Intelligent examination method for communication work ticket based on cooperation of large model and small model

Abstract

The invention is suitable for the technical field of electric power communication work ticket review, and provides a large and small model collaborative communication work ticket intelligent review method, which comprises the following steps: acquiring multi-modal data of a communication work ticket, and analyzing and aggregating the multi-modal data; carrying out normalization check on the analyzed multi-modal work order data, and if the check is not passed, generating corresponding correction information and returning the work order; if the check is passed, analyzing the data of different dimensions by using a plurality of special small models, and outputting an analysis result and a corresponding confidence coefficient; carrying out integration and correlation analysis on the analysis result of each dimension by utilizing a large model, judging the overall compliance of the work order, and generating an overall compliance judgment result; the confidence output by each small model is integrated, the overall confidence of the work order is calculated and graded, and a corresponding automatic processing or manual rechecking process is executed on the work order; the optimization of accurate and automatic distribution and manual rechecking of the work orders is realized, and the processing efficiency is greatly improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of electric power communication work ticket review, and particularly relates to a communication work ticket intelligent review method based on large and small models. BACKGROUND

[0002] In the current communication network operation and maintenance field, the compliance review of work tickets is a key link to ensure operation safety and process specification. With the surge in the number of work orders and the increasing complexity and variability of formats and content, artificial intelligence is gradually used to assist in review. Currently, two paths are mainly adopted: one is to rely on small models to process structured data and other single-dimensional tasks, which is efficient but difficult to understand cross-modal semantic association, but easy to cause local compliance misjudgment; the other is to use large models with strong semantic understanding ability for end-to-end analysis, but the efficiency is low when processing format checking, rule matching and other standardized tasks, and the adaptation accuracy of domain-specific rules is insufficient. When facing multi-modal complex work orders that integrate structured data, unstructured text and images, these two methods cannot effectively analyze and check multi-dimensional logic based on the collaborative analysis of multi-source information, which results in a large number of work orders that need to be manually reviewed, and the overall processing efficiency is low.

[0003] In view of this, a communication work ticket intelligent review method based on large and small models is proposed. SUMMARY

[0004] The present application provides a communication work ticket intelligent review method based on large and small models, which is used to solve the problem of a large number of work orders that need to be manually reviewed and low overall processing efficiency in work ticket review.

[0005] The present application provides a communication work ticket intelligent review method based on large and small models, which comprises: Obtaining multi-modal data of a communication work ticket, and analyzing and aggregating the multi-modal data, wherein the multi-modal data at least includes structured data, unstructured text data and image data; Performing a specification check on the analyzed multi-modal work order data, and if the check fails, generating corresponding correction information and returning the work order; if the check passes, entering the subsequent processing step; Using multiple special small models to analyze different dimensional data in the multi-modal work order data that passes the check, and outputting the analysis results of each dimension and the corresponding confidence; Integrating and correlating the analysis results of each dimension using a large model, and based on cross-dimensional logical verification to determine the overall compliance of the work order, and generating an overall compliance determination result; Comprehensively considering the confidence output by each small model, calculating the overall confidence of the work order and performing grade division; Based on the overall compliance assessment result and the overall confidence level, the work order will be processed through corresponding automated or manual review procedures.

[0006] Furthermore, the standardization check of the parsed multimodal work order data includes: The system checks whether any required fields in the multimodal work order data are missing. If any are missing, a supplementary template containing the missing fields is generated and returned to the work order. The field content in the multimodal work order data is compared with the predefined format rules to detect whether there are format errors; if there are format errors, a format correction suggestion is generated and returned to the work order.

[0007] Furthermore, the process involves using multiple dedicated mini-models to analyze different dimensions of the multimodal work order data that have passed the inspection, and outputting the analysis results and corresponding confidence levels for each dimension, including: Using a small model of structured data, verify whether the business logic of the structured data in the multimodal work order data conforms to predefined business rules; Output the business logic compliance conclusion of the structured data and the confidence level corresponding to the small model of the structured data.

[0008] Furthermore, the step of using multiple dedicated small models to analyze different dimensions of the multimodal work order data that have passed the inspection, and outputting the analysis results and corresponding confidence scores for each dimension, also includes: Using a text mini-model, semantic parsing is performed on the unstructured text data in the multimodal work order data; Extract entities from semantically parsed text and establish semantic relationships between those entities; Output the entity extraction list, semantic association score, and confidence score corresponding to the text mini-model.

