Automatic grading and shunting method and system for supervision clues

Abstract

The invention relates to the technical field of supervision, and particularly provides an automatic grading and shunting method and system for supervision clues, and the method comprises the following steps: S1, establishing a multi-modal data fusion model; s2, processing information driven by a large model; and S3, setting a dynamic rule engine architecture. Compared with the prior art, the method can improve the efficiency, the standardization degree and the adaptability of monitoring clue processing.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of supervision, and specifically provides a supervision clue automatic grading and shunting method and system. BACKGROUND

[0002] The current supervision clue processing has the following problems: 1. Strong dependence on manual work: manual reading of text, pictures, voice and other multi-modal clues is low in efficiency and easy to miss key information; 2. Low utilization rate of historical case data: unstructured historical records (such as past case texts) need to be processed manually before they can be used for query and retrieval, and the data utilization cost of historical cases is high; 3. Inconsistent rule execution: grading standards depend on personal experience, and the same clue may be determined as different levels by different personnel; 4. Poor dynamic adaptability: traditional rule engines rely on static rule libraries and cannot optimize decision logic based on historical processing results. SUMMARY

[0003] The present application is aimed at the deficiencies of the prior art and provides a supervision clue automatic grading and shunting method with strong practicality.

[0004] The further technical task of the present application is to provide a supervision clue automatic grading and shunting system with reasonable design and safety.

[0005] The technical solution adopted by the present application to solve its technical problems is: A supervision clue automatic grading and shunting method, comprising the following steps: S1, a multi-modal data fusion model; S2, information processing driven by a large model; S3, setting up a dynamic rule engine architecture.

[0006] Further, in step S1, it includes: S1-1, text clue processing, based on a domain keyword recognition model of a special knowledge base, extracting reflection objects, duties, amounts and misconducts; S1-2, using OCR to extract text and speech-to-text for image and voice clue processing.

[0007] Further, in step S2, it includes: S2-1, converting the clue information to be judged into structured data through a pre-trained large model, and optimizing the information input to the rule engine; S2-2, automatically parsing and generating structured fields for the existing historical clue processing information through a large model, and correcting them through a verification module for dynamic rule weight training.

[0008] Further, in step S3, comprising: S3-1, static rule layer: mandatory requirements of regulations; S3-2, dynamic rule layer: structured historical data generated based on large model, training logistic regression model, generating rule weight; S3-3, conflict resolution strategy: priority sorting + weighted total score determination.

[0009] A kind of supervision clue automatic grading and shunting method, first, carry out the construction multi-modal data fusion model, carry out large model driven information processing, finally, set dynamic rule engine architecture.

[0010] Further, when constructing multi-modal data fusion model, comprising: (1), text clue processing, based on the domain keyword identification model of special knowledge base, extract reflection object, post, amount and misconduct behavior; (2), using OCR to extract text and speech-to-text to process image, voice clues.

[0011] Further, when carrying out large model driven information processing, comprising: (1), the clue information needing discrimination is converted into structured data by pre-training large model, optimization information input to rule engine; (2), the existing historical clue processing information is automatically parsed and structured field is generated by large model, and after being corrected by verification module, it is used for dynamic rule weight training.

[0012] Further, when setting dynamic rule engine architecture, comprising: (1), static rule layer: mandatory requirements of regulations; (2), dynamic rule layer: structured historical data generated based on large model, training logistic regression model, generating rule weight; (3), conflict resolution strategy: priority sorting + weighted total score determination.

[0013] Compared with prior art, the supervision clue automatic grading and shunting method and system of the present application have the following outstanding beneficial effects: The clue processing time of the present application is shortened from an average of 2 hours per piece to 5 minutes per piece, the historical data processing efficiency is improved by 90% (compared with manual labeling), the consistency rate of grading results and manual review is 92% (the traditional method is only 65%), the historical data is automatically processed by large model, 80% of rule maintenance cost is saved, the system automatically generates grading report, records triggered rules and weight calculation logic. BRIEF DESCRIPTION OF DRAWINGS

[0014] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings described below are only some embodiments of the present application, and all other drawings obtained by those of ordinary skill in the art without creative work based on these drawings are within the protection scope of the present application.

[0015] Figure 1 It is a flowchart of a supervision clue automatic grading and shunting method. DETAILED DESCRIPTION

[0016] In order to make the person in the art better understand the scheme of the present application, the present application will be further described in detail below in combination with specific embodiments. Obviously, the described embodiments are only some embodiments of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work belong to the protection scope of the present application.

[0017] The following gives a best embodiment: As Figure 1 shown, the supervision clue automatic grading and shunting method in the embodiment has the following steps: S1, a multi-modal data fusion model; including: S1-1, text clue processing, based on a domain keyword recognition model of a special knowledge base, extracting reflection objects, positions, amounts and misconducts; S1-2, using OCR to extract text and speech-to-text to process image and voice clues.

[0018] S2, information processing driven by a large model; including: S2-1, converting the clue information to be judged into structured data through a pre-trained large model, and optimizing the information input to the rule engine; S2-2, automatically analyzing and generating structured fields for the existing historical clue processing information through a large model, and correcting them through a verification module to be used for dynamic rule weight training, and increasing the accuracy of dynamic rules.

