. It is unable to automatically feedback and optimize according to historical processing results, and has high operation and maintenance cost and slow response. SUMMARY

[0004] The embodiments of the present application provide an intelligent anti-duplicate submission method and system based on multi-feature analysis and decision feedback, to solve the problems that the existing technology is difficult to effectively deal with different users submitting requests with highly similar business intentions, or the same user submitting requests with the same intention in different ways, the macro-predictable duplicate request flood caused by the periodicity of the business (such as concentrated declaration at the end of the quarter), the lack of flexibility in dynamic decision adjustment according to real-time system load, business risk level and other factors, and the inability to automatically feedback and optimize according to historical processing results, high operation and maintenance cost and slow response.

[0005] In one aspect, the embodiments of the present application provide an intelligent anti-duplicate submission method based on multi-feature analysis and decision feedback, comprising: generating a real-time multi-dimensional feature vector according to the context features of the accessed business request; precisely matching and fuzzy matching the real-time multi-dimensional feature vector through the precise matching layer and the fuzzy matching and frequency analysis layer of the multi-level rule decision engine to obtain a matching result; dynamically selecting a business submission strategy according to the matching result and the real-time state of the system; The execution result and the final state of the business are obtained by executing the dynamically selected business submission strategy. The rule parameters of the multi-level rule decision engine are optimized based on the execution results and the final business state.

[0006] In one possible implementation, the real-time multidimensional feature vector is represented as: V_request=Concat(T,S,C,Sys); The real-time multidimensional feature vector V_request includes time dimension feature T, spatial dimension feature S, business content feature C, and system state feature Sys, where Concat is the concatenation operation of feature vectors.

[0007] In one possible implementation, the time dimension feature T includes the timestamp of the request, the business cycle, and the time interval since the last similar request; The spatial dimension feature S includes user ID, department ID, geographic location information, IP address range, and access link node; The business content feature C includes key business fields; The system state characteristics Sys include the current system load, target database pressure, and message queue backlog depth.

[0008] In one possible implementation, the precise matching layer of the multi-level rule decision engine performs matching based on preset strong interception rules or precise fingerprints of request content. The fuzzy matching and frequency analysis layer of the multi-level rule decision engine calculates the similarity and frequency of current business requests and historical business requests in the feature dimension based on a configurable threshold rule set. The fuzzy matching and frequency analysis layer is used to compare the similarity and frequency with a dynamic threshold.

[0009] In one possible implementation, the dynamic selection of the business submission strategy includes direct rejection of requests, manual review, challenge verification, and delayed processing queues.

[0010] In one possible implementation, the delayed processing queue corresponds to a non-high-risk matching result, and the delayed processing queue satisfies the system overload condition; The system overload conditions include: The CPU utilization rate in the system state feature Sys continuously exceeds the first preset threshold, or the rate of change of the CPU utilization rate in the system state feature Sys within a unit time exceeds the fourth preset threshold, the database connection occupancy rate exceeds the second preset threshold, or the growth rate of the database connection number exceeds the fifth preset threshold, or the number of messages accumulated in the message queue exceeds the third preset threshold, or one or more of the following:

[0011] In one possible implementation, the direct rejection request corresponds to an exact match or a high-frequency rule; The manual review and challenge verification correspond to medium-risk matching in fuzzy matching.

[0012] On the other hand, embodiments of the present invention also provide an intelligent anti-duplicate submission system based on multi-feature analysis and decision feedback, including: The feature extraction module is used to generate real-time multidimensional feature vectors based on the contextual features of the incoming business requests; The feature matching module is used to perform precise matching and fuzzy matching on the real-time multidimensional feature vector through the precise matching layer and fuzzy matching and frequency analysis layer of the multi-level rule decision engine to obtain the matching result. The dynamic selection module for submission strategy is used to dynamically select the business submission strategy based on the matching results and the real-time status of the system. The engine optimization module is used to execute the dynamically selected business submission strategy to obtain the execution result and the final business state; and to optimize the rule parameters of the multi-level rule decision engine based on the execution result and the final business state.

[0013] The intelligent anti-duplicate submission method and system based on multi-feature analysis and decision feedback in this invention have the following advantages: By introducing spatiotemporal context feature extraction, multi-level rule engines, and policy feedback loops, a paradigm shift from passive prevention to proactive collaborative processing is achieved, fundamentally improving the system's processing efficiency, operational stability, and automation level, and solving problems such as blind spots, static rigidity, and unpredictable surges of duplicate requests. Attached Figure Description

[0014] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0015] Figure 1 A flowchart of an intelligent anti-duplicate submission method based on multi-feature analysis and decision feedback provided in an embodiment of this application; Figure 2 This is a schematic diagram of an intelligent anti-duplicate submission system based on multi-feature analysis and decision feedback provided in an embodiment of this application. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and 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.

