An AI-based analysis report quality dynamic optimization method

Abstract

The application discloses an AI-based analysis report quality dynamic optimization method, which balances weights to build a user preference vector based on user historical report scoring behavior and user type description; utilizes multiple AI models to generate a candidate report set in parallel; calculates an objective score based on multiple quality dimensions for each candidate report in the candidate report set, and shows a candidate report abstract or key features to a user through an interactive interface to collect the user's subjective original score on the abstract; calculates the similarity value of the user preference vector and the report features for each candidate report; combines the similarity value and the user's subjective score to obtain an adjusted subjective score; and uses dynamic weights to fuse the objective score and the adjusted subjective score to obtain a fusion score. The application can realize three deployment values: a significant reduction in artificial proofreading cost, a systematic jump in user satisfaction index, and a reduction in the self-adaptive iteration period of the weight rule to the rate of business demand change.

Description

TECHNICAL FIELD

[0001] The application relates to the field of AI, in particular to an AI-based analysis report quality dynamic optimization method. BACKGROUND

[0002] The existing AI report generation system relies on unified standard to evaluate report quality, which has the following defects: first, individual needs are ignored, financial experts need rigorous data, and marketing personnel prefer creative analysis, but the existing technology cannot distinguish these differences, resulting in doctors receiving medical reports full of marketing terms and pilots seeing complex financial charts; second, the evaluation mechanism is rigid, and the system cannot adjust itself according to user feedback, for example, when investors start to focus on emerging markets, the old system still generates analysis of traditional industries according to historical standards; third, diversity is lost, and the system will fall into a fixed routine in long-term operation. Civil aviation marketing reports gradually become uniform promotion templates, losing the keen judgment of market changes, and improvement is urgently needed. SUMMARY

[0003] To solve the technical problems in the background art, the application provides an AI-based analysis report quality dynamic optimization method, which includes: S1, based on user historical report scoring behavior and user type description, a user preference vector is constructed by balancing weights; S2, a plurality of AI models are used to generate a candidate report set in parallel; S3, the objective score based on multiple quality dimensions is calculated for each candidate report in the candidate report set, and the candidate report abstract or key features are displayed to the user through an interactive interface, and the user's subjective original score on the abstract is collected; S4, for each candidate report, the similarity value of the user preference vector and the report features is calculated; the adjusted subjective score is obtained by combining the similarity value and the user's subjective score; the objective score and the adjusted subjective score are fused using a dynamic weight to obtain a fusion score; S5, the candidate report with the highest fusion score is selected as the final report output.

[0004] Further, it also includes: S6, calculating the arithmetic mean of the objective scores of all candidate reports at the current time as the average objective score; collecting the user's subjective score on the final report; updating the dynamic weight according to the difference between the subjective score and the average objective score, and applying the updated dynamic weight to the subsequent fusion score calculation.

[0005] Further, S2 specifically includes: S21, configuring the difference parameters of the plurality of AI models, the difference parameters including training data distribution difference and algorithm architecture difference; S22, in response to a report generation instruction, sending a parallel trigger signal to all AI models; S23, each AI model independently generates a report entity according to the input data, the report entity including text content and feature vector; S24, aggregating all report entities to form a candidate report set.

[0006] Further, S21 specifically comprises: S211, obtaining an algorithm architecture descriptor and a training data distribution descriptor of each AI model; S212, calculating an architecture difference value based on the algorithm architecture descriptor; S213, calculating a data distribution difference value based on the training data distribution descriptor; and S214, integrating the architecture difference value and the data distribution difference value to generate a difference parameter.

[0007] Further, S1 comprises: S11, extracting a user historical report scoring behavior feature; S12, extracting a user occupation type related description; S13, normalizing the user historical report scoring behavior feature; S14, vectorizing and encoding the user occupation type related description; S15, weighting and fusing the normalized user historical report scoring behavior feature and the vectorized and encoded user occupation type related description by a balance coefficient;

[0008] S16, generating a user preference vector; wherein a construction formula of the user preference vector is: ; represents the normalized user historical report scoring behavior feature; represents the vectorized and encoded user occupation type related description; is the balance coefficient.