[0009] Furthermore, the step of using multiple dedicated small models to analyze different dimensions of the multimodal work order data that have passed the inspection, and outputting the analysis results and corresponding confidence scores for each dimension, also includes: Using image mini-models, identify whether predefined key elements exist within the image data of the multimodal work order data; Output the element identification list and matching score of the key element; output the confidence level of the image mini-model in evaluating the identification result.

[0010] Furthermore, the integration and correlation analysis of the analytical results of each dimension using a large model includes: The parsing results of the structured data mini-model, the text mini-model, and the image mini-model are aggregated; Based on the knowledge graph of communication services, cross-dimensional associations are performed on the aggregated parsing results.

[0011] Furthermore, the method of determining the overall compliance of a work order based on cross-dimensional logical verification and generating an overall compliance determination result includes: Determine whether the key dimensions are fully compliant; Determine whether the compliance rate of non-critical dimensions has reached the predetermined percentage; Verify whether the logic of cross-dimensional data is consistent; If all the above conditions are met, the work order is deemed compliant overall; otherwise, it is deemed non-compliant overall.

[0012] Furthermore, the process of combining the confidence levels output by each sub-model to calculate the overall confidence level of the work order and classify it into levels includes: Assign preset initial weights to the confidence scores output by the structured data mini-model, the text mini-model, and the image mini-model, respectively; Multiply the confidence score of each small model by its corresponding initial weight, and sum the products to obtain the preliminary confidence score. The preliminary confidence score is compared with the matching degree calculated based on historical work order data, and the preliminary confidence score is calibrated according to the comparison result to obtain the overall confidence score. Based on a preset score threshold range, the overall confidence level is divided into different confidence levels.

[0013] Furthermore, based on a preset score threshold range, the overall confidence level is divided into different confidence levels, including: If the overall confidence level is not lower than the first threshold, it is classified as a high confidence level; If the overall confidence level is lower than the first threshold but not lower than the second threshold, it is classified as a medium confidence level. If the overall confidence level is lower than the second threshold, it is classified as a low confidence level.

[0014] Furthermore, the step of performing corresponding automated processing or manual review of work orders based on the overall compliance determination result and the overall confidence level includes: If the overall compliance determination result is compliant and the overall confidence level is high confidence level, then the work order will be automatically transferred to the next business process. If the overall compliance determination result is non-compliant and the overall confidence level is high confidence level, a structured violation report is generated and the work order is returned. The structured violation report includes at least a description of the violation and rectification suggestions. If the overall confidence level is medium confidence level, the work order will be pushed to the manual review queue. If the overall confidence level is low, the work order will be pushed to the manual deep review queue and marked as a model optimization sample.

[0015] As can be seen from the above technical solutions, the present invention has the following advantages: This invention achieves comprehensive processing of structured data, unstructured text, and images by acquiring and parsing multimodal work order data. It automatically intercepts format errors and missing fields through pre-processing standardization checks, reducing the influx of basic errors into subsequent stages. By utilizing dedicated small models to process data from different dimensions in parallel, and then performing cross-dimensional correlation analysis through a large model, it overcomes the limitations of a single model's local perspective, significantly improving the accuracy of overall compliance judgment. A confidence assessment method based on weighted calculation and historical data calibration enhances the reliability of the judgment results. Finally, based on compliance results and confidence levels, it achieves differentiated automatic routing of work orders, significantly reducing the workload of manual review. This invention improves the accuracy of review while significantly optimizing the work order processing workflow and increasing its automation level. Attached Figure Description

[0016] Figure 1 This is a schematic flowchart of an embodiment of an intelligent review method for communication work tickets based on the collaboration of large and small models in this invention; Figure 2 This is a flowchart illustrating another embodiment of the intelligent review method for communication work tickets based on the collaboration of large and small models in this invention; Figure 3 This is a flowchart illustrating another embodiment of the intelligent review method for communication work tickets based on the collaboration of large and small models in this invention; Figure 4 This is a flowchart illustrating another embodiment of the intelligent review method for communication work tickets based on the collaboration of large and small models in this invention. Detailed Implementation