[0017] Figure 1 A flowchart illustrating an intelligent method for preventing duplicate submissions based on multi-feature analysis and decision feedback provided in an embodiment of the present invention; the embodiment of the present invention provides an intelligent method for preventing duplicate submissions based on multi-feature analysis and decision feedback, including: Generate real-time multidimensional feature vectors based on the contextual characteristics of the accessed business requests; The matching results are obtained by performing precise matching and fuzzy matching on the real-time multidimensional feature vector through the precise matching layer and the fuzzy matching and frequency analysis layer of the multi-level rule decision engine. Dynamically select the business submission strategy based on the matching results and the real-time system status; The execution result and the final state of the business are obtained by executing the dynamically selected business submission strategy. The rule parameters of the multi-level rule decision engine are optimized based on the execution results and the final business state.

[0018] The real-time multidimensional feature vector is represented as follows: V_request=Concat(T,S,C,Sys); The real-time multidimensional feature vector V_request includes time dimension feature T, spatial dimension feature S, business content feature C, and system state feature Sys, where Concat is the concatenation operation of feature vectors.

[0019] The time dimension feature T includes the timestamp of the request, the business cycle, and the time interval since the last similar request; The spatial dimension feature S includes user ID, department ID, geographic location information, IP address range, and access link node; The business content feature C includes key business fields; The system state characteristics Sys include the current system load, target database pressure, and message queue backlog depth.

[0020] The precise matching layer of the multi-level rule decision engine performs matching based on preset strong interception rules or precise fingerprints of request content; The fuzzy matching and frequency analysis layer of the multi-level rule decision engine calculates the similarity and frequency of current business requests and historical business requests in the feature dimension based on a configurable threshold rule set. The fuzzy matching and frequency analysis layer is used to compare the similarity and frequency with a dynamic threshold.

[0021] The dynamic selection of the business submission strategy includes direct rejection of requests, manual review, challenge verification, and delayed processing queue.

[0022] The delayed processing queue corresponds to non-high-risk matching results, and the delayed processing queue satisfies the system overload condition; The system overload conditions include: The CPU utilization rate in the system state feature Sys continuously exceeds the first preset threshold, or the rate of change of the CPU utilization rate in the system state feature Sys within a unit time exceeds the fourth preset threshold, the database connection occupancy rate exceeds the second preset threshold, or the growth rate of the database connection number exceeds the fifth preset threshold, or the number of messages accumulated in the message queue exceeds the third preset threshold, or one or more of the following:

[0023] The direct rejection request corresponds to an exact match or a high-frequency rule; The manual review and challenge verification correspond to medium-risk matching in fuzzy matching.

[0024] For example, S101: Generate real-time multidimensional feature vectors.

[0025] When the system receives a new business request (e.g., an HTTP API request for file submission or data saving), the feature extraction module is immediately activated. It extracts and fuses multi-dimensional features from the request and its context to generate a structured real-time multi-dimensional feature vector V_request.

[0026] Specifically, V_request uses the concatenation function Concat to fuse the following four types of feature vectors: ①Time dimension feature T: includes the precise timestamp of the request arrival (millisecond level), the system's predefined business cycle identifier (such as "end of quarter", "last three days of the month"), and the time interval (Δt) since the user last submitted a similar request.

[0027] ② Spatial dimension feature S: includes the unique identifier of the user initiating the request (UserID), the identifier of the organization department (DeptID), the client IP address, the geographical location or internal network segment to which the IP address belongs, and the identifier of the gateway or service node through which the request passes.

[0028] ③ Business Content Features C: Extract key business fields from the request payload. For example, for a project application request, extract fields such as "Project Number," "Project Name," "Applicant Name," and "Application Amount." After standardizing and cleaning these fields (e.g., removing spaces and unifying case), a fixed-length content fingerprint is generated using a hash algorithm (e.g., MD5, SHA-256) or a semantic fingerprint algorithm (e.g., SimHash). This fingerprint represents the core business semantics of the request.

[0029] ④ System Status Characteristics Sys: Real-time collected system health indicators, including but not limited to: average CPU utilization of application server cluster (CPU_Load), connection pool utilization of core database (DB_Conn_Ratio), and number of messages to be processed in critical business message queue (MQ_Backlog).