[0009] Further, S3 comprises: S31, extracting an information coverage feature vector, a coherence feature vector, and a data criticism degree feature vector for each candidate report; S32, processing the information coverage feature vector to generate an information coverage score; S33, processing the coherence feature vector to generate a coherence score; S34, processing the data criticism degree feature vector to generate a data criticism degree score; S35, weighting and fusing the information coverage score, the coherence score, and the data criticism degree score based on a preset weight coefficient to generate an objective score; S36, collecting a user's subjective original score for the candidate report through an interactive interface; the interactive interface only displays a core feature vector or a text summary of the candidate report, and the time length is controlled within a range that can be quickly evaluated by the user, so as to efficiently collect subjective feedback.

[0010] S37, standardizing the subjective original score to generate a user subjective score.

[0011] Further, S4 comprises: S41, calculating a cosine similarity value between the user preference vector and a feature vector of each candidate report; S42, multiplying the cosine similarity value by a user subjective score to generate an adjusted subjective score; and S43, calculating a fusion score by a dynamic weight fusion formula, wherein the fusion score formula is: ; represents the objective score; represents the adjusted subjective score; represents the dynamic weight.

[0012] Further, S5 comprises: S51, performing descending order sorting on fusion scores of all reports in the candidate report set; S52, selecting the candidate report with the first fusion score as the final report; S53, sending the final report to the user terminal through the report output interface.

[0013] Further, S6 comprises: S61, calculating the arithmetic mean of objective scores of all reports in the candidate report set at the current time, to generate an average objective score; S62, collecting the subjective original score of the user on the final report; S63, performing standardization processing on the subjective original score to generate a user subjective score; S64, calculating the updated dynamic weight through a weight updating formula, wherein the updating formula is: ; represents the user subjective score; represents the average objective score; represents the dynamic weight before updating; represents the weight adjustment rate, wherein the weight adjustment rate is a preset constant parameter, the value range is set to be between 0 and 1, and is determined through experimental calibration or system administrator configuration; S65, storing the updated dynamic weight into a weight configuration library; S66, calling the dynamic weight in the weight configuration library in subsequent fusion score calculation.

[0014] The application further provides a non-transitory computer readable storage medium, which stores a computer program, and when the program is executed by a processor, the method for dynamically optimizing analysis report quality based on AI can be realized.

[0015] The application establishes a precise demand capturing mechanism by proposing a method for dynamically optimizing analysis report quality based on AI, significantly enhances the matching accuracy of reports and user needs (for example, in the aviation marketing scene, the adoption rate of route revenue analysis reports is greatly improved), and establishes a precise demand capturing mechanism; based on real-time feedback, the evaluation weight is dynamically adjusted, when the user continuously gives high score evaluation on the data rigor of the report, the system automatically improves the weight coefficient of the data rigor degree, and gives the system continuous evolution ability; by controlling the difference parameters of multiple models, the innovation index of single scene report is improved on the premise of maintaining the accuracy of core data, and a diversity guarantee mechanism is formed; finally, three deployment values are realized: artificial review cost is greatly compressed, user satisfaction index is systematically jumped, and weight rule self-adaptive iteration period is shortened to within the business demand change rate. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 The application provides a main flowchart of the method for dynamically optimizing analysis report quality based on AI;

[0017] Figure 2 A local flowchart of an AI-based analysis report quality dynamic optimization method proposed by the present application;

[0018] Figure 3 A local flowchart of an AI-based analysis report quality dynamic optimization method proposed by the present application;

[0019] Figure 4 A local flowchart of an AI-based analysis report quality dynamic optimization method proposed by the present application;

[0020] Figure 5 A local flowchart of an AI-based analysis report quality dynamic optimization method proposed by the present application;

[0021] Figure 6 A local flowchart of an AI-based analysis report quality dynamic optimization method proposed by the present application;

[0022] Figure 7 A local flowchart of an AI-based analysis report quality dynamic optimization method proposed by the present application;

[0023] Figure 8 A local flowchart of an AI-based analysis report quality dynamic optimization method proposed by the present application. DETAILED DESCRIPTION

[0024] REFERENCE Figures 1-8 The present application proposes an AI-based analysis report quality dynamic optimization method, which includes:

[0025] S1, based on user historical report scoring behavior and user type description, a user preference vector is constructed by balancing weights. Specifically, it includes:

[0026] S11, extract user historical report scoring behavior features. This step extracts key behavior patterns from user past scoring records, such as frequently focused report types or scoring preferences, to form a structured feature set, improving the accuracy of subsequent user preference modeling, avoiding report recommendation deviation caused by ignoring historical behavior, and making the system more in line with long-term user needs.

