Structured text segmentation verification method and system based on logarithmic linear tolerance budget
By employing a log-linear tolerance budgeting mechanism and boundary protection adjustments, the problem of instability in text segmentation methods under different document sizes is resolved, achieving efficient and accurate text segmentation verification. This approach is suitable for a unified monitoring framework that covers documents ranging from short to long.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- CHENGDU AIRCRAFT DESIGN INST OF AVIATION IND CORP OF CHINA
- Filing Date
- 2025-12-27
- Publication Date
- 2026-05-12
AI Technical Summary
Existing text segmentation methods cannot simultaneously achieve high accuracy for short documents and low false alarms for long documents when processing documents of different sizes, resulting in unstable detection results and high maintenance costs.
A log-linear tolerance budgeting mechanism is adopted, which calculates the tolerance rate through a log-linear mathematical model and combines it with boundary protection adjustments to ensure that the tolerance rate decays smoothly under different document sizes. The amount of content loss and duplication is calculated independently, and absolute values are used for comparison to generate verification results.
It achieves stable and accurate text segmentation verification under different document sizes, reduces computational complexity and supports batch parallel processing, and is suitable for a unified monitoring framework from short documents to long documents.
Smart Images
Figure CN122021618A_ABST
Abstract
Description
Technical Field
[0001] This application belongs to the field of text processing technology, and in particular relates to a structured text segmentation verification method and system based on logarithmic linear tolerance budget. Background Technology
[0002] Currently, many systems require splitting long documents into smaller chunks for processing. However, this process can lead to content loss or duplication.
[0003] Existing detection methods either miss errors in short documents or generate too many false positives in long documents, as detailed below: 1. Fixed absolute threshold content integrity monitoring The basic idea is to pre-set a fixed upper limit for the difference between characters (or words) (such as 1000 characters), and after segmentation, calculate the absolute difference between the length of the original document and the sum of the lengths of all text blocks. If the difference exceeds the threshold, an alarm or rollback will be triggered. Main drawback: Lack of sensitivity to document length – short documents will not trigger an alarm even if a high percentage is lost, while long documents will trigger a false alarm even if only a small percentage is lost, resulting in both missed detections and false alarms, which affects the accuracy of subsequent tasks such as retrieval and answer extraction. 2. Fixed percentage threshold verification mechanism The basic idea is to define the tolerable error as a fixed percentage (e.g., 1%) of the document length, and calculate the allowable error according to this percentage after segmentation. If the error is exceeded, an alarm will be triggered. Main drawback: The linear scaling model cannot meet the needs of documents of different sizes. Small documents may trigger false alarms due to minor punctuation differences, while large documents may lose several pages of content without triggering an alarm. It is both "too strict" and "too lenient", and cannot meet the adaptive requirements of both small and large documents. 3. Linear decay adaptive model The basic idea is to use a linear function to make the tolerance decrease linearly with the document length. After segmentation, the relative tolerance rate is calculated based on the current length, and then converted into an absolute threshold for comparison. Main drawbacks: There is a "cliff effect" - at a certain critical length, even a difference of 1 character can cause a sudden change in tolerance, resulting in discontinuous verification results; at the same time, different document types require separate parameter tuning, parameters are difficult to be universal across domains, maintenance costs are high and results are unstable. 4. Multi-level threshold content validation framework The basic idea is to divide the document into several intervals based on its length (such as short, medium, and extra-large), configure independent absolute or relative thresholds for each interval, determine the interval to which the document belongs after segmentation, and then load the corresponding thresholds for verification. Main drawbacks: The interval boundaries can cause sudden changes in tolerance – adjacent sides differing by only a few characters use completely different thresholds, resulting in inconsistent fault tolerance; as the number of partitions increases, the configuration and management complexity increases exponentially. Each of the four solutions has its own advantages and disadvantages, but none of them can simultaneously meet the requirements of "high accuracy for short documents" and "low false alarms for long documents" for documents ranging from hundreds of words to millions of words. Summary of the Invention
[0004] The purpose of this invention is to provide a unified monitoring framework for structured text segmentation pipelines. By introducing a log-linear tolerance budget mechanism, this invention can adaptively handle the problems of content loss and content duplication in documents of different sizes.