[0030] Finally, V_request = Concat(T,S,C,Sys). This vector serves as the input basis for all subsequent intelligent decisions.

[0031] S102: Intelligent matching is performed through the decision engine to obtain matching results.

[0032] The decision engine receives V_request and internally employs a multi-level collaborative decision-making architecture. It executes the following matching process sequentially and generates a structured matching result object.

[0033] ①Exact Matching Layer: Content fingerprint comparison: The content fingerprint C in V_request is compared with a "global request fingerprint database" stored in a distributed cache (such as Redis). This fingerprint database has a time-to-live (TTL), for example, storing the fingerprints of all successfully processed requests within the last 24 hours. If an exact match fingerprint is found, it is immediately determined to be an "exact duplicate," and the matching result is marked as EXACT_MATCH (high risk).

[0034] ② Static strong rule interception: Parallel checks are performed on the S features (such as UserID, IP) in V_request to see if they match preset static rules. For example, a match may be made with an "IP blacklist" or trigger a "forced secondary verification for sensitive operations" rule. If a match is found, the result is marked as RULE_BLOCK (high risk).

[0035] Fuzzy matching and frequency analysis layer (core): If the request is not intercepted at the exact matching layer, it proceeds to this layer for in-depth analysis. This layer relies on a set of configurable, dynamic threshold rules.

[0036] Similarity calculation: For requests that do not match precisely, the similarity between their business content feature C and recent historical requests is calculated. The system maintains a queue of historical request content for each user or department.

[0037] In implementation, the SimHash algorithm can be used. This algorithm generates a fixed-bit (e.g., 64-bit) fingerprint for text, and the similarity of the content is determined by calculating the Hamming distance between two fingerprints. The smaller the Hamming distance, the higher the similarity. For example, a Hamming distance less than 3 can be considered highly similar.

[0038] Alternatively, cosine similarity can be used. First, the text keywords are vectorized, and then the cosine value of the angle between the vectors is calculated.

[0039] The SimHash algorithm first standardizes and cleans the extracted key business fields, merging them into a structured text string. Then, it segments the text string and assigns weights to each keyword based on the importance of the business fields. Next, it performs a hash mapping on each keyword, generating a fixed-length binary hash value. Then, it initializes a fixed-dimensional zero vector and performs weighted addition and subtraction operations on the corresponding bits of the vector based on each bit of the hash value. Afterward, it binarizes the processed vector to obtain a fixed-length SimHash fingerprint. Finally, it calculates the Hamming distance between the current request fingerprint and historical request fingerprints, and uses the Hamming distance to calculate the similarity; the smaller the Hamming distance, the higher the similarity.

[0040] Cosine similarity calculation first involves building a global keyword dictionary based on key fields from historical business requests, which needs to be updated regularly. Then, keywords are extracted from the current request's business content text, and their frequencies are calculated. The TF-IDF value (term frequency - inverse document frequency) for each keyword is then calculated using the dictionary: first, the frequency of the keyword in the current text is calculated (term frequency TF), then the prevalence of the keyword across all texts is calculated (inverse document frequency IDF; the more prevalent the keyword, the lower the IDF value), and finally, the two are multiplied to obtain the TF-IDF value. Based on the TF-IDF value, a text vector with the same dimensions as the dictionary is generated; historical request texts are also generated in the same way. Finally, the cosine value is calculated as the ratio of the vector dot product to the vector magnitude; this cosine value represents the similarity between the two.

[0041] Frequency statistics: Utilizing T and S characteristics, multiple sliding time window counters are maintained in the distributed cache. For example: Counter_user_op: The key is "User ID: Operation Type", which counts the number of times a user performs a specific operation in the last 5 minutes.

[0042] Counter_dept_keyword: The key is "Department ID: Keyword", which counts the number of requests containing a certain keyword submitted by this department in the last 10 minutes (the specific value is only an example parameter used to illustrate the judgment logic).

[0043] Threshold rules and risk assessment: The decision engine loads pre-configured rules. Each rule defines matching conditions, a time window, statistical dimensions, and a dynamic threshold. For example: Rule A: IF (same user && content similarity > θ_sim_A) THEN Risk Level = Medium Risk. θ_sim_A is the similarity threshold, with an initial value of 0.8.

[0044] Rule B: IF(same department && "submit" operation frequency > θ_freq_B) THEN Risk Level = High Risk. θ_freq_B is the frequency threshold, with an initial value of 50 times / 10 minutes.