[0027] S12, extract user occupation type related description. This step collects user's occupation background information, such as financial analyst or medical practitioner label, and converts it into quantifiable description text, thereby enhancing the comprehensiveness of user portrait, reducing business scenario adaptation error, and better integrating professional characteristics into preference analysis.

[0028] S13, normalize the user historical report scoring behavior features. This step processes the user historical scoring features through a standardization algorithm, such as min-max scaling to eliminate dimensional differences and make the data distribution more uniform, thereby improving the comparability between different users and reducing the risk of weight imbalance caused by excessively large or small feature values.

[0029] S14, vectorize the user occupation type related description. This step converts the occupation description into a numerical vector through embedding coding technology, such as Word2Vec model processing text description, which converts abstract attributes into computable form, thereby increasing the feasibility of feature fusion and avoiding text ambiguity interference in preference modeling.

[0030] S15, balance coefficient weighted fusion of normalized user historical report scoring behavior features and vectorized user occupation type related description. This step dynamically adjusts the weight proportion of normalized features and vectorized description through a balance coefficient, such as adjusting the contribution of the two, to ensure balanced fusion of user behavior and occupation background, thereby improving the representativeness of the preference vector and reducing model bias caused by a single factor.

[0031] S16, generate a user preference vector. The construction formula of the user preference vector is: . represents the normalized user historical report scoring behavior features. represents the vectorized user occupation type related description. is the balance coefficient. This step generates the final user preference vector through the weighted result, such as integrating historical scoring and occupation coding data to form a unique identifier, so that the vector can be directly used for similarity matching, thereby improving the personalization degree of subsequent report recommendation and avoiding the neglect of user uniqueness by general models.

[0032] S2, generate a candidate report set using multiple AI models in parallel. Specifically, it includes:

[0033] S21, configure the difference parameters of multiple AI models, including training data distribution difference and algorithm architecture difference. Specifically, it includes:

[0034] S211, obtain the algorithm architecture descriptor and training data distribution descriptor of each AI model. This step quantitatively stores the model characteristics by obtaining the architecture descriptor (such as Transformer or RNN type) and data distribution descriptor (such as training set source distribution) of the AI model, thereby increasing the basis for difference parameter calculation and reducing the report singularity caused by model repetition.

[0035] S212, calculate the architecture difference value based on the algorithm architecture descriptor. This step calculates the difference value through the algorithm architecture description, such as comparing the number of layers or the parameter size difference, so as to convert the architecture characteristics into numerical indicators, improve the model diversity control accuracy, and reduce the risk of similar reports generated by homogeneous models.

[0036] S213, calculate the data distribution difference value based on the training data distribution descriptor. This step calculates the difference value through the training data distribution description, such as analyzing the data field coverage or sample balance, so as to make the data characteristics measurable, increase the innovation of the candidate report, and avoid the limitation of the report content caused by data bias.

[0037] S214, integrate the architecture difference value and the data distribution difference value to generate the difference parameter. This step generates a comprehensive parameter by integrating the difference values of architecture and data, such as weighted average method outputting a single indicator to guide the model selection strategy, so as to improve the parallel generation efficiency and reduce the waste of resources caused by invalid model calling.

[0038] S22, in response to the report generation instruction, send a parallel trigger signal to all AI models. This step triggers all AI models at the same time after receiving the generation instruction, such as broadcasting a signal to wake up the standby model, so as to start the calculation task in parallel, thereby shortening the report generation delay and avoiding the response lag caused by sequential execution.

[0039] S23, each AI model independently generates a report entity based on the input data, and the report entity contains text content and feature vector. This step outputs a complete report entity containing text and feature vector by processing input data independently for each model, such as LLM generating financial analysis content with vector field, so as to increase the diversity and structure of the candidate set content, thereby increasing the report selection range and avoiding the limitation of single output for user selection.

[0040] S24, aggregate all report entities to form a candidate report set. This step aggregates the report entities generated by the models to form a unified set, such as listing all report texts and features, so that subsequent evaluation can be processed in batches, thereby improving system throughput and reducing processing delay caused by data dispersion.

