A data intelligent classification and grading method and system based on deep learning

Abstract

The present invention relates to the field of data processing technology, specifically to a method and system for intelligent data classification and grading based on deep learning, comprising the following steps: configuring a hybrid neural network, performing data feature recognition, extracting features that have a key impact on classification by weighted processing of features of input data, and forming a key feature weighted graph. In the present invention, by constructing a hybrid neural network and utilizing an autoencoder, fine data feature weighting and automatic classification hierarchical division are achieved, highlighting the optimization in the extraction and processing of key features, so that the accuracy of data analysis and the delicacy of processing are significantly improved, and by dynamically adjusting model parameters to adapt to changes in data streams, highly adaptive and real-time updating of data processing is achieved, and the introduction of iterative learning optimizes the model's ability to handle abnormal data, enhances the model's robustness and adaptability, and especially maintains the stability of classification performance when facing changes in data structure or new types of data.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a method and system for intelligent data classification and grading based on deep learning. Background Art

[0002] The field of data processing technology involves the entire process from data collection, storage, management to analysis and output. In modern computing, data processing is not limited to basic data input and output operations, but includes complex processing flows to extract valuable information from large amounts of data, using various algorithms and technical models, including machine learning and deep learning, to achieve data classification, prediction, optimization and decision support. Deep learning, as a core technology in data processing, relies on building, training and applying deep neural networks to identify complex patterns and relationships in data, and to achieve efficient and automated processing of large data sets.

[0003] Among them, the data intelligent classification and grading method based on deep learning refers specifically to the technology of automatically classifying and grading data using deep learning models. By building and training multi-layer neural networks, it is possible to learn the hidden features and structures in the data, so as to accurately classify the data. It has a wide range of uses, such as automatically identifying and classifying pathological images in medical diagnosis, or for risk assessment and customer stratification in financial services, as well as optimizing user experience and personalized recommendations in retail and e-commerce, which helps to improve decision-making efficiency and accuracy while reducing labor costs.

[0004] The existing conventional data processing process lacks the ability to quickly adapt to dynamic data flows, especially in terms of parameter adjustment and feature engineering, which often rely on manual intervention, resulting in low processing efficiency when the data volume increases and the types diversify. In terms of abnormal data processing and error rate control, traditional models lack effective automation mechanisms, which makes them prone to errors in the face of complex or unknown data, affecting the overall processing quality and decision-making accuracy. In addition, existing systems often fail to fully utilize the multi-level structural information of the data, resulting in limited classification accuracy and application flexibility, especially in areas that require fine classification decision support such as financial risk assessment or medical diagnosis, resulting in the omission of key information, affecting the reliability and practicality of the results. Summary of the invention

[0005] The purpose of the present invention is to solve the shortcomings of the prior art and to propose a data intelligent classification and grading method based on deep learning.

[0006] In order to achieve the above object, the present invention adopts the following technical solution: a data intelligent classification and grading method based on deep learning, comprising the following steps:

[0007] S1: Configure a hybrid neural network to identify data features, extract features that have a key impact on classification by weighting the features of the input data, and form a key feature weighted graph;

[0008] S2: Perform feature analysis on the key feature weighted graph, identify multi-level classification information in the data, automatically divide the data into corresponding classification levels, perform feature enhancement processing on the data in each category according to the division results, mark the key information, and create an initial hierarchical data classification and grading model;

[0009] S3: Based on the initial hierarchical data classification and grading model, the autoencoder calibration error data points are integrated, and the model is iteratively learned to optimize the error rate, the abnormal data features are re-encoded and adjusted according to the iterative learning results, the consistency of the classification results is tested, and the optimized data classification and grading model is generated;

[0010] S4: deploy the optimized data classification and grading model, monitor the data classification and grading effect in real time, dynamically adjust the model parameters according to the changes in data flow, continuously optimize the response speed and accuracy of data classification and grading, and output a real-time optimized data intelligent classification and grading solution.

[0011] As a further solution of the present invention, the step of obtaining the key feature weighted graph is:

[0012] S111: configuring a hybrid neural network, analyzing the input data features, preliminarily identifying the data features, and generating a data feature set;

[0013] S112: introducing an attention mechanism to generate a weighted data feature set by weighting the features in the data feature set;

[0014] S113: Based on the weighted data feature set, using the formula:

[0015]

[0016] Calculate the weighted mean of the features, filter the features that have a key impact on classification, and construct and output the key feature weighted graph, where: Representative The weight of the feature, Representative The value of the feature, Represents the total number of features.