[0005] In a first aspect, this application provides a structured text segmentation verification method based on logarithmic linear tolerance budgeting, the method comprising: Step 1: Access the text blocks output by the external segmentation module, read and record the total length of the original document and the length of each text block, and provide input for subsequent fault tolerance calculations; Step 2: Based on the log-linear mathematical model, calculate the relative tolerance rate of the current document, so that the tolerance rate decreases smoothly with the document size, thereby balancing the accuracy of short documents and the robustness of long documents. Step 3: Apply the lower limit of absolute loss and the upper limit of relative tolerance to the initially calculated tolerance rate to correct for upper and lower bounds, so as to prevent abrupt changes in the tolerance rate at extremely small or extremely large document lengths. Step 4: Calculate the amount of content loss and content duplication introduced after text segmentation independently, and output them in absolute numerical form to ensure the accuracy of downstream tolerance comparison; Step 5: Compare the calculated amount of content loss and content duplication with the allowable error adjusted for boundary protection to determine whether the segmentation quality passes or needs to be reviewed. Step 6: Based on the decision results, route the "verification passed" text block to downstream components such as the question-answering system and vector index, and trigger alarms or rollback mechanisms for "verification failed" cases to complete the entire verification pipeline.
[0006] Preferably, step 2 includes: Calculate the log-linear tolerance rate using formula (1). (1) Where n is the length of the source document, n min Where T is the reference length, T0 is the reference tolerance, and s is the logarithmic decay slope. To calculate the tolerance rate.
[0007] Preferably, step 3 includes: After calculating the initial tolerance rate Next, boundary protection adjustments need to be performed to ensure that the final tolerance rate does not exceed the maximum relative tolerance limit C due to insufficient document length. rel It will not fall below the minimum absolute loss constant C due to excessive document length. abs This smooths the tolerance curve, avoids abrupt changes at extreme lengths, and ensures the stability and consistency of the segmentation verification results.
[0008] Preferably, step 4 includes: By independently calculating the absolute values of content loss and content duplication, accurate input is provided for subsequent tolerance comparisons, as follows: Using formula (2), calculate the content loss. - abs: When the sum of the lengths of all text blocks is less than the length of the original document, (2) Where n is the length of the original document. The sum of the lengths of all text blocks; Calculate the content repetition rate using formula (3). When the sum of the lengths of all text blocks is greater than the length of the original document, (3).
[0009] Preferably, step 5 includes: The relative tolerance rate T calculated upstream final Convert to the corresponding absolute allowance, and then calculate the content loss separately. - Abs and content duplication A double comparison is performed to generate the verification results, as follows: Absolute tolerance for conversion:
[0010] loss of content - Abs and content duplication Perform independent comparisons to ensure that both are within acceptable limits; Generate clear decision signals for subsequent processing: If the verification passes, it is considered "verification passed"; otherwise, it is considered "verification failed" and an alarm or rollback mechanism is triggered.
[0011] Secondly, this application also provides a structured text segmentation verification system based on logarithmic linear tolerance budgeting, the system comprising: The text block data acquisition module is used to access the text blocks output by the external segmentation module, read and record the total length of the original document and the length information of each text block, and provide input for subsequent fault tolerance calculations; The log-linear tolerance calculation module is used to calculate the relative tolerance rate of the current document based on a log-linear mathematical model, so that the tolerance decreases smoothly with the document size, thus balancing the accuracy of short documents and the robustness of long documents. The boundary protection module is used to apply the lower limit of absolute loss and the upper limit of relative tolerance to the initially calculated tolerance rate, so as to prevent abrupt changes in tolerance at extremely small or extremely large document lengths. The loss and duplication measurement module is used to independently calculate the amount of content loss and content duplication introduced after text segmentation, and output them in absolute numerical form to ensure the accuracy of downstream tolerance comparison. The threshold comparison decision module is used to compare the calculated amount of content loss and content duplication with the allowable error adjusted by boundary protection, and to determine whether the segmentation quality passes or needs to be reviewed. The verification result routing module is used to route the "verification passed" text block to downstream components such as the question-answering system and vector index based on the decision result, and to trigger alarms or rollback mechanisms for "verification failed" cases, thus completing the entire verification pipeline.
[0012] Preferably, the logarithmic linear tolerance calculation module is further used to calculate the logarithmic linear tolerance rate using formula (1). (1) Where n is the length of the source document, n min Where T is the reference length, T0 is the reference tolerance, and s is the logarithmic decay slope. To calculate the tolerance rate.