[0041] S3, calculate the objective score based on multiple quality dimensions for each candidate report in the candidate report set, and show the candidate report summary or key features to the user through the interactive interface, and collect the user's subjective original score on the summary. Specifically, it includes:

[0042] S31, extract the information coverage feature vector, the coherence feature vector and the data rigor feature vector for each candidate report. This step extracts the feature vector of the candidate report, covers the information completeness, logical coherence and data rigor dimensions, such as NLP tool analysis of keyword coverage, so that the multi-dimensional quality can be quantified, thereby increasing the objectivity of the evaluation and avoiding subjective judgment ignoring key defects.

[0043] S32, processing the information coverage feature vector to generate an information coverage score. This step generates a score by processing the coverage features, such as calculating the proportion of missing key information, converting it to a 0-1 score, so that the information completeness is quantifiable and comparable, thereby improving the report utility and reducing the decision-making errors caused by missing data.

[0044] S33, processing the coherence feature vector to generate a coherence score. This step generates a score by coherence features, such as analyzing the fluency of paragraph connection, assigning values based on the integrity of logical chain, so that the readability of the report is measurable, thereby increasing the user understanding efficiency and avoiding the cognitive burden caused by structural confusion.

[0045] S34, processing the data rigor feature vector to generate a data rigor score. This step generates a score by data rigor features, such as verifying the reliability of data reference, scoring according to the strength of evidence support, so that the report rigor is materialized, thereby reducing the probability of incorrect inference and avoiding the impact of unsupported statements on user trust.

[0046] S35, based on the preset weight coefficient, the information coverage score, the coherence score and the data rigor score are weighted and fused to generate an objective score. This step fuses the three-dimensional scores by preset weight, such as giving higher weight to coverage, outputs the comprehensive objective score, so that the quality evaluation takes into account multiple factors, thereby improving the fairness of the ruling and avoiding the misjudgment of high-quality reports dominated by a single dimension.

[0047] S36, collecting user's subjective original score on the candidate report through an interactive interface; the interactive interface only shows the core feature vector or text summary of the candidate report, and the time length is controlled within the range that the user can quickly evaluate, so as to efficiently collect subjective feedback. This step collects the user's original score through an interactive interface, such as a five-level scale to collect accuracy feedback, so that the subjective perception is directly input into the system, thereby increasing the user participation and reducing the optimization lag caused by feedback delay.

[0048] S37, standardizing the subjective original score to generate a user subjective score. This step standardizes the original score, such as normalizing to a unified dimension, eliminating individual scoring bias, so that the subjective data is comparable and reliable, thereby improving the accuracy of the fused score and avoiding the distortion of weight adjustment caused by abnormal values.

[0049] S4, for each candidate report, calculate the similarity value of the user preference vector and the report features. Combine the similarity value with the user subjective score to obtain the adjusted subjective score. Use dynamic weight to fuse the objective score and the adjusted subjective score to obtain the fusion score. Specifically, it includes:

[0050] S41, calculate the cosine similarity value of the user preference vector and the feature vector of each candidate report. This step calculates the cosine similarity of the user preference vector and the report feature vector, such as vector dot product divided by length, outputs 0-1 matching value, so that the personalized fit degree can be quantified, thereby increasing the relevance of recommendation and reducing irrelevant report interference in user decision-making.

[0051] S42, multiply the cosine similarity value by the user subjective score to generate the adjusted subjective score. This step generates the adjusted score by multiplying the similarity value by the user subjective score, such as amplifying the subjective score weight with high similarity, dynamically strengthening the preference influence, thereby improving the report satisfaction and avoiding the static method ignoring the user immediate feedback.

[0052] S43, calculate the fusion score by the dynamic weight fusion formula, wherein the fusion score formula is: . represents the objective score. represents the adjusted subjective score. represents the dynamic weight. This step fuses the objective score and the adjusted subjective score by the dynamic weight formula, such as balancing the contribution of both, outputs the final fusion value, so that the evaluation takes into account both quality and individuality, thereby reducing one-sided decision and increasing report acceptance rate.

[0053] S5, select the candidate report with the highest fusion score as the final report output. Specifically, it includes:

[0054] S51, sort the fusion scores of all reports in the candidate report set in descending order. This step sorts the fusion scores of all reports in descending order, such as using quicksort algorithm to process the score list, generates a sequence of advantages and disadvantages, so that the optimization process is orderly and efficient, thereby shortening the decision-making time and avoiding random selection to reduce the output quality.