[0017] As a further solution of the present invention, the step of automatically dividing the data is:

[0018] S211: Analyze the key feature weighted graph using a deep learning model, confirm feature combinations and associations in the weighted graph, identify potential data classification levels, and generate feature relationship records;

[0019] S212: automatically assigning the data to the identified classification hierarchy according to the feature relationship record, accurately classifying each data point according to its features, and generating a classification data set;

[0020] S213: Perform verification analysis on the classification data set, using the formula:

[0021]

[0022] Calculate the silhouette coefficient , evaluate the classification effect and generate structured classification hierarchy results, where It is The average distance between a data point and other points in the same category, It is The average distance between a data point and all points in the nearest neighbor class.

[0023] As a further solution of the present invention, the step of obtaining the initial hierarchical data classification and grading model is:

[0024] S221: Based on the structured classification hierarchy result, perform feature enhancement processing on each category, analyze the new feature weights of the data points in each category, and generate a feature-enhanced data set;

[0025] S222: marking key information according to the feature-enhanced data set, identifying outliers and key points in the data, and generating a marked data set;

[0026] S223: Using the labeled data set, design a multi-layer classification model to match the data characteristics of the difference level, using the formula,

[0027]

[0028] Calculate the overall classification efficiency of the model , output the initial hierarchical data classification and grading model, where is the number of categories, Yes Category The number of correctly classified data points in Yes Category The expected number of data points in is the total number of data points in all categories.

[0029] As a further solution of the present invention, the iterative learning steps of the model are:

[0030] S311: calling the initial hierarchical data classification and grading model to perform error correction and performance enhancement, optimizing classification accuracy through cyclic tuning, and outputting a pre-tuned data classification model;

[0031] S312: Based on the pre-tuned data classification model, the autoencoder is integrated to identify and correct the error data points, the data labels are adjusted by analyzing the deviation between the data points and the expected output, and the calibrated data set is output;

[0032] S313: Perform model iterative learning in combination with the calibrated data set, adjust network parameters, and output an updated model with minimized error.

[0033] As a further solution of the present invention, the step of obtaining the optimized data classification and grading model is:

[0034] S321: combining the updated model after minimizing the error, re-encoding the abnormal data features, adjusting the encoder weights and thresholds, matching the new data features, and obtaining an abnormal data re-encoding feature set;

[0035] S322: performing a classification consistency test based on the abnormal data re-encoding feature set, checking the consistency and accuracy of the new encoding features through cross-validation, evaluating the generalization ability and classification efficiency of the model, and obtaining a model evaluation and verification result;

[0036] S323: Integrate the model evaluation and verification results, perform integrated processing, upgrade the decision framework, and update and generate an optimized data classification and grading model.

[0037] As a further solution of the present invention, the step of obtaining the real-time optimized data intelligent classification and grading solution is:

[0038] S411: Initialize and deploy the optimized data classification and grading model, import the real-time data stream into the model, classify and grade the data, and generate a primary classification result of the real-time data;

[0039] S412: Based on the primary classification result of the real-time data, monitor the data classification effect, continuously track the change of the data flow, compare the deviation between the real-time and the preset performance standard, and obtain the deviation data information;

[0040] S413: According to the deviation data information, adjust the model parameters, using the formula:

[0041]

[0042] Calculate new model decision threshold , perform data classification and grading processing, and obtain real-time optimized data intelligent classification and grading solutions, among which, represents the old model decision threshold, represents the target decision threshold, Represents the average probability of the current model output, which indicates the average predicted probability of the current model on the test data set. Represents the adjustment factor.