[0013] Preferably, the boundary protection module is further configured to calculate the initial tolerance rate. Next, boundary protection adjustments need to be performed to ensure that the final tolerance rate does not exceed the maximum relative tolerance limit C due to insufficient document length. rel It will not fall below the minimum absolute loss constant C due to excessive document length. abs This smooths the tolerance curve, avoids abrupt changes at extreme lengths, and ensures the stability and consistency of the segmentation verification results.
[0014] Preferably, the loss and duplication measurement module is further used to provide accurate input for subsequent tolerance comparison by independently calculating the absolute measurement values of content loss and content duplication, as follows: Using formula (2), calculate the content loss. - abs: When the sum of the lengths of all text blocks is less than the length of the original document, (2) Where n is the length of the original document. The sum of the lengths of all text blocks; Calculate the content repetition rate using formula (3). When the sum of the lengths of all text blocks is greater than the length of the original document, (3).
[0015] Preferably, the threshold comparison decision module is used to replace the relative tolerance rate T calculated upstream. final Convert to the corresponding absolute allowance, and then calculate the content loss separately. - Abs and content duplication A double comparison is performed to generate the verification results, as follows: Absolute tolerance for conversion:
[0016] loss of content - Abs and content duplication Perform independent comparisons to ensure that both are within acceptable limits; Generate clear decision signals for subsequent processing: If the verification passes, it is considered "verification passed"; otherwise, it is considered "verification failed" and an alarm or rollback mechanism is triggered.
[0017] The beneficial technical effects of this application are as follows: 1. Elegant mathematical model: Based on natural mathematical laws on a logarithmic scale, similar to scientific measurements such as earthquake magnitude and decibels.
[0018] 2. Simple parameter settings: Only 4 parameters are required, and the default values are suitable for most scenarios.
[0019] 3. Predictable results: The tolerance rate decreases proportionally for every 10-fold increase in document length.
[0020] 4. High efficiency: Low computational complexity, supports batch parallel processing. Attached Figure Description
[0021] Figure 1 A flowchart illustrating a structured text segmentation verification method based on logarithmic linear tolerance budgeting, provided in this application embodiment; Figure 2 A detailed flowchart for calculating the log-linear tolerance is provided for embodiments of this application; Figure 3A schematic diagram of the boundary protection dual constraint mechanism provided in the embodiments of this application; Figure 4 Detailed flowchart of the measurement calculator provided in the embodiments of this application; Figure 5 The threshold comparator decision flowchart provided in the embodiments of this application is shown. Detailed Implementation
[0022] The present invention will be further described in detail below with reference to the embodiments and accompanying drawings, but the embodiments of the present invention are not limited thereto.
[0023] It should be noted that, to ensure accuracy in reading comprehension, the following are the definitions of key terms used in this document:
[0024] Please see the appendix Figure 1 -Appendix Figure 5 This invention develops a structured text segmentation and verification method and system based on logarithmic linear tolerance budgeting, which can automatically adjust the error tolerance according to the document length. Short document: rigorously checked (to prevent overlooking minor errors) Long documents: Relatively lenient checks (to avoid false positives) Adjustment pattern: Logarithmic scaling is used for adjustment, similar to the calculation method for earthquake magnitude. Key innovation points This is achieved using a mathematical formula: when the document length increases tenfold, the allowable error rate decreases by a predictable proportion. 500-word short document: 8% error tolerance allowed (40 characters) 5000-word article: Allow 5% error tolerance (250 characters) A 500,000-word book: 2% error tolerance allowed (10,000 characters). It can effectively prevent error detection problems in documents of all sizes, from short documents (hundreds of characters) to books (millions of characters).
[0025] 3.1 Log-linear tolerance budget calculation (core algorithm) 3.1.1 Parameter Description
[0026] 3.1.2 Core Calculation Formula Main formulas:
[0027] Formula explanation: Calculate tolerance rate = Baseline tolerance rate × (-Attenuation coefficient × Logarithmic difference of document length) When document length = baseline length (1000 characters), tolerance rate = baseline tolerance rate (4%). When the document length increases tenfold, the tolerance rate is calculated as follows: Decreasing by a factor of two For example, when s=0.5, for every tenfold increase in document length, the tolerance rate decreases to 31.6% of its original value. Numerical calculation example:
[0028] 3.1.3 Detailed Explanation of Calculation Steps Step 1: Calculate the scale difference
[0029] n is the length of the current document. min This refers to the baseline length. This can be understood as how many orders of magnitude the length n of the given text differs from the baseline length.