[0055] S52, select the candidate report with the highest fusion score as the final report. This step selects the report with the highest index as the final output, such as the highest score item, to ensure optimal content delivery, thereby increasing user satisfaction and avoiding waste of resources by suboptimal reports.

[0056] S53, send the final report to the user terminal through the report output interface. This step sends the final report to the user terminal through the report interface, such as API transmission file data, so that the result is immediately available, thereby improving the smoothness of interaction and reducing the impact of delivery delay on user experience.

[0057] S6, calculate the arithmetic mean of all candidate report objective scores at the current time as the average objective score. Collect the user's subjective score for the final report. Update the dynamic weight according to the difference between the subjective score and the average objective score, and apply the updated dynamic weight to the subsequent fusion score calculation. Specifically, it includes:

[0058] S61, calculate the arithmetic mean of all report objective scores in the candidate report set at the current time, and generate the average objective score. This step generates a reference quality value by calculating the arithmetic mean of all report objective scores, such as sum divided by the number, so that the system state can be monitored, thereby increasing the basis for weight adjustment and avoiding isolated data misleading the optimization direction.

[0059] S62, collect the user's subjective raw score for the final report. This step makes the real-time feeling entry system by collecting the user's subjective raw score for the final report, such as interface pop-up collecting feedback, thereby reducing the feedback loss rate and ensuring the completeness of the closed-loop optimization data.

[0060] S63, standardize the subjective raw score to generate the user's subjective score. This step generates a comparable subjective score by standardizing the raw score, such as scaling to a fixed range, so that the data adapts to the update formula, thereby reducing noise interference and improving the accuracy of weight calculation.

[0061] S64, calculate the updated dynamic weight by the weight update formula, wherein the update formula is: . represents the user's subjective score. represents the average objective score. represents the dynamic weight before updating. represents the weight adjustment rate, wherein the weight adjustment rate is a preset constant parameter, with a value range set between 0 and 1, and determined by experimental calibration or system administrator configuration; for example, in the initial deployment stage, is set to 0.05 to balance the weight update speed and stability. The weight adjustment rate is a hyperparameter, which is preset when the system is initialized and does not need to be calculated in real time. The typical value is determined based on business scenario testing, such as in the aviation marketing report, to quickly respond to changes in user preferences. This step calculates the new dynamic weight by the weight update formula, such as based on the difference between the subjective score and the average objective score and the old weight adjustment, so that the parameter evolves dynamically, thereby increasing the adaptability of the system and avoiding the lag of rigid rules from user demand changes.

[0062] S65, store the updated dynamic weight to the weight configuration library. This step ensures consistency in subsequent calls and reduces the risk of data loss by storing the updated weight to the configuration library, such as a database, writing new lambda values, and making parameters persistent.

[0063] S66, call the dynamic weight in the weight configuration library in subsequent fusion score calculation. This step makes the optimization results effective in real time by calling the configuration library weight in subsequent fusion score calculation, such as reading the latest value and applying it to the formula, thereby increasing the system response agility and avoiding the reduction of decision quality caused by historical parameters.

[0064] The application also provides a non-transitory computer-readable storage medium storing a computer program, which, when executed by a processor, implements the AI-based analysis report quality dynamic optimization method.

[0065] The above is only a preferred specific embodiment of the present application, but the protection scope of the present application is not limited thereto, and any person skilled in the art can make equivalent replacements or changes within the technical range disclosed by the present application according to the technical solution and inventive concept of the present application, which should be covered within the protection scope of the present application.

Claims

1. An AI-based analysis report quality dynamic optimization method, characterized in that, Comprise: S1, based on user history report scoring behavior and user type description, construct user preference vector by balancing weight; S2, utilize multiple AI models to generate candidate report set in parallel; S3, calculate the objective score based on multiple quality dimensions for each candidate report in the candidate report set, and show the candidate report summary or key features to the user through the interactive interface, and collect the user's subjective original score of the candidate report; S4, for each candidate report, calculate the similarity value of the user preference vector and the report features; combine the similarity value with the user's subjective score to obtain the adjusted subjective score; Use dynamic weight to fuse objective score and adjusted subjective score to obtain fusion score; S5, select the candidate report with the highest fusion score as the final report output.