[0043] A data intelligent classification and grading system based on deep learning, comprising:

[0044] The feature analysis module configures a hybrid neural network to preliminarily identify data features, assign weights to features, screen features that have a key impact on classification, and construct and output a weighted graph of key features;

[0045] The hierarchical parsing module uses a deep learning model to analyze the key feature weighted graph, identify potential data classification hierarchies, accurately classify each data point according to its characteristics, evaluate the classification effect, and generate a structured classification hierarchical result;

[0046] The multi-layer classification design module performs feature enhancement processing on each category based on the structured classification hierarchical results, performs key information marking, designs a multi-layer classification model, matches the data characteristics of the different hierarchical levels, and outputs an initial hierarchical data classification and grading model;

[0047] The performance enhancement module calls the initial hierarchical data classification and grading model to perform error correction and performance enhancement, integrates the autoencoder, identifies and corrects the error data points, performs model iterative learning, adjusts the network parameters, and outputs an updated model after minimizing the error;

[0048] The model optimization module combines the updated model after the error minimization, recodes the abnormal data features, verifies the consistency and accuracy of the new coded features through cross-validation, upgrades the decision framework, and updates and generates an optimized data classification and grading model;

[0049] The real-time monitoring module initializes the optimized data classification and grading model, imports real-time data streams, monitors data classification effects, continuously tracks changes in data streams, adjusts model parameters, performs data classification and grading processing, and obtains real-time optimized data intelligent classification and grading solutions.

[0050] Compared with the prior art, the advantages and positive effects of the present invention are:

[0051] In the present invention, by constructing a hybrid neural network and utilizing an autoencoder, fine data feature weighting and automatic classification hierarchical division are achieved, and the optimization in the extraction and processing of key features is highlighted, so that the accuracy of data analysis and the delicacy of processing are significantly improved. By dynamically adjusting model parameters to adapt to changes in data streams, high adaptability and real-time updating of data processing are achieved, and the response speed and accuracy are improved. The introduction of iterative learning optimizes the model's ability to handle abnormal data, enhances the model's robustness and adaptability, and especially maintains the stability of classification performance when facing changes in data structure or new types of data. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 It is a main step flow chart of the present invention;

[0053] Figure 2 The flowchart of obtaining the weighted graph of key features of the present invention is as follows;

[0054] Figure 3 The automatic division flow chart of data of the present invention;

[0055] Figure 4 It is a flow chart of obtaining the initial level data classification and grading model of the present invention;

[0056] Figure 5 It is an iterative learning flow chart of the model of the present invention;

[0057] Figure 6 This is a flow chart for obtaining the optimized data classification and grading model of the present invention;

[0058] Figure 7 A flow chart for obtaining the real-time optimized data intelligent classification and grading solution of the present invention. DETAILED DESCRIPTION

[0059] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0060] In the description of the present invention, it should be understood that the terms "length", "width", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, in the description of the present invention, "multiple" means two or more, unless otherwise clearly and specifically defined.

[0061] See also Figure 1 , a data intelligent classification and grading method based on deep learning, comprising the following steps:

[0062] S1: Configure a hybrid neural network to identify data features, extract features that have a key impact on classification by weighting the features of the input data, and form a key feature weighted graph;

[0063] S2: Perform feature analysis on the key feature weighted graph, identify multi-level classification information in the data, automatically divide the data into corresponding classification levels, perform feature enhancement processing on the data in each category according to the division results, mark the key information, and create an initial hierarchical data classification and grading model;

[0064] S3: Based on the initial hierarchical data classification and grading model, the autoencoder calibration error data points are integrated, and the model is iteratively learned to optimize the error rate. According to the iterative learning results, the abnormal data features are re-encoded and adjusted, the consistency of the classification results is tested, and the optimized data classification and grading model is generated;

[0065] S4: Deploy the optimized data classification and grading model, monitor the data classification and grading effect in real time, dynamically adjust the model parameters according to the changes in data flow, continuously optimize the response speed and accuracy of data classification and grading, and output real-time optimized data intelligent classification and grading solutions.

[0066] The key feature weighted graph includes feature weight records, key feature extraction results and feature graph visualization output records. The initial hierarchical data classification and grading model includes category division results, feature enhancement processing results and key information marking records. The optimized data classification and grading model includes error rate optimization results, abnormal data recoding records and classification consistency test results. The real-time optimized data intelligent classification and grading scheme includes parameter dynamic adjustment records and classification accuracy enhancement records.

