A data classification method based on fitting a classification model and a user portrait

Abstract

The application provides a data classification method based on a fitting classification model and a user portrait, comprising the following steps: step 1, obtaining target data of a user, and dividing the target data into digital data and analog data according to the data type; step 2, classifying the digital data based on the fitting classification model to obtain a first classification result; step 3, classifying the analog data based on the user portrait to obtain a second classification result; and step 4, taking the first classification result and the second classification result as a final classification result of the target data. The method can realize more accurate classification of digital data and accurate classification of analog data of a non-digital data type.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a data classification method based on a fitted classification model and user profiles. Background Technology

[0002] Data classification, generally speaking, involves analyzing the attributes or features of a single data point to group it with other data possessing the same attributes or features, thus differentiating the data based on these attributes or features. In practical applications, the types of data requiring classification are increasing, going beyond simple categorization of numerical data. General data classification methods are no longer sufficient to accurately classify multi-type data. To achieve accurate classification of multi-type data, the first challenge is how to more accurately classify numerical data and how to accurately classify analog data (non-numerical data types). Summary of the Invention

[0003] This invention aims to at least partially solve one of the technical problems in the aforementioned technologies. Therefore, the objective of this invention is to propose a data classification method based on a fitted classification model and user profiles, aiming to achieve more accurate classification of digital data and accurate classification of analog data of non-digital data types.

[0004] To achieve the above objectives, embodiments of the present invention propose a data classification method based on a fitted classification model and user profiles, comprising:

[0005] Step 1: Obtain the user's target data and categorize it into numerical data and analog data based on data type;

[0006] Step 2: Based on the fitted classification model, classify the numerical data to obtain the first classification result;

[0007] Step 3: Based on user profiles, classify the simulated data to obtain the second classification results;

[0008] Step 4: Use the first classification result and the second classification result as the final classification result for the target data.

[0009] According to some embodiments of the present invention, the simulation data includes graphic data, text data, symbol data, and image data.

[0010] According to some embodiments of the present invention, step 2 includes:

[0011] Input the numerical data into the fitted classification model;

[0012] The fitting classification model is used to fit digital data using a nonlinear least squares fitting method to obtain several fitting classification curves.

[0013] The numerical data corresponding to each fitted classification curve is used as sub-classification data;

[0014] Based on several sub-classification data, the first classification result is obtained.

[0015] According to some embodiments of the present invention, step 3 includes:

[0016] Extract the feature information from the simulation data;

[0017] The feature information is matched with several tags in the user profile;

[0018] The simulated data corresponding to the successfully matched feature information is used as the tag class data for the matched tags;

[0019] The aforementioned labeled data are used as the second classification result.

[0020] According to some embodiments of the present invention, the training method for the fitted classification model includes:

[0021] Construct an initial fitted classification model and obtain several sets of sample numerical data and corresponding data classification results;

[0022] A set of sample numerical data is input into the initial fitted classification model, and the actual data classification result is output. The actual data classification result is compared with the numerical classification result corresponding to the set of sample numerical data. The model parameters of the initial fitted classification model are adjusted according to the comparison result to obtain the corrected fitted classification model.

[0023] Input another set of sample numerical data into the corrected fitted classification model, and continue iterative training until the output actual data classification result is consistent with the corresponding data classification result, thus obtaining the trained fitted classification model.

[0024] According to some embodiments of the present invention, the method for generating the user profile includes:

[0025] Obtain the user's data source;

[0026] Based on the data source, several category labels are determined;

[0027] User profiles are built based on tags from several categories.

[0028] According to some embodiments of the present invention, step 1, obtaining the user's target data, includes:

[0029] Based on several data acquisition interfaces, several sets of data to be processed are obtained, and a data cleaning task is generated.

[0030] Construct several processing processes and dynamically allocate data cleaning tasks based on processing load;

[0031] The processing process handles the assigned data cleaning tasks to obtain the target data.