[0030] Function: Quantifies the order-of-magnitude difference between the current document and a standard document (1000 characters). Example: 10,000-character document: log 10 (10000) - log 10 (1000) = 4 - 3 = 1 Step 2: Calculate the initial tolerance rate If document length ≤ base length: = Baseline tolerance rate (4%) else: = Baseline tolerance rate × (-attenuation slope × log_ratio) Step 3: Boundary Constraints Ensure the tolerance rate is within a reasonable range (0% < ≤ 100%) Detailed calculation demonstration (10,000-character document): Step 1: Determine the input parameters Current document length: n = 10,000 characters Base length: n_min = 1,000 characters Benchmark tolerance rate: T0 = 4% Attenuation coefficient: s = 0.5 Step 2: Calculate the scale difference log_ratio = log 10(current length) - log 10 (Base length) log_ratio = log 10 (10,000)- log 10 (1,000) = 4 - 3 = 1 Step 3: Apply the core formula = T0× (-s × log_ratio) = 4% × (-0.5 × 1) = 4% × (-0.5) = 4% × 0.316 = 1.26% Step 4: Boundary Check 1.26% is within the valid range (0,1] and requires no adjustment. Step 5: Calculate the final allowable error Tolerance = Tolerance Rate × Document Length = 1.26% × 10,000 = 126 characters Interpretation of Results: For a document of 10,000 characters, the system allows a maximum of 126 characters of content to be lost or duplicated.
[0031] Adjustment of the logarithmic decay slope s The smaller the value of s, the smoother the exponential decay, and the slower the rate of decrease in tolerance due to the increase in length. This is suitable for application scenarios where there is no obvious distinction between long and short documents and the tolerance rate for long and short documents is not expected to be significantly different.
[0032] The larger the value of s, the steeper the decay, and the more sensitive the tolerance rate decreases to the increase in document length, approaching the zero tolerance rate faster. This is suitable for application scenarios with extremely high error requirements for long documents.
[0033] 3.2 Boundary protection adjustment (tolerance rate upper and lower limit correction) like Figure 3 As shown, this module implements boundary protection adjustment for logarithmic linear tolerance. After calculating the initial tolerance rate... Next, boundary protection adjustments need to be performed to ensure that the final tolerance rate will not exceed the maximum relative tolerance rate limit due to the document length being too small, nor will it fall below the minimum absolute loss constant due to the document length being too large. This will smooth the tolerance curve, avoid abrupt changes at extreme lengths, and ensure the stability and consistency of the segmentation verification results.
[0034] 3.2.1 Parameter Definition
[0035] 3.2.2 Calculation process and formulas The boundary protection adjustment module described in this invention aims to correct the tolerance distortion problem that may occur in log-linear models under extreme document length conditions. The specific approach involves two steps: Upper and lower limit boundary protection parameter settings Set an absolute loss lower bound to ensure that any document allows at least a fixed minimum loss (e.g., 100 characters) to avoid being overly sensitive to short documents; "Oversensitivity to short documents" means that when a document is very short, the number of characters that can be lost based on the relative tolerance rate will be very small, or even lower than the normal deviation generated during actual segmentation, which will cause the system to frequently misjudge "exceeding tolerance" and report errors.
[0036] For example: Suppose we set the relative tolerance C_rel=1%, which means that we can allow a maximum loss of 1% of the document length.
[0037] If a document has only 300 characters, then proportionally, a loss of up to 300 × 1% = 3 characters is allowed.
[0038] However, in the actual segmentation process, due to differences in punctuation, spaces, or encoding, it is easy to produce a normal deviation of 5–10 characters.
[0039] At this point, the system will mistakenly judge the segmentation as failed because 5 characters are greater than 3 characters, clearly showing "oversensitivity to short documents".
[0040] By introducing an absolute loss lower bound C_abs (e.g., 100 characters), at least 100 characters of loss are allowed regardless of how short the document is, so that frequent errors will not occur due to small deviations in short documents.
[0041] Set a relative tolerance limit to limit the maximum acceptable loss percentage (e.g., 8%) to prevent excessive leniency for long documents.
[0042] "Overly lenient treatment of long documents" refers to the situation where, when a document is very long, the number of characters that can be lost proportionally increases, potentially reaching thousands or tens of thousands of characters, without triggering an alarm, thus masking a large amount of information loss.