[0067] See also Figure 2 ,The steps to obtain the key feature weighted graph are,

[0068] S111: configuring a hybrid neural network, analyzing the input data features, preliminarily identifying the data features, and generating a data feature set;

[0069] Configuring a hybrid neural network involves accurately selecting the type and number of network layers. The selection of network layers is based on the characteristics of the input data and the complexity of the required processing. By setting different types of network layers, including convolutional layers, pooling layers, or fully connected layers, and adjusting the number of nodes in each layer, the network's response to the input data is optimized. During implementation, special attention is paid to data characteristics such as size, type, and processing speed requirements. Network parameters are configured based on the characteristics, including the selection of learning rate and activation function, and how to effectively connect the layers to form an overall data processing flow. After configuration, the network gradually adjusts its weights through iterative training until it learns patterns that can accurately identify and classify input data. This process ensures that the network can effectively predict and classify new and unseen data.

[0070] S112: Introduce the attention mechanism to generate a weighted data feature set by assigning weights to the features in the data feature set;

[0071] The purpose of introducing the attention mechanism is to enhance the network's ability to respond to specific features in the input data, analyze the data feature set, and assign weights to each feature. The basis for weight assignment is the performance of each feature in the training data and the impact on the output results. During the implementation process, the weight of each feature is obtained through supervised learning, that is, the weight is adjusted by comparing the difference between the weighted output of the feature and the expected output. Through continuous training iterations, the model learns how to allocate appropriate attention to different features, so that the network output is more accurate. The final weighted feature set reflects the importance of each feature in solving specific problems.

[0072] S113: Based on the weighted data feature set, using the formula,

[0073]

[0074] Calculate the weighted mean of the features, filter the features that have a key impact on classification, and construct and output the key feature weighted graph, where: Representative The weight of the feature, Representative The value of the feature, represents the total number of features;

[0075] There is a feature set containing 3 features, where the feature weight , the corresponding eigenvalue .

[0076] Calculate the absolute value and square root of the weight and get:

[0077]

[0078]

[0079]

[0080] Compute the numerator part of the weighted eigenvalue:

[0081]

[0082]

[0083] At the same time, calculate the sum of the absolute values ​​of the weights as the denominator:

[0084]

[0085] Compute the weighted mean of the features:

[0086]

[0087] The results show that by considering the absolute value and square root of the weight to weight the eigenvalue, a comprehensive index that combines the weight influence and the eigenvalue importance is obtained. This index can be used to screen key classification influencing features (eigenvalue greater than is the key influencing feature), which reflects the contribution of the weighted feature set to the final output.

[0088] See also Figure 3 , the steps for automatic data partitioning are,

[0089] S211: Use deep learning models to analyze key feature weighted graphs, confirm feature combinations and associations in weighted graphs, identify potential data classification hierarchies, and generate feature relationship records;

[0090] The deep learning model is used to perform detailed feature analysis on the existing key feature weighted graph. The model draws a potential classification graph of the data by identifying the association and combination of features in the graph. The multi-layer feature abstraction capability of the convolutional neural network is used to learn complex feature relationships from the original data. After generating feature relationship records, the dependencies between data features and the influence of each feature on the final classification are listed in detail, providing a scientific basis for the next step of data processing. Through this method, the inherent structure and classification standards of the data can be accurately grasped, making data analysis more accurate and efficient.

[0091] S212: automatically assigning the data to the identified classification hierarchy based on the feature relationship records, accurately classifying each data point based on its features, and generating a classification data set;

[0092] Based on the feature relationship records, the data is automatically classified to the corresponding level, and the cluster center is optimized through statistical methods to enhance the similarity of data points within the class, while reducing the similarity between different classes, thereby improving the accuracy of classification. The distance from each data point to the cluster center is iteratively adjusted, and the intra-class aggregation degree of each iteration is evaluated to determine the final data classification. This method not only ensures that the data points are accurately classified, but also makes the entire classification structure clear. The data points in each category are closely connected, thus providing a high-quality classification data set.

[0093] S213: Perform validation analysis on the classification data set using the formula,

[0094]

[0095] Calculate the silhouette coefficient , evaluate the classification effect and generate structured classification hierarchy results, where It is The average distance between a data point and other points in the same category, It is The average distance between a data point and all points in the nearest neighbor category;

[0096] Identify data points The average distance to other data points in its class , and the average distance to all points in the nearest neighbor category is , according to the formula, calculate the cohesion and separation The difference between the two is then divided by the maximum value between the two. The specific calculation is as follows:

[0097]

[0098] This silhouette coefficient value shows that although the data points It is relatively close to the points in its class, but there is still room for improvement in the distance to the nearest other classes, because the closer the silhouette coefficient is to 1, the more reasonable the classification is. This result shows that the rationality of the classification can still be improved, which provides a quantitative basis for the subsequent optimization of the clustering algorithm and adjustment of the classification level.