[0032] According to some embodiments of the present invention, the processing process processes the assigned data cleaning task to obtain target data, including:

[0033] During the processing, determine the type of data to be processed corresponding to the data cleaning task;

[0034] When the data is determined to be unstructured, quantization is performed, and the first cleaning rule is determined based on the quantized data to be processed.

[0035] During the cleaning process based on the first cleaning rule, the data cleaning progress is obtained and compared with a preset threshold. The first cleaning rule is adjusted according to the comparison result, a second cleaning rule is generated, and a second data cleaning is performed. The data obtained from the first and second data cleaning are used as the target data.

[0036] According to some embodiments of the present invention, when the data is determined to be of the unstructured type, quantization processing is performed, and a first cleaning rule is determined based on the quantized data to be processed, including:

[0037] When the data type is determined to be unstructured, quantization processing is performed, and quantization features are extracted from the quantized data to obtain corresponding quantization feature information. The quantization feature information is compared with the quantization feature information of historical processed data in the data cleaning model included in the processing process. When the comparison is successful, the data cleaning rule of the historical processed data that was successfully compared is called as the first cleaning rule. When the comparison fails, a corresponding data cleaning rule is generated based on the quantization feature information, and it is added to the data cleaning model library of the data cleaning model.

[0038] According to some embodiments of the present invention, it further includes:

[0039] The final classification results are stored on a data server;

[0040] Based on the data servers currently running in the network, obtain historical traffic logs; parse the packets in the first traffic data included in the historical traffic logs and perform sensitive data detection;

[0041] Based on the detection results, generate sensitive data behavior standards;

[0042] The system detects the access status of preset traffic transmission nodes in the data server and obtains the corresponding sessions; records the traffic behavior data of the sessions and obtains the current traffic log; performs sensitivity analysis on the second traffic data included in the current traffic log based on the sensitive data behavior standard; and executes preset countermeasures and issues warning information when it is determined that there is a risk of data leakage.

[0043] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the written description and the accompanying drawings.

[0044] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0045] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0046] Figure 1 This is a flowchart of a data classification method based on a fitted classification model and user profile according to an embodiment of the present invention;

[0047] Figure 2 This is a flowchart of step 2 according to an embodiment of the present invention;

[0048] Figure 3 This is a flowchart of step 3 according to an embodiment of the present invention. Detailed Implementation

[0049] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0050] like Figure 1 As shown, this embodiment of the invention provides a data classification method based on a fitted classification model and user profiles, including:

[0051] Step 1: Obtain the user's target data and categorize it into numerical data and analog data based on data type;

[0052] Step 2: Based on the fitted classification model, classify the numerical data to obtain the first classification result;

[0053] Step 3: Based on user profiles, classify the simulated data to obtain the second classification results;

[0054] Step 4: Use the first classification result and the second classification result as the final classification result for the target data.

[0055] The working principle of the above technical solution is as follows: Step 1: Obtain the user's target data and classify it into digital data and analog data according to the data type; the target data is the data to be classified; this facilitates obtaining effective data to be classified before data classification; Step 2: Classify the digital data based on the fitted classification model to obtain the first classification result; this facilitates the reasonable classification of digital data according to the fitted model; Step 3: Classify the analog data based on the user profile to obtain the second classification result; this facilitates the classification of analog data that is not suitable for classification by the fitted classification model according to the user profile; Step 4: Use the first classification result and the second classification result as the final classification result for the target data. This facilitates the classification of different types of data and obtains reliable data classification results.

[0056] The beneficial effects of the above technical solution are as follows: By determining the basis for data classification according to data type, the accuracy of digital data classification is improved. Through user profiling, analog data can be easily classified, ultimately enabling the rapid and accurate classification of both digital and non-digital analog data.

[0057] According to some embodiments of the present invention, the simulation data includes graphic data, text data, symbol data, and image data.