[0043] For example Assuming a document is 100,000 characters long, without a relative upper limit, and according to the linear model allowing for a 10% loss, then 10,000 characters can be lost while the system still judges it as "passing".
[0044] However, such a high degree of "leniency" can easily lead to the omission of key information, affecting the quality of subsequent searches and question-and-answer sessions.
[0045] When the relative tolerance limit C_rel = 8%, a document with the same 100,000 characters is allowed a loss of up to 8,000 characters. Any loss exceeding this limit will trigger an alarm, thus preventing overly lenient validation results for long documents.
[0046] Calculate the final tolerance rate The calculation process is as follows:
[0047] Calculation process description:
[0048] Output example:
[0049] 3.3 Loss and Repeatability Calculator like Figure 4 As shown, this invention provides accurate input for subsequent tolerance comparisons by independently calculating the absolute values of content loss and content duplication. The specific definitions are as follows: Content loss: When the sum of the lengths of all text blocks is less than the original document length,
[0050] Where n is the length of the original document (number of characters or number of tokens). This is the sum of the lengths of all text blocks (number of characters or words). Content duplication rate: When the sum of the lengths of all text blocks exceeds the length of the original document,
[0051] The advantages of using absolute value as a metric are: Maintaining consistent units (number of characters or number of tokens) ensures direct compatibility with downstream tolerance calculations; Avoid making percentage metrics overly sensitive in short document scenarios; Explicitly distinguishing between "missing" and "duplicate" errors facilitates the adoption of targeted repair strategies (such as supplementary segmentation or deduplication) based on different scenarios.
[0052] 3.4 Threshold Comparator and Verification Logic This invention provides the final decision logic for segmentation quality verification. Its core idea is to convert the relative tolerance rate calculated upstream into the corresponding absolute allowance, and then perform a dual comparison of "content loss" and "content duplication" to generate the verification result. The specific implementation needs to address the following technical challenges: Absolute tolerance for conversion:
[0053] The amount of lost content and the amount of duplicate content were compared independently to ensure that both were within acceptable limits. Generate clear decision signals for subsequent routing: like If the verification passes, it is considered "verification passed"; otherwise, it is considered "verification failed" and an alarm or rollback mechanism is triggered.
[0054] In terms of technical implementation, All comparisons are performed independently using the same Allowance threshold to ensure consistent evaluation criteria for both types of errors. Pure scalar comparisons are used to avoid complex model calculations, achieving modularity and low computational overhead. The decision logic and tolerance calculation module are decoupled, making it easy to reuse in various pipelines such as QA systems and vector indexes.
Claims
1. A structured text segmentation verification method based on logarithmic linear tolerance budgeting, characterized in that, The method includes: Step 1: Access the text blocks output by the external segmentation module, read and record the total length of the original document and the length of each text block, and provide input for subsequent fault tolerance calculations; Step 2: Based on the log-linear mathematical model, calculate the relative tolerance rate of the current document, so that the tolerance rate decreases smoothly with the document size, thereby balancing the accuracy of short documents and the robustness of long documents. Step 3: Apply the lower limit of absolute loss and the upper limit of relative tolerance to the initially calculated tolerance rate to correct for upper and lower bounds, so as to prevent abrupt changes in the tolerance rate at extremely small or extremely large document lengths. Step 4: Calculate the amount of content loss and content duplication introduced after text segmentation independently, and output them in absolute numerical form to ensure the accuracy of downstream tolerance comparison; Step 5: Compare the calculated amount of content loss and content duplication with the allowable error adjusted for boundary protection to determine whether the segmentation quality passes or needs to be reviewed. Step 6: Based on the decision results, route the "verification passed" text block to downstream components such as the question-answering system and vector index, and trigger alarms or rollback mechanisms for "verification failed" cases to complete the entire verification pipeline.
2. The method according to claim 1, characterized in that, Step 2 includes: Calculate the log-linear tolerance rate using formula (1). （1） Where n is the length of the source document, n min Where T is the reference length, T0 is the reference tolerance, and s is the logarithmic decay slope. To calculate the tolerance rate.