[0099] See also Figure 4 ,The steps to obtain the initial hierarchical data classification and grading model are,

[0100] S221: Based on the structured classification hierarchy results, feature enhancement processing is performed on each category, new feature weights of data points in each category are analyzed, and a feature-enhanced data set is generated;

[0101] Principal component analysis (PCA) is used to perform feature enhancement processing on each category in the classification data set obtained above. By extracting and enhancing the features that have the greatest impact on the classification decision, the feature weights of the data points in each category are ensured to be accurately calculated. Through mathematical transformation, important features are emphasized, while insignificant features are compressed. Ultimately, a feature-enhanced data set with the potential to improve classification accuracy is generated. While maintaining the diversity of the original data, this data set enhances the model's sensitivity to key features, providing a basis for optimization and adjustment for subsequent data processing and model training.

[0102] S222: Mark key information based on the feature-enhanced data set, identify outliers and key points in the data, and generate a marked data set;

[0103] Based on the feature-enhanced data set, automatic labeling is used to accurately mark the key information in the data and identify the key data points and outliers in the data set. The automatic labeling process involves in-depth analysis of the behavioral patterns, statistical characteristics and abnormal indicators of the data points. The marked key data points include but are not limited to abnormal fluctuations, extreme values ​​and pattern deviations, ensuring that the data set can quickly identify and respond to key information in further analysis and decision making. The labeling results greatly improve the accuracy of data processing.

[0104] S223: Using the labeled data set, design a multi-layer classification model to match the data characteristics of the difference level, using the formula,

[0105]

[0106] Calculate the overall classification efficiency of the model , output the initial hierarchical data classification and grading model, where is the number of categories, Yes Category The number of correctly classified data points in Yes Category The expected number of data points in is the total number of data points in all categories;

[0107] In an actual data set, there are categories, and the actual and expected classification data points for each category are , , , , , , total number of data points ,So:

[0108]

[0109]

[0110] The results show that the classification model has a certain degree of accuracy and efficiency, but the accuracy is not high. By comparing the actual performance of data point classification with the expected performance, the performance of the model on specific categories and the potential optimization space can be seen.

[0111] See also Figure 5 , the iterative learning steps of the model are,

[0112] S311: calling the initial level data classification and grading model, performing error correction and performance enhancement, optimizing classification accuracy through cyclic tuning, and outputting a pre-tuned data classification model;

[0113] The first step of analysis is based on the existing initial hierarchical data classification and grading model, and preliminary verification and adjustment are carried out to ensure its high accuracy and stability in practical applications. Through the structure and parameter settings of the model, the response and processing mechanism of the model to different categories of data are analyzed in detail, and the deviations and errors in the classification process of the model are identified to ensure that the model can effectively integrate and utilize the automatic adjustment of errors by the autoencoder when processing new data sets. The detailed steps include the evaluation of the input-output relationship of the model, the feasibility analysis of internal parameter adjustment, and the adaptability test of the pre-processed data, which provide accurate basic data and operation guidelines for the next step of autoencoder integration.

[0114] S312: Based on the pre-tuned data classification model, the autoencoder is integrated to identify and correct the error data points, adjust the data labels by analyzing the deviation between the data points and the expected output, and output the calibrated data set;

[0115] On the basis of the adjusted model, the autoencoder is integrated to process the calibration error data points. The autoencoder effectively identifies and modifies the data points that deviate from the normal range through unsupervised learning. The specific process includes feature learning of error data points, identification and processing of outliers, and quality control after correction of data. It not only enhances the accuracy of the data, but also optimizes the overall quality of the data input model. Finally, the corrected data set is output, which will be directly applied to model training to reduce the error rate in the training process and improve the generalization ability of the model.

[0116] S313: Combine the calibrated data set, perform model iterative learning, adjust network parameters, and output an updated model with minimized error;

[0117] The model is iteratively learned using calibrated data. The error rate of the model is refined by adjusting the learning rate and optimization algorithm. Detailed processing includes monitoring each iteration of the learning process, analyzing the changing trend of the error, adjusting the weights and bias of the network, and verifying the performance of the model after each iteration to ensure that each iteration can significantly reduce the classification error. The final result is an optimized model with a lower error rate and higher data processing capabilities, which is suitable for complex and dynamically changing data classification tasks.