[0017] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “corresponding to,” and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0018] Example 1 The implementation method in this embodiment can be implemented in a system, on a server, or on a terminal; no specific limitation is made. The method in this application will be described below from the perspective of system implementation. Please refer to... Figure 1 The method provided in this application includes the following steps: S11. Obtain multimodal data of communication work tickets, and parse and aggregate the multimodal data. The multimodal data includes at least structured data, unstructured text data, and image data. S12. Perform a standardization check on the parsed multimodal work order data; S13. If the check fails, generate the corresponding correction information and return the work order; S14. If the inspection passes, use multiple dedicated small models to analyze the data of different dimensions in the multimodal work order data that has passed the inspection, and output the analysis results and corresponding confidence scores for each dimension. S15. Utilize the large model to integrate and correlate the analysis results of each dimension, verify the overall compliance of the work order based on cross-dimensional logic, and generate the overall compliance judgment result. S16. Combine the confidence scores of each sub-model to calculate the overall confidence score of the work order and classify it into levels; S17. Based on the overall compliance assessment results and the overall confidence level, execute the corresponding automated processing or manual review process for the work order.

[0019] The implementation principle and process of the above steps are as follows: First, through RPA robots, OCR engines, and feature extraction technologies, the system automatically acquires and parses multimodal data, including structured data, unstructured text, and images, in the work order, completing data aggregation and preliminary structuring. Then, the system performs a rigorous compliance check on the parsed multimodal work order data, focusing on detecting missing fields and format errors. If the check fails, a work order with example values ​​or correction suggestions is automatically generated and returned to the submission end, achieving early interception and rapid correction of errors. For work orders that pass the check, the system concurrently calls multiple dedicated small models for structured data, text, and images, performing in-depth analysis on the corresponding dimensions of data and outputting analysis results with confidence levels. The large model, based on a communication business knowledge graph, integrates and performs cross-dimensional correlation analysis on the above multi-dimensional analysis results, determining the overall compliance of the work order by verifying the logical consistency between data. The system further integrates the confidence levels of each small model, combining preset weights and historical data for weighted calculation and calibration, ultimately classifying the overall confidence level into three levels: high, medium, and low. Ultimately, based on the overall compliance assessment results and confidence level, the system performs differentiated processing, such as fully automated processing, generating violation reports for return, or pushing them to manual review queues of different priorities.

[0020] Example 2 Please see Figure 2 The normative checks in this invention include the following: 1. Check if any required fields in the multimodal work order data are missing; if so, generate a supplementary template containing the missing fields and return it to the work order. 2. The field content in the multimodal work order data is compared with the predefined format rules to detect whether there are format errors; if there are format errors, format correction suggestions are generated and returned to the work order.

[0021] First, the system checks if any required fields in the multimodal work order data are missing; this is the field missing judgment node in the flowchart. If a missing field is found, a supplementary template is generated. This template highlights the missing fields and provides example values ​​based on historical data. The work order is then returned, and the process ends. If the fields are complete, the system proceeds to the format error judgment node, comparing the work order field content with predefined format rules (such as IP address, device number format, etc.). If a format error is found, format correction suggestions are generated, for example, displaying the standard format through a floating prompt box and providing a one-click correction entry. The work order is then returned, and the process ends. Only when both checks pass (i.e., the basic checks are passed) can the work order proceed to the next processing stage. This standardized check step, as a pre-filter, effectively intercepts basic errors, significantly improving the efficiency and accuracy of the subsequent intelligent review process.

[0022] Example 3 Please see Figure 3 After passing the basic checks, the work order enters the second-level type adaptation check and small model single-dimensional parsing stage. This process is implemented as follows: 1. A small model of structured data is used to verify whether the business logic of structured data in multimodal work order data conforms to predefined business rules; 2. Conclusions on the compliance of business logic with structured data and the confidence level of the small model of structured data.

[0023] Specifically, predefined business rules are stored in a dynamically updated business rule base, whose content originates from the telecommunications industry's operating procedures, security specifications, and specific internal workflows. The structured data mini-model verifies the compliance of business logic by matching fields in the work order (such as device number, operation type, and timestamp) with constraints in the rule base. For example, it verifies whether the device number of the operation exists in the currently valid asset list, or whether the sequence of subtasks in a complex work order conforms to the safety procedure of powering off before operating. After verification, the model not only outputs a binary conclusion of compliance or non-compliance, but also outputs a quantified confidence score based on factors such as the accuracy of rule matching and data integrity.