[0045] The engine compares the calculated similarity value or frequency statistics with the threshold of the corresponding rule. Based on the comparison result, the matching result may be marked as FUZZY_HIGH_RISK (high risk), FUZZY_MID_RISK (medium risk), or LOW_RISK (low risk).

[0046] Distributed collaboration: During cluster deployment, all service nodes write a feature digest of the request (such as user ID: content fingerprint) to a shared cache. When making decisions, the system queries whether other nodes have recently processed highly similar requests to identify cross-node distributed duplicate submissions.

[0047] ③ Output matching result object: This object contains key fields: matchType (matching type enumeration), riskLevel (risk level), hitRuleId (hit rule ID), similarityScore (similarity score), frequency (statistical frequency), etc.

[0048] S103: Dynamically select a processing strategy based on the matching results and system status.

[0049] The system combines the matching results output in step S102 with the Sys features acquired in real time to execute a strategy selection logic: Strategy 1: Direct rejection. When matchType is EXACT_MATCH, RULE_BLOCK, or FUZZY_HIGH_RISK, unless the system is in an extremely idle state (determined by the strategy configuration), a "Request suspected of being duplicated, rejected" response will be returned directly to the client.

[0050] Strategy Two: Transfer to manual review or challenge verification. When matchType is FUZZY_MID_RISK, the system will not reject the request directly, but will suspend the request and trigger a secondary process. This could be: The request details will be pushed to the administrator console for manual review.

[0051] Return a CAPTCHA (image verification code) or SMS verification code to the user's client, requiring the user to complete the verification to continue.

[0052] Strategy 3: Delayed Processing. This strategy is executed when matchType is LOW_RISK or FUZZY_MID_RISK, but the system status characteristic Sys indicates system overload. The specific criterion for determining system overload is a configurable composite condition, for example: (CPU_Load>85% for 30 seconds) OR (DB_Conn_Ratio>90%) OR (MQ_Backlog>10000).

[0053] When any overload condition is met, the system places the request into a priority delayed queue (such as a RabbitMQ delayed queue) and informs the user that "the request has been submitted and is being queued for processing." Once the system load decreases, the requests in the queue are processed in order. This strategy is used for "peak shaving and valley filling" to ensure overall system stability.

[0054] Strategy 4: Allow access directly. When matchType is LOW_RISK and the system status is normal, the request is allowed to proceed directly to the backend business system for processing.

[0055] S104: Optimize the decision engine based on execution feedback.

[0056] The system establishes a continuous feedback optimization loop: ① Data Collection: Asynchronously record the "audit log" for each request, including its V_request, matching results, execution strategy, and most importantly—the final status of the request in the backend business system. The final status is obtained by listening to database operation results, business logs, or manual review interfaces, with tags such as: "SUCCESS", "DB_DUPLICATE_KEY_ERROR", and "MANUAL_REJECT".

[0057] The final status is obtained by monitoring the transaction commit logs of the business database, capturing specific business error codes, or receiving confirmation signals from the manual review module, to ensure the authenticity and timeliness of the feedback data.

[0058] For example, by parsing the SQLSTATE error code 23000 returned by the database (indicating an integrity constraint violation), it can be automatically marked as DB_DUPLICATE_KEY_ERROR.

[0059] ②Effect Evaluation: The optimization module analyzes audit logs periodically (e.g., hourly) and generates visual reports. Key metrics include: Precision and recall of each rule: For example, if rule B is triggered 100 times and the final state is a conflict in 95 of them, then its precision is high; if there are still 50 conflicts that are not captured by rule B, then its recall needs to be improved.

[0060] Strategy effectiveness analysis: such as "the percentage of requests that were delayed but were eventually successfully processed", to evaluate the impact of the delay strategy on user experience.

[0061] ③ Parameter tuning: Manual tuning: Administrators can dynamically adjust rule parameters in the decision engine based on reports via the web management interface. For example, if rule A is found to be causing a lot of false blocking, the similarity threshold θ_sim_A can be increased from 0.8 to 0.85.

[0062] Automatic optimization (optional advanced function): The system can set optimization goals (such as maximizing overall accuracy), utilize historical data, and automatically find a better set of dynamic threshold parameters through algorithms (such as grid search, Bayesian optimization), which will take effect after administrator confirmation.