[0118] See also Figure 6 ,The steps to obtain the optimized data classification and grading model are:

[0119] S321: combining the updated model after error minimization, re-encoding the abnormal data features, adjusting the encoder weights and thresholds, matching the new data features, and obtaining the abnormal data re-encoding feature set;

[0120] Integrate the data of iterative learning, determine the density and distribution characteristics of abnormal data points through statistical analysis, apply the autoencoder to calibrate the data points, adjust the parameter settings of the encoder, ensure that each data point is expressed most accurately in the model, and adjust the parameters involved, including the learning rate of the encoder and the selection of the activation function. Through continuous trial and error comparison, find the best combination of parameters, thereby optimizing the error rate and learning efficiency of the overall model. The execution of this step ensures the accuracy and stability of the model when processing actual data.

[0121] S322: Re-encode the feature set according to the abnormal data, perform classification consistency test, verify the consistency and accuracy of the new encoding features through cross-validation, evaluate the generalization ability and classification efficiency of the model, and obtain the model evaluation and verification results;

[0122] The newly adjusted model is applied to conduct a comprehensive test on the consistency of data classification. The cross-validation method is used to test the generalization ability and classification accuracy of the model to ensure that all categories of data can be accurately identified and classified. The success rate and error rate of each classification are recorded in detail, and the results of different test sets are compared to confirm the effectiveness of the new feature encoding. This improves the adaptability and robustness of the model in diverse data environments, making the model more suitable for actual application scenarios.

[0123] S323: Integrate the model evaluation and verification results, perform integrated processing, upgrade the decision framework, and update and generate an optimized data classification and grading model;

[0124] Integrate the results of all modifications and tests, update the data classification and grading model to include all the latest feature adjustments and verification results, including formatting and standardizing the final output of the model to ensure that the output data can be directly used for subsequent applications or further analysis. The generated optimized model will have higher accuracy and better user experience. After the model is updated, conduct a final performance evaluation to ensure its stable operation and efficient performance under various conditions.

[0125] See also Figure 7 ,The steps for obtaining the real-time optimized data intelligent classification and grading scheme are,

[0126] S411: Initialize and deploy the optimized data classification and grading model, import the real-time data stream into the model, classify and grade the data, and generate the primary classification results of the real-time data;

[0127] When the model is initialized, real-time data streams are imported and preliminarily classified and graded, which involves data preprocessing, including data cleaning, normalization, and type calibration. For example, statistical analysis is used to identify and separate outliers, and cluster analysis is used to identify similar patterns in the data to ensure the quality and consistency of the data when it is input into the model. In addition, the algorithm for preliminary classification and grading of data needs to be selected according to the characteristics of the data. For example, for time series data, time window-based segmentation technology is used to identify different data stages. For classification tasks, decision trees or random forest models are used to stratify data to ensure that each type of data is correctly identified and graded according to preset standards. The final generated primary classification results provide a basis for subsequent analysis.

[0128] S412: Based on the primary classification results of real-time data, monitor the data classification effect, continuously track the changes in data flow, compare the deviation between real-time and preset performance standards, and obtain deviation data information;

[0129] Based on the preliminary classification results, monitoring the data classification and grading effects involves continuously tracking changes in data streams, using real-time data monitoring technology to capture changes in each data point through continuous comparison with preset performance standards, including monitoring the rate and quality of data streams, and using data anomaly detection technology to promptly identify data points that deviate from the normal range. These technologies allow us to update the accuracy and timeliness of data classification in real time, thereby achieving dynamic monitoring of classification effects. Through comparison, we can accurately identify data classification information that needs to be adjusted, provide a basis for adjusting model parameters, and ensure that the data classification system can adapt to the continuously changing data environment, thereby maintaining high-performance operation of the system.

[0130] S413: According to the deviation data information, adjust the model parameters, using the formula,

[0131]

[0132] Calculate new model decision threshold , perform data classification and grading processing, and obtain real-time optimized data intelligent classification and grading solutions, among which, represents the old model decision threshold, represents the target decision threshold, Represents the average probability of the current model output, which indicates the average predicted probability of the current model on the test data set. represents the adjustment factor;

[0133] There is a simple logistic regression model for a binary classification task where:

[0134] : The old decision threshold, initially set to 0.5, a common initial threshold, means that if the probability output by the model is greater than 0.5, the positive class is predicted.