[0024] 1. Utilize a small text model to perform semantic parsing on unstructured text data in multimodal work order data; 2. Extract entities from the semantically parsed text and establish semantic relationships between entities; 3. Output the entity extraction list, semantic association score, and confidence score of the text mini-model.

[0025] In practice, the text-based mini-model employs named entity recognition technology based on a pre-trained language model to extract key entities from free text such as fault descriptions and work summaries in work orders. These entities include faulty equipment names, maintenance personnel employee numbers, and fault codes. Furthermore, a relation extraction model analyzes the semantic relationships between entities, such as determining whether a person and their performed operation are pointing to the correct equipment. Semantic association scoring quantifies the degree of confidence in these relationships. The confidence score of the entire parsing process integrates the accuracy of entity recognition and the reliability of relation extraction.

[0026] 1. Utilize image mini-models to identify whether predefined key elements exist within the image data of multimodal work order data; 2. Output a list of key elements for identification and a matching score; output the confidence level of the image mini-model in evaluating the identification results.

[0027] The image mini-model, based on deep convolutional neural networks or object detection algorithms, analyzes scanned images in work order attachments to identify predefined key visual elements such as official company seals, handwritten signatures of responsible persons, and photos of on-site safety measures. The matching score reflects the degree of similarity between the identified elements and standard templates in terms of shape, position, and sharpness. The confidence score of the model output reflects the reliability of the recognition results and is affected by factors such as image quality and element occlusion.

[0028] After completing the single-dimensional analysis, the process enters the large-scale model cross-dimensional correlation analysis stage: 1. Aggregate the parsing results of the structured data mini-model, text mini-model, and image mini-model; 2. Based on the knowledge graph of communication services, cross-dimensional associations are performed on the aggregated parsing results.

[0029] Large-scale models (such as large-scale language models, LLMs) act as information fusion and inference engines. They structurally aggregate the outputs of all smaller models (including conclusions, lists, scores, etc.) to form a unified set of contextual information. Then, they access a built-in communication business knowledge graph, which contains entity relationships such as device topology, personnel permissions, and workflows. The large-scale model uses this knowledge for cross-dimensional logical verification; for example, it checks the consistency between the faulty device A extracted from the text smaller model and the operating device number in the structured data, or verifies whether the signatory identified in an image matches the approver recorded in the text.

[0030] Finally, an overall compliance assessment is conducted. This step includes: Determine whether the key dimensions are fully compliant; Determine whether the compliance rate of non-critical dimensions has reached the predetermined percentage; Verify whether the logic of cross-dimensional data is consistent; If all the above conditions are met, the work order is deemed compliant overall; otherwise, it is deemed non-compliant overall.

[0031] Based on the results of the aforementioned correlation analysis, the large model executes a multi-condition decision-making process. First, it determines that key dimensions involving safety and regulations (such as confirmation of power outage operations) must be 100% compliant. Second, for non-critical dimensions (such as certain descriptive fields), a certain percentage of non-compliance is allowed (e.g., a compliance rate threshold is set at 90%). Finally, it comprehensively verifies the logical consistency between all dimensions to avoid contradictions such as textual records indicating completed repairs while images show warning signs not being removed. Only when all conditions are met is the work order deemed compliant overall, and the process then proceeds to the confidence assessment stage.

[0032] Example 4 Please see Figure 4The system then enters the confidence assessment and differentiation processing stage, which is implemented as follows: 1. Assign preset initial weights to the confidence scores of the structured data mini-model, text mini-model, and image mini-model respectively; 2. Multiply the confidence score of each small model output by its corresponding initial weight, and sum the products to obtain the preliminary confidence score; 3. Compare the preliminary confidence score with the matching degree calculated based on historical work order data, and calibrate the preliminary confidence score according to the comparison results to obtain the overall confidence score; 4. Based on the preset score threshold range, the overall confidence level is divided into different confidence levels.