[0058] The working principle and beneficial effects of the above technical solution are as follows: The simulated data includes graphic data, text data, symbol data, and image data. This facilitates the classification of non-numerical data, achieving diversified data classification.

[0059] like Figure 2 As shown, according to some embodiments of the present invention, step 2 includes:

[0060] Step 2-1: Input the numerical data into the fitted classification model;

[0061] Step 2-2: Using the aforementioned fitting classification model, fit the digital data using a nonlinear least squares fitting method to obtain several fitting classification curves;

[0062] Steps 2-3: Take the numerical data corresponding to each fitted classification curve as sub-classification data;

[0063] Steps 2-4: Obtain the first classification result based on several sub-classification data.

[0064] The working principle of the above technical solution is as follows: Step 2-1: Input the digital data into the fitting classification model; Step 2-2: Fit the digital data using the fitting classification model based on the nonlinear least squares method to obtain several fitting classification curves; this facilitates the classification of the digital data according to the fitting classification curves; Step 2-3: Use the digital data corresponding to each fitting classification curve as sub-classification data; Step 2-4: Obtain the first classification result based on the several sub-classification data. For example, if there are 5 digital data corresponding to a fitting classification curve, then these 5 digital data are considered as a sub-classification result.

[0065] The beneficial effects of the above technical solution are: by fitting a classification model, it is convenient to classify digital data using a nonlinear least squares fitting method, thereby improving the accuracy and efficiency of digital data classification.

[0066] like Figure 3 As shown, according to some embodiments of the present invention, step 3 includes:

[0067] Step 3-1: Extract the feature information of the simulated data;

[0068] Step 3-2: Match the feature information with several tags in the user profile;

[0069] Step 3-3: Use the simulated data corresponding to the successfully matched feature information as the tag class data for the matching tags;

[0070] Steps 3-4: Use the aforementioned label data as the second classification result.

[0071] The working principle of the above technical solution is as follows: Step 3-1: Extract the feature information of the simulated data; the feature information includes the source information, data content, and data attributes of the simulated data; this facilitates matching user profiles using the feature information; Step 3-2: Match the feature information with several tags in the user profile; Step 3-3: Use the simulated data corresponding to the successfully matched feature information as the tag class data of the matched tags; this facilitates data classification of the simulated data according to the tags in the user profile based on the feature information; Step 3-4: Use the several tag class data as the second classification result. This facilitates obtaining the data classification result of the simulated data; for example, a picture of a teddy bear, after being classified by the user profile, is classified into the tag class data labeled "toy".

[0072] The beneficial effects of the above technical solution are: by classifying simulated data through user profiles, it is easier to achieve diversified data classification, obtain classification results for different non-numerical data, and improve the reference value of data classification in practical applications.

[0073] According to some embodiments of the present invention, the training method for the fitted classification model includes:

[0074] Construct an initial fitted classification model and obtain several sets of sample numerical data and corresponding data classification results;

[0075] A set of sample numerical data is input into the initial fitted classification model, and the actual data classification result is output. The actual data classification result is compared with the numerical classification result corresponding to the set of sample numerical data. The model parameters of the initial fitted classification model are adjusted according to the comparison result to obtain the corrected fitted classification model.

[0076] Input another set of sample numerical data into the corrected fitted classification model, and continue iterative training until the output actual data classification result is consistent with the corresponding data classification result, thus obtaining the trained fitted classification model.

[0077] The working principle of the above technical solution is as follows: An initial fitting classification model is constructed, and several sets of sample digital data and corresponding data classification results are obtained. A set of sample digital data is input into the initial fitting classification model, and the actual data classification result is output. The actual data classification result is compared with the corresponding digital classification result of the set of sample digital data. Based on the comparison result, the model parameters of the initial fitting classification model are adjusted to obtain a corrected fitting classification model. This facilitates obtaining the most suitable fitting classification model. Another set of sample digital data is input into the corrected fitting classification model, and iterative training continues until the output actual data classification result matches the corresponding data classification result, resulting in a well-trained fitting classification model. This facilitates training the most accurate and suitable fitting classification model by continuously adjusting the model parameters.