[0063] The automatic optimization algorithm first clarifies the optimization objective, sets the rule parameters to be optimized, and their boundaries. Then, it extracts a sample set containing features, rule parameters, and the final business state from audit logs. During initialization, it randomly samples some parameters for training to obtain initial objective function values. Next, it constructs a proxy model, fitting a mapping relationship between the sampled parameters and the objective function values ​​to predict the objective function values ​​and uncertainties of unsampled parameters. Then, it selects the next set of parameters to be tested according to a preset sampling strategy, executes the test, and records the objective function values. This sampling and testing process is repeated until the iteration termination condition is met, outputting the optimal parameter combination. Finally, the optimal parameters are pushed to the configuration center to update the rule parameters, and the effect of the new parameters is verified through traffic splitting. Based on the verification results, it is determined whether the parameters are effective.

[0064] ④ Model Iteration (if a machine learning model is integrated): If the decision engine integrates a classification model such as a random forest, the feedback optimization module will periodically add new "feature-result" data to the training set to incrementally update the model, adapting it to changes in the distribution of business data. The statistical analysis model or machine learning model serves as a rule calculation or auxiliary evaluation method in the rule decision engine, and its output is uniformly decided by the rule decision engine.

[0065] Thus, the "dynamic" characteristic of the dynamic threshold described in this application is fully realized, specifically in that: the initial threshold value can be flexibly configured according to the business scenario, and during system operation, it can continuously adjust automatically or manually based on real business result data through the aforementioned feedback optimization loop. This mechanism enables the system to break free from the rigidity of static rules, possessing adaptive learning capabilities that perceive the environment and self-optimize, thereby enabling it to respond accurately and consistently to the evolution of repetitive business submission patterns over the long term.

[0066] Through the closed-loop execution of the above four steps S101-S104, intelligent anti-duplicate submission is achieved, moving from passive identification to proactive perception, from static rules to dynamic decision-making, and from fixed strategies to continuous optimization.

[0067] Figure 2 A schematic diagram of an intelligent anti-duplicate submission system based on multi-feature analysis and decision feedback provided in this application embodiment; this application embodiment provides an intelligent anti-duplicate submission system based on multi-feature analysis and decision feedback, including: The feature extraction module is used to generate real-time multidimensional feature vectors based on the contextual features of the incoming business requests; The feature matching module is used to perform precise matching and fuzzy matching on the real-time multidimensional feature vector through the precise matching layer and fuzzy matching and frequency analysis layer of the multi-level rule decision engine to obtain the matching result. The dynamic selection module for submission strategy is used to dynamically select the business submission strategy based on the matching results and the real-time status of the system. The engine optimization module is used to execute the dynamically selected business submission strategy to obtain the execution result and the final business state; and to optimize the rule parameters of the multi-level rule decision engine based on the execution result and the final business state.

[0068] For example, such as Figure 2 As shown, ① Feature extraction module: corresponding to method step S101. This module is deployed in the API gateway or request interceptor and is responsible for capturing the request stream in real time, extracting and fusing four-dimensional features of time, space, content, and system state to generate the V_request vector.

[0069] ② Feature Matching Module (Decision Engine): Corresponding to method step S102. This module is an independent service or library that encapsulates the logic of the precise matching layer, fuzzy matching layer, and frequency analysis layer. It receives feature vectors, accesses the distributed cache for comparison and statistics, and outputs structured matching results. The rule set and threshold parameters of this module can be dynamically updated through the configuration center.

[0070] ③ Dynamic Strategy Selection Module: Corresponds to method step S103. This module executes strategy selection logic based on the matching results and real-time system monitoring data (Sys), and calls the corresponding processor to perform operations such as rejection, transfer to manual intervention, initiating a challenge, or submitting to the delay queue.

[0071] ④ Engine Optimization Module: Corresponds to method step S104. This module is a background service responsible for collecting audit logs, calculating and analyzing indicators, generating optimization reports, and providing a configuration interface for administrators to adjust rule parameters or automatically execute parameter tuning tasks.

[0072] In addition to the above, supporting components include distributed caches (such as Redis, a high-performance key-value distributed in-memory database often used for caching, message queues, etc.) for storing fingerprint databases, counters, and shared digests; message queues (such as RabbitMQ / Kafka, two mainstream message queue middleware. RabbitMQ focuses on high reliability and supports multiple protocols; Kafka focuses on high throughput and is suitable for big data scenarios) for implementing latency processing and decoupling; and a monitoring system for providing real-time Sys feature data.

[0073] The modules mentioned above can be deployed on the same server or in a distributed cluster and communicate and collaborate through an internal network.