[0033] In practice, the system first assigns initial weights to the outputs of the smaller models in different dimensions, reflecting their relative importance. For example, in the scenario of reviewing communication work orders, the accuracy of structured data is usually the most critical, so it can be assigned a relatively high weight of 0.4; while unstructured text and image data, as important supplements, can each be assigned a weight of 0.3. The system then multiplies the confidence score of each smaller model's output by its corresponding weight and sums the results to obtain the preliminary confidence score.

[0034] Next, the system introduces a dynamic calibration mechanism based on historical data to improve the reliability of the assessment. This mechanism calculates the matching degree between the feature vector of the current work order and the historical work order database, comparing features such as work order type, involved equipment, and key operations. If the feature matching degree between the current work order and historical compliant work orders is high, for example, exceeding 80%, it indicates that the current situation is common and the model's judgment is based on sufficient evidence; the preliminary confidence score can be directly used as the overall confidence score. Conversely, if the matching degree is low, it indicates that the current work order belongs to a special or marginal case, increasing the uncertainty of the model's judgment. The system will appropriately lower the preliminary score based on the degree of matching degree deviation, such as by 5 to 20 points, thereby obtaining a calibrated and more robust overall confidence score.

[0035] In this embodiment, based on a preset score threshold range, the overall confidence level is divided into different confidence levels; wherein: if the overall confidence level is not lower than the first threshold, it is classified as a high confidence level; if the overall confidence level is lower than the first threshold but not lower than the second threshold, it is classified as a medium confidence level; if the overall confidence level is lower than the second threshold, it is classified as a low confidence level.

[0036] In practical applications, the threshold range needs to be set based on extensive experiments and experience. For example, the first threshold can be set to 80 points, and the second threshold to 60 points. This means that when the overall confidence level reaches 80 points or above, it is classified as a high confidence level, indicating that the judgment result is very reliable; when the score falls between 60 and 79 points, it is classified as a medium confidence level, indicating that the judgment result has a certain degree of reliability but still has uncertainty; and when the score is below 60 points, it is classified as a low confidence level, indicating that the model has a very low grasp on the judgment of the work order and requires manual review. This classification provides a key basis for implementing subsequent differentiated processing procedures.

[0037] In this embodiment, the system executes the differentiated processing flow shown in the attached figure based on the aforementioned overall compliance determination result and the predefined overall confidence level: If the judgment result is "compliant" and the confidence level is "high confidence level", then "automatically transfer to the next stage" will be executed; If the judgment result is "non-compliant" and the confidence level is "high confidence level", then "generate a structured violation report" and "return" the work order. The report will clearly list the violation items, violation dimensions, rule basis and rectification suggestions, and support one-click modification of correctable fields; If the confidence level is "medium confidence level", regardless of the compliance result, the work order will be "pushed to the manual review queue" and marked as "medium priority" to help reviewers quickly locate the points of contention; If the confidence level is "low confidence level", the work order will be "pushed to the manual deep review queue" and marked as "high priority", and at the same time, the work order will be "marked as a model optimization sample". After the manual review is completed, its conclusion will be "synchronized to the model training library" for continuous optimization of the judgment ability of the large and small models.

[0038] The method of this invention significantly improves the accuracy and reliability of compliance review of complex multimodal work orders by dividing the work and coordinating large and small models and cross-dimensional logical verification, and effectively reduces the false judgment rate. Through intelligent confidence assessment and result-driven differentiated processing, it realizes accurate automatic diversion of work orders and optimizes manual review, which greatly improves processing efficiency.

[0039] It is understood that those skilled in the art can combine various implementation methods in the above embodiments under the guidance of the above examples to obtain technical solutions with multiple implementation methods.