[0019] S3, setting a dynamic rule engine architecture; including: S3-1, static rule layer: hard requirements of laws and regulations; S3-2, dynamic rule layer: training a logistic regression model based on structured historical data generated by a large model to generate rule weights (such as "amount ≥ 100,000", then weight +0.6); S3-3, conflict resolution strategy: priority ranking (static rule > dynamic rule) + weighted total score determination.

[0020] For example: S1, information processing module; Use a large model to analyze the text and extract keywords such as "Zhang XX" and "rebate 1 million yuan". Use OCR to identify the transfer amount in the screenshot as 1 million yuan, and detect that the transfer party is a construction company.

[0021] S2, rule determination; (1) Use a large model to search the knowledge base and retrieve 3 similar anonymous report cases that have been finally verified, dynamically adjust the "anonymous" deduction weight from -0.3 to -0.1.

[0022] (2) Automatically match static rules according to keywords, trigger "involving... → high risk" (3) Dynamic rule calculation total score = duty (0.7) + amount (1.0) - anonymous (0.1) = 1.6.

[0023] S3, structure output; The total score exceeds the threshold value, and is automatically pushed to the management department, and a report is generated to explain the basis for the determination.

[0024] Based on the above method, the supervision clue automatic grading and shunting method in this embodiment first constructs a multi-modal data fusion model, performs large model driven information processing, and finally sets up a dynamic rule engine architecture.

[0025] When constructing a multi-modal data fusion model, it includes: (1) Text clue processing, based on a domain keyword recognition model of a special knowledge base, to extract reflection objects, duties, amounts, and misconducts; (2) Use OCR to extract text and speech-to-text for image and voice clue processing.

[0026] When performing large model driven information processing, it includes: (1) Convert the clue information to be judged into structured data through a pre-trained large model, and optimize the information input to the rule engine; (2) Process the existing historical clue information through a large model to automatically analyze and generate structured fields, which are corrected by a verification module and used for dynamic rule weight training.

[0027] When setting up a dynamic rule engine architecture, it includes: (1) Static rule layer: legal hard requirements; (2) Dynamic rule layer: based on the structured historical data generated by the large model, train the logistic regression model to generate the rule weight; (3) Conflict resolution strategy: priority ranking + weighted total score determination.

[0028] The above specific embodiments are only specific cases of the present application, and the patent protection scope of the present application includes but is not limited to the above specific embodiments. Any technical solution meeting the above specific embodiments of the present application and any appropriate changes or replacements made by ordinary technical personnel in the art shall fall within the patent protection scope of the present application.

[0029] Although embodiments of the present application have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. A method for automatically grading and diverting monitoring clues, characterized in that: The steps are as follows: S1, multimodal data fusion model; S2, large model-driven information processing; S3. Set up the dynamic rule engine architecture.

2. The method for automatically grading and diverting monitoring clues according to claim 1 is characterized in that: In step S1, it includes: S1-1. Text clue processing: Based on the domain keyword recognition model of the proprietary knowledge base, extract the information reflecting the object, position, amount and disciplinary behavior; S1-2. Use OCR to extract text and speech-to-text to process image and speech clues.

3. The method for automatically grading and diverting monitoring clues according to claim 2 is characterized in that: In step S2, it includes: S2-1. Convert the clue information that needs to be identified into structured data through a pre-trained large model to optimize the information input to the rule engine; S2-2. The existing historical clue processing information is automatically parsed and generated into structured fields through the large model, which are then corrected by the verification module and used for dynamic rule weight training.

4. The method for automatically grading and diverting monitoring clues according to claim 3 is characterized in that: In step S3, it includes: S3-1, static rule layer: rigid regulatory requirements; S3-2, Dynamic Rule Layer: Based on the structured historical data generated by the large model, the logistic regression model is trained to generate rule weights; S3-3. Conflict resolution strategy: priority ranking + weighted total score determination.

5. An automatic grading and diversion system for monitoring clues, characterized by: First, build a multimodal data fusion model and perform large model-driven information processing. Finally, set up a dynamic rule engine architecture.

6. The automatic grading and diversion system for monitoring clues according to claim 5 is characterized in that: When building a multimodal data fusion model, it includes: (1) Text clue processing: Based on the domain keyword recognition model of the proprietary knowledge base, extract the information reflecting the object, position, amount and disciplinary behavior; (2) Use OCR to extract text and speech-to-text to process image and speech clues.

7. The automatic classification and diversion system for monitoring clues according to claim 6 is characterized in that: When performing large model-driven information processing, including: (1) Convert the clue information that needs to be judged into structured data through a pre-trained large model to optimize the information input to the rule engine; (2) The existing historical clue processing information is automatically parsed and generated into structured fields through the large model, which are then corrected by the verification module and used for dynamic rule weight training.

8. The automatic grading and diversion system for monitoring clues according to claim 7 is characterized in that: When setting up the dynamic rules engine architecture, include: (1) Static rule layer: rigid requirements of regulations; (2) Dynamic rule layer: Based on the structured historical data generated by the large model, the logistic regression model is trained to generate rule weights; (3) Conflict resolution strategy: priority sorting + weighted total score determination.

Images