2. The AI-based analysis report quality dynamic optimization method of claim 1, wherein, Also include: S6, calculate the arithmetic mean of the objective scores of all candidate reports at the current time as the average objective score; Collect the user's subjective score of the final report; According to the difference between the subjective score and the average objective score, update the dynamic weight, and apply the updated dynamic weight to the subsequent fusion score calculation. 3.The AI-based analytical report quality dynamic optimization method of claim 1, wherein S2 specifically includes: S21, configure the difference parameters of multiple AI models, which include training data distribution difference and algorithm architecture difference; S22, in response to the report generation instruction, send a parallel trigger signal to all AI models; S23, each AI model independently generates a report entity based on the input data, which includes text content and feature vector; S24, aggregate all report entities to form a candidate report set. 4.The AI-based analytical report quality dynamic optimization method of claim 3, wherein, S21 specifically includes: S211, obtain the algorithm architecture descriptor and training data distribution descriptor of each AI model; S212, calculate the architecture difference value based on the algorithm architecture descriptor; S213, calculate the data distribution difference value based on the training data distribution descriptor; S214, integrate the architecture difference value and the data distribution difference value to generate the difference parameter. 5.The AI-based analytical report quality dynamic optimization method of claim 1, wherein, S1 includes: S11, extract user historical report scoring behavior features; S12, extract user occupation type related description; S13, normalize the user historical report scoring behavior features; S14, vectorize the user occupation type related description; S15, weighted fusion of normalized user historical report scoring behavior features and vectorized user occupation type related description through balance coefficient; S16, generate user preference vector; Wherein, the construction formula of the user preference vector is: ; Indicates the normalized user historical report scoring behavior characteristics; Indicates the vectorized and encoded user occupation type related description; is a balance coefficient. 6.The AI-based analysis report quality dynamic optimization method according to claim 1, wherein S3 Include: S31, extract information coverage feature vector, coherence feature vector and data criticism degree feature vector for each candidate report; S32, process the information coverage feature vector to generate the information coverage score; S33, process the coherence feature vector to generate the coherence score; S34, process the data criticism degree feature vector to generate the data criticism degree score; S35, weighted fusion of information coverage score, coherence score and data criticism degree score based on preset weight coefficient to generate objective score; S36, collect the user's subjective original score of the candidate report through the interactive interface; the interactive interface only shows the core feature vector or text summary of the candidate report, and the time length is controlled within the range that the user can quickly evaluate, so as to efficiently collect subjective feedback; S37, standardize the subjective original score to generate a user subjective score. 7.The AI-based analytical report quality dynamic optimization method of claim 1, wherein S4 Comprising: S41, calculate the cosine similarity value between the user preference vector and each candidate report feature vector; S42, multiply the cosine similarity value by the user subjective score to generate an adjusted subjective score; S43, calculate the fusion score through a dynamic weight fusion formula, wherein the fusion score formula is: ; denotes the objective score; denotes the adjusted subjective score; denotes the dynamic weight. 8.The AI-based analysis report quality dynamic optimization method of claim 1, wherein S5 Comprising: S51, sort the fusion scores of all reports in the candidate report set in descending order; S52, select the candidate report with the highest fusion score as the final report; S53, send the final report to the user terminal through the report output interface. 9.The AI-based analytical report quality dynamic optimization method of claim 2, wherein S6 Comprising: S61, calculate the arithmetic mean of the objective scores of all reports in the candidate report set at the current time to generate an average objective score; S62, collect the user's subjective original score for the final report; S63, standardize the subjective original score to generate a user subjective score; S64, calculate the updated dynamic weight through a weight updating formula, wherein the updating formula is: ; represents the user subjective score; represents the average objective score; represents the dynamic weight before updating; represents the weight adjustment rate, wherein the weight adjustment rate is a preset constant parameter, the value range is set to be between 0 and 1, and is determined through experimental calibration or system administrator configuration. S65, store the updated dynamic weight in the weight configuration library; S66, call the dynamic weight in the weight configuration library in subsequent fusion score calculation.

10. A non-transitory computer-readable storage medium, characterized in that, The computer program is stored, and when the program is executed by the processor, the AI-based analysis report quality dynamic optimization method of any one of claims 1-9 is realized.

Images