3. The method according to claim 2, characterized in that, Step 3 includes: After calculating the initial tolerance rate Next, boundary protection adjustments need to be performed to ensure that the final tolerance rate does not exceed the maximum relative tolerance limit C due to insufficient document length. rel It will not fall below the minimum absolute loss constant C due to excessive document length. abs This smooths the tolerance curve, avoids abrupt changes at extreme lengths, and ensures the stability and consistency of the segmentation verification results.
4. The method according to claim 3, characterized in that, Step 4 includes: By independently calculating the absolute values of content loss and content duplication, accurate input is provided for subsequent tolerance comparisons, as follows: Using formula (2), calculate the content loss. - abs: When the sum of the lengths of all text blocks is less than the length of the original document, （2） Where n is the length of the original document. This is the sum of the lengths of all text blocks; Calculate the content repetition rate using formula (3). When the sum of the lengths of all text blocks is greater than the length of the original document, （3）。 5. The method according to claim 4, characterized in that, Step 5 includes: The relative tolerance rate T calculated upstream final Convert to the corresponding absolute allowance, and then calculate the content loss separately. - Abs and content duplication A double comparison is performed to generate the verification results, as follows: Absolute tolerance for conversion: loss of content - Abs and content duplication Perform independent comparisons to ensure that both are within acceptable limits; Generate clear decision signals for subsequent processing: If the verification passes, it is considered "verification passed"; otherwise, it is considered "verification failed" and an alarm or rollback mechanism is triggered.
6. A structured text segmentation and verification system based on logarithmic linear tolerance budgeting, characterized in that, The system includes: The text block data acquisition module is used to access the text blocks output by the external segmentation module, read and record the total length of the original document and the length information of each text block, and provide input for subsequent fault tolerance calculations; The log-linear tolerance calculation module is used to calculate the relative tolerance rate of the current document based on a log-linear mathematical model, so that the tolerance decreases smoothly with the document size, thus balancing the accuracy of short documents and the robustness of long documents. The boundary protection module is used to apply the lower limit of absolute loss and the upper limit of relative tolerance to the initially calculated tolerance rate, so as to prevent abrupt changes in tolerance at extremely small or extremely large document lengths. The loss and duplication measurement module is used to independently calculate the amount of content loss and content duplication introduced after text segmentation, and output them in absolute numerical form to ensure the accuracy of downstream tolerance comparison. The threshold comparison decision module is used to compare the calculated amount of content loss and content duplication with the allowable error adjusted by boundary protection, and to determine whether the segmentation quality passes or needs to be reviewed. The verification result routing module is used to route the "verification passed" text block to downstream components such as the question-answering system and vector index based on the decision result, and to trigger alarms or rollback mechanisms for "verification failed" cases, thus completing the entire verification pipeline.
7. The system according to claim 6, characterized in that, The log-linear tolerance calculation module is also used to calculate the log-linear tolerance rate using formula (1). （1） Where n is the length of the source document, n min Where T is the reference length, T0 is the reference tolerance, and s is the logarithmic decay slope. To calculate the tolerance rate.
8. The system according to claim 7, characterized in that, The boundary protection module is also used to calculate the initial tolerance rate. Next, boundary protection adjustments need to be performed to ensure that the final tolerance rate does not exceed the maximum relative tolerance limit C due to insufficient document length. rel It will not fall below the minimum absolute loss constant C due to excessive document length. abs This smooths the tolerance curve, avoids abrupt changes at extreme lengths, and ensures the stability and consistency of the segmentation verification results.
9. The system according to claim 8, characterized in that, The loss and duplication measurement module is also used to provide accurate input for subsequent tolerance comparison by independently calculating the absolute measurement values of content loss and content duplication, as follows: Using formula (2), calculate the content loss. - abs: When the sum of the lengths of all text blocks is less than the length of the original document, （2） Where n is the length of the original document. This is the sum of the lengths of all text blocks; Calculate the content repetition rate using formula (3). When the sum of the lengths of all text blocks is greater than the length of the original document, （3）。 10. The system according to claim 9, characterized in that, The threshold comparison decision module is used to replace the relative tolerance rate T calculated upstream. final Convert to the corresponding absolute allowance, and then calculate the content loss separately. - Abs and content duplication A double comparison is performed to generate the verification results, as follows: Absolute tolerance for conversion: loss of content - Abs and content duplication Perform independent comparisons to ensure that both are within acceptable limits; Generate clear decision signals for subsequent processing: If the verification passes, it is considered "verification passed"; otherwise, it is considered "verification failed" and an alarm or rollback mechanism is triggered.