[0135] : Target decision threshold, hoping that the model will make classification decisions with a higher degree of confidence.

[0136] : The current model outputs the average probability, which represents the average predicted probability of the current model on the test data set.

[0137] : Adjustment coefficient, used to control the sensitivity and rate of parameter adjustment.

[0138] Using the above parameter values, calculate the update decision threshold :

[0139]

[0140]

[0141] New decision threshold This means that the confidence required by the model to make a positive decision is slightly higher than before. This subtle adjustment helps the model to slightly improve the rigor of its decision while maintaining its original performance. Through such adjustments, the model can more accurately reflect real-time data changes and improve classification accuracy.

[0142] A data intelligent classification and grading system based on deep learning, comprising:

[0143] The feature analysis module configures a hybrid neural network to preliminarily identify data features, assign weights to features, screen features that have a key impact on classification, and construct and output a weighted graph of key features;

[0144] The hierarchical parsing module uses a deep learning model to analyze the key feature weighted graph, identify potential data classification hierarchies, accurately classify each data point based on its characteristics, evaluate the classification effect, and generate structured classification hierarchical results;

[0145] The multi-layer classification design module performs feature enhancement processing on each category based on the structured classification hierarchical results, marks key information, designs a multi-layer classification model, matches the data characteristics of different levels, and outputs the initial hierarchical data classification and grading model;

[0146] The performance enhancement module calls the initial hierarchical data classification and grading model to perform error correction and performance enhancement, integrates the autoencoder, identifies and corrects the error data points, performs model iterative learning, adjusts the network parameters, and outputs the updated model after the error is minimized;

[0147] The model optimization module combines the updated model after error minimization, recodes the abnormal data features, verifies the consistency and accuracy of the newly encoded features through cross-validation, upgrades the decision framework, and updates and generates an optimized data classification and grading model;

[0148] The real-time monitoring module initializes the optimized data classification and grading model, imports real-time data streams, monitors data classification effects, continuously tracks changes in data streams, adjusts model parameters, performs data classification and grading processing, and obtains real-time optimized data intelligent classification and grading solutions.