[0040] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A size model cooperative communication work ticket intelligent review method, characterized in that, The method comprises the following steps: acquiring multi-modal data of a communication work ticket and performing parsing and aggregation on the multi-modal data, wherein the multi-modal data at least includes structured data, unstructured text data and image data; performing a specification check on the parsed multi-modal work order data, and if the check fails, generating corresponding correction information and returning the work order; if the check passes, entering a subsequent processing step; using multiple special small models to respectively parse different dimensional data in the multi-modal work order data that passes the check, and outputting the parsing results of each dimension and the corresponding confidence; using a large model to integrate and analyze the parsing results of each dimension, and based on cross-dimensional logical verification, determining the overall compliance of the work order, and generating an overall compliance determination result; comprehensively calculating the overall confidence of the work order based on the confidence output by each small model and performing level division; based on the overall compliance determination result and the overall confidence level, performing corresponding automatic processing or manual review process on the work order. 2.The size model coordinated communication work ticket intelligent review method of claim 1, wherein, The specification check on the parsed multi-modal work order data comprises: detecting whether the content of the multi-modal work order data is missing in the required field; if it is missing, generating a supplement template containing the missing field and returning the work order; comparing the field content in the multi-modal work order data with the pre-defined format rule to detect whether there is a format error; if there is a format error, generate a format correction suggestion and return the work order. 3.The size model coordination communication work ticket intelligent review method of claim 1, wherein, The use of multiple special small models to respectively parse different dimensional data in the multi-modal work order data that passes the check, and output the parsing results of each dimension and the corresponding confidence, comprises: using a structured data small model to check whether the business logic of the structured data in the multi-modal work order data conforms to the pre-defined business rules; output the business logic compliance conclusion of the structured data and the confidence of the structured data small model. 4.The size model coordination communication work ticket intelligent review method of claim 3, wherein, The use of multiple special small models to respectively parse different dimensional data in the multi-modal work order data that passes the check, and output the parsing results of each dimension and the corresponding confidence, further comprises: using a text small model to perform semantic analysis on the unstructured text data in the multi-modal work order data; extracting entities from the text after semantic analysis and establishing semantic association between the entities; output the entity extraction list, semantic association score and confidence of the text small model.

5. The size model coordinated communication work ticket intelligent review method of claim 4, wherein, The use of multiple special small models to respectively parse different dimensional data in the multi-modal work order data that passes the check, and output the parsing results of each dimension and the corresponding confidence, further comprises: using an image small model to identify whether there is a pre-defined key element in the image data in the multi-modal work order data; output the element identification list and matching degree score of the key element; output the confidence of the image small model in judging the recognition result. 6.The size model coordination communication work ticket intelligent review method of claim 1, wherein, The use of a large model to integrate and analyze the parsing results of each dimension comprises: aggregating the parsing results of the structured data small model, the text small model and the image small model; Based on the communication service knowledge graph, the aggregated analysis results are cross-dimensionally associated.

7. The size model coordinated communication work ticket intelligent review method of claim 6, wherein, The cross-dimension logic verification is used to determine the overall compliance of the work order, and an overall compliance determination result is generated, including: determining whether the key dimension is fully compliant; determining whether the compliance rate of the non-key dimension reaches a predetermined proportion; checking whether the logic of cross-dimension data is consistent; if the above conditions are met, the work order is determined to be overall compliant, otherwise it is determined to be overall non-compliant. 8.The size-model-coordinated communication work ticket intelligent review method according to claim 1, wherein, The confidence levels output by the small models are integrated to calculate the overall confidence level of the work order and perform level division, including: assigning a preset initial weight to the confidence level output by the structured data small model, the text small model, and the image small model, respectively; multiply the confidence level output by each small model by the corresponding initial weight, and sum the products to obtain a preliminary confidence score; compare the preliminary confidence score with the matching degree calculated based on historical work order data, and calibrate the preliminary confidence score based on the comparison result to obtain an overall confidence level; based on a preset score threshold interval, the overall confidence level is divided into different confidence levels. 9.The size model coordinated communication work ticket intelligent review method of claim 8, wherein, based on a preset score threshold interval, the overall confidence level is divided into different confidence levels, including: if the overall confidence level is not less than a first threshold, it is divided into a high confidence level; if the overall confidence level is lower than the first threshold but not lower than a second threshold, it is divided into a medium confidence level; if the overall confidence level is lower than the second threshold, it is divided into a low confidence level. 10.The size model coordination communication work ticket intelligent review method of claim 1, wherein, According to the overall compliance determination result and the overall confidence level, the work order is executed with corresponding automatic processing or manual review process, including: if the overall compliance determination result is compliant and the overall confidence level is high, the work order is automatically forwarded to the next business link; if the overall compliance determination result is non-compliant and the overall confidence level is high, a structured violation report is generated and the work order is returned, the structured violation report including at least violation item description and rectification suggestion; if the overall confidence level is medium, the work order is pushed to the manual review queue; if the overall confidence level is low, the work order is pushed to the manual deep review queue, and the work order is marked as a model optimization sample.

Images