[0074] It should be noted that the specific values ​​mentioned in the embodiments (such as a threshold of 85% and a time window of 10 minutes) are merely examples and can be adjusted according to specific business scenarios and system performance in actual applications. The third-party components involved (such as Redis and the SimHash algorithm) are all well-known technologies in the field, and their specific implementation details are not the focus of this application. The focus is on how these components are organically integrated and applied to the specific problem of "preventing duplicate submissions" to form a complete intelligent closed-loop system with perception, decision-making, execution, and feedback capabilities.

[0075] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of this invention.

[0076] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. An intelligent method for preventing duplicate submissions based on multi-feature analysis and decision feedback, characterized in that: include Generate real-time multidimensional feature vectors based on the contextual characteristics of the accessed business requests; The matching results are obtained by performing precise matching and fuzzy matching on the real-time multidimensional feature vector through the precise matching layer and the fuzzy matching and frequency analysis layer of the multi-level rule decision engine. Dynamically select the business submission strategy based on the matching results and the real-time system status; The execution result and the final state of the business are obtained by executing the dynamically selected business submission strategy. The rule parameters of the multi-level rule decision engine are optimized based on the execution results and the final business state.

2. The intelligent anti-duplicate submission method based on multi-feature analysis and decision feedback according to claim 1, characterized in that, The real-time multidimensional feature vector is represented as follows: V_request=Concat(T,S,C,Sys); The real-time multidimensional feature vector V_request includes time dimension feature T, spatial dimension feature S, business content feature C, and system state feature Sys, where Concat is the concatenation operation of feature vectors.

3. The intelligent anti-duplicate submission method based on multi-feature analysis and decision feedback according to claim 2, characterized in that, The time dimension feature T includes the timestamp of the request, the business cycle, and the time interval since the last similar request; The spatial dimension feature S includes user ID, department ID, geographic location information, IP address range, and access link node; The business content feature C includes key business fields; The system state characteristics Sys include the current system load, target database pressure, and message queue backlog depth.

4. The intelligent anti-duplicate submission method based on multi-feature analysis and decision feedback according to claim 1, characterized in that, The precise matching layer of the multi-level rule decision engine performs matching based on preset strong interception rules or precise fingerprints of request content; The fuzzy matching and frequency analysis layer of the multi-level rule decision engine calculates the similarity and frequency of current business requests and historical business requests in the feature dimension based on a configurable threshold rule set. The fuzzy matching and frequency analysis layer is used to compare the similarity and frequency with a dynamic threshold.

5. The intelligent anti-duplicate submission method based on multi-feature analysis and decision feedback according to claim 2, characterized in that, The dynamic selection of the business submission strategy includes direct rejection of requests, manual review, challenge verification, and delayed processing queue.

6. The intelligent anti-duplicate submission method based on multi-feature analysis and decision feedback according to claim 5, characterized in that, The delayed processing queue corresponds to non-high-risk matching results, and the delayed processing queue satisfies the system overload condition; The system overload conditions include: The CPU utilization rate in the system state feature Sys continuously exceeds the first preset threshold, or the rate of change of the CPU utilization rate in the system state feature Sys within a unit time exceeds the fourth preset threshold, the database connection occupancy rate exceeds the second preset threshold, or the growth rate of the database connection number exceeds the fifth preset threshold, or the number of messages accumulated in the message queue exceeds the third preset threshold, or one or more of the following:

7. The intelligent anti-duplicate submission method based on multi-feature analysis and decision feedback according to claim 5, characterized in that, The direct rejection request corresponds to an exact match or a high-frequency rule; The manual review and challenge verification correspond to medium-risk matching in fuzzy matching.

8. An intelligent anti-duplicate submission system based on multi-feature analysis and decision feedback, the system being used to implement the method described in any one of claims 1-7, characterized in that, include: The feature extraction module is used to generate real-time multidimensional feature vectors based on the contextual features of the incoming business requests; The feature matching module is used to perform precise matching and fuzzy matching on the real-time multidimensional feature vector through the precise matching layer and fuzzy matching and frequency analysis layer of the multi-level rule decision engine to obtain the matching result. The dynamic selection module for submission strategy is used to dynamically select the business submission strategy based on the matching results and the real-time status of the system. The engine optimization module is used to execute the dynamically selected business submission strategy to obtain the execution result and the final business state; The rule parameters of the multi-level rule decision engine are optimized based on the execution results and the final business state.

Images