[0149] The above are only preferred embodiments of the present invention and are not intended to limit the present invention in other forms. Any technician familiar with the profession may use the technical contents disclosed above to change or modify them into equivalent embodiments with equivalent changes and apply them to other fields. However, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention without departing from the technical solution of the present invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. A data intelligent classification and grading method based on deep learning, characterized in that: The following steps are involved: Configure a hybrid neural network to identify data features, extract features that have a key impact on classification by weighting the features of the input data, and form a key feature weighted graph; Performing feature analysis on the key feature weighted graph, identifying multi-level classification information in the data, automatically dividing the data into corresponding classification levels, performing feature enhancement processing on the data in each category according to the division results, marking the key information, and creating an initial hierarchical data classification and grading model; Based on the initial hierarchical data classification and grading model, the autoencoder calibration error data points are integrated, and the model is iteratively learned to optimize the error rate, the abnormal data features are re-encoded and adjusted according to the iterative learning results, the consistency of the classification results is tested, and the optimized data classification and grading model is generated; Deploy the optimized data classification and grading model, monitor the data classification and grading effect in real time, dynamically adjust the model parameters according to the changes in data flow, continuously optimize the response speed and accuracy of data classification and grading, and output a real-time optimized data intelligent classification and grading plan; The steps of obtaining the key feature weighted graph are: Configure a hybrid neural network to analyze the input data features, preliminarily identify the data features, and generate a data feature set; Introducing an attention mechanism to generate a weighted data feature set by assigning weights to features in the data feature set; Based on the weighted data feature set, the formula is adopted: Calculate the weighted mean of the features, filter the features that have a key impact on classification, and construct and output the key feature weighted graph, where: Representative The weight of the feature, Representative The value of the feature, represents the total number of features; The steps of automatically dividing the data are as follows: Analyzing the key feature weighted graph using a deep learning model, confirming feature combinations and associations in the weighted graph, identifying potential data classification levels, and generating feature relationship records; According to the feature relationship records, the data is automatically assigned to the identified classification hierarchy, and each data point is accurately classified according to its features to generate a classified data set; A validation analysis was performed on the classification dataset using the formula, Calculate the silhouette coefficient , evaluate the classification effect and generate structured classification hierarchy results, where It is The average distance between a data point and other points in the same category, It is The average distance between a data point and all points in the nearest neighbor category; The steps for obtaining the initial hierarchical data classification and grading model are: Based on the structured classification hierarchical results, feature enhancement processing is performed on each category, new feature weights of data points in each category are analyzed, and a feature-enhanced data set is generated; Marking key information based on the feature-enhanced data set, identifying outliers and key points in the data, and generating a marked data set; Using the labeled data set, a multi-layer classification model is designed to match the data characteristics of the difference level, using the formula, Calculate the overall classification efficiency of the model , output the initial hierarchical data classification and grading model, where is the number of categories, Yes Category The number of correctly classified data points in Yes Category The expected number of data points in is the total number of data points in all categories; The iterative learning steps of the model are: Calling the initial level data classification and grading model to perform error correction and performance enhancement, optimizing classification accuracy through cyclic tuning, and outputting a pre-tuned data classification model; Based on the pre-tuned data classification model, an autoencoder is integrated to identify and correct the error data points, adjust the data labels by analyzing the deviation between the data points and the expected output, and output a calibrated data set; Combined with the calibrated data set, iterative model learning is performed to adjust network parameters and output an updated model with minimized error; The steps for obtaining the optimized data classification and grading model are: Combining the updated model after the error minimization, re-encoding the abnormal data features, adjusting the encoder weights and thresholds, matching the new data features, and obtaining the abnormal data re-encoding feature set; According to the abnormal data re-encoding feature set, a classification consistency test is performed, the consistency and accuracy of the new encoding features are checked through cross-validation, the generalization ability and classification efficiency of the model are evaluated, and the model evaluation and verification results are obtained; Integrate the model evaluation and verification results, perform integrated processing, upgrade the decision framework, and update and generate an optimized data classification and grading model; The steps for obtaining the real-time optimized data intelligent classification and grading scheme are: Initialize and deploy the optimized data classification and grading model, import the real-time data stream into the model, classify and grade the data, and generate real-time data primary classification results; Based on the primary classification results of the real-time data, monitor the data classification effect, continuously track the changes in data flow, compare the deviation between the real-time and preset performance standards, and obtain deviation data information; According to the deviation data information, the model parameters are adjusted, and the formula is adopted. Calculate new model decision threshold , perform data classification and grading processing, and obtain real-time optimized data intelligent classification and grading solutions, among which, represents the old model decision threshold, represents the target decision threshold, Represents the average probability of the current model output, which indicates the average predicted probability of the current model on the test data set. Represents the adjustment factor.

2. A data intelligent classification and grading system based on deep learning, characterized in that: The data intelligent classification and grading system based on deep learning is used to execute the data intelligent classification and grading method based on deep learning according to claim 1, comprising: The feature analysis module configures a hybrid neural network to preliminarily identify data features, assign weights to features, screen features that have a key impact on classification, and construct and output a weighted graph of key features; The hierarchical parsing module uses a deep learning model to analyze the key feature weighted graph, identify potential data classification hierarchies, accurately classify each data point according to its characteristics, evaluate the classification effect, and generate a structured classification hierarchical result; The multi-layer classification design module performs feature enhancement processing on each category based on the structured classification hierarchical results, performs key information marking, designs a multi-layer classification model, matches the data characteristics of the different hierarchical levels, and outputs an initial hierarchical data classification and grading model; The performance enhancement module calls the initial hierarchical data classification and grading model to perform error correction and performance enhancement, integrates the autoencoder, identifies and corrects the error data points, performs model iterative learning, adjusts the network parameters, and outputs an updated model after minimizing the error; The model optimization module combines the updated model after the error minimization, recodes the abnormal data features, verifies the consistency and accuracy of the new coded features through cross-validation, upgrades the decision framework, and updates and generates an optimized data classification and grading model; The real-time monitoring module initializes the optimized data classification and grading model, imports real-time data streams, monitors data classification effects, continuously tracks changes in data streams, adjusts model parameters, performs data classification and grading processing, and obtains real-time optimized data intelligent classification and grading solutions.

Images