[0078] The beneficial effects of the above technical solution are as follows: by inputting sample data and continuously adjusting the model parameters based on the actual classification results and data classification results, it is easier to train the most suitable fitting classification model, and the fitting classification model helps to obtain more accurate classification results for digital data.

[0079] According to some embodiments of the present invention, the method for generating the user profile includes:

[0080] Obtain the user's data source;

[0081] Based on the data source, several category labels are determined;

[0082] User profiles are built based on tags from several categories.

[0083] The working principle and beneficial effects of the above technical solution are as follows: It obtains user data sources; facilitates the generation of rich user data for user profiles; determines several category labels based on the data sources; facilitates the classification of simulated data according to the labels; constructs user profiles based on the several category labels; and facilitates the obtaining of more comprehensive user profiles. By obtaining user data sources, it is easier to acquire rich user data, improve the comprehensiveness of user profiles, and thus enhance the reliability of using user profiles to classify simulated data.

[0084] According to some embodiments of the present invention, step 1, obtaining the user's target data, includes:

[0085] Based on several data acquisition interfaces, several sets of data to be processed are obtained, and a data cleaning task is generated.

[0086] Construct several processing processes and dynamically allocate data cleaning tasks based on processing load;

[0087] The processing process handles the assigned data cleaning tasks to obtain the target data.

[0088] The working principle of the above technical solution is as follows: Based on several data acquisition interfaces, several sets of data to be processed are obtained, and data cleaning tasks are generated; this facilitates data acquisition from multiple interfaces; several processing processes are constructed, and data cleaning tasks are dynamically allocated based on processing load; this facilitates more efficient allocation of data cleaning tasks; the processing processes process the allocated data cleaning tasks to obtain the target data. This also facilitates the deletion of duplicate, erroneous, and incomplete data from the data to be processed.

[0089] The beneficial effects of the above technical solution are: dynamically allocating data cleaning tasks, which facilitates improved data cleaning efficiency; and by cleaning the data to be processed, duplicate, erroneous, and incomplete data can be removed, ensuring the integrity and reliability of the target data, as well as the reliability of the data classification results.

[0090] According to some embodiments of the present invention, the processing process processes the assigned data cleaning task to obtain target data, including:

[0091] During the processing, determine the type of data to be processed corresponding to the data cleaning task;

[0092] When the data is determined to be unstructured, quantization is performed, and the first cleaning rule is determined based on the quantized data to be processed.

[0093] During the cleaning process based on the first cleaning rule, the data cleaning progress is obtained and compared with a preset threshold. The first cleaning rule is adjusted according to the comparison result, a second cleaning rule is generated, and a second data cleaning is performed. The data obtained from the first and second data cleaning are used as the target data.

[0094] The working principle of the above technical solution is as follows: During the processing, the type of data to be processed corresponding to the data cleaning task is determined; when the type is determined to be unstructured data, quantization processing is performed, and the first cleaning rule is determined based on the quantized data to be processed; since feature extraction from unstructured data is relatively difficult, quantization is performed on unstructured data to easily obtain its feature information; the quantization processing includes the quantization of three parts: general attribute category quantization description, image attribute quantization description, and semantic attribute quantization description; the general attribute category quantization description includes general attribute category number, general attribute category feature item, and general attribute category feature data content; the image attribute quantization description includes image attribute category label, image attribute feature name, image attribute feature data structure, and image attribute feature data content; the semantic attribute quantization description includes semantic attribute category number, semantic attribute category feature category, and semantic attribute category keyword; this facilitates the extraction of data feature information from unstructured data; during the cleaning process based on the first cleaning rule, the data cleaning progress is obtained, and the corresponding data cleaning rate is obtained based on the data cleaning progress;

[0095] When the difference between the data cleaning rate and the preset cleaning rate is greater than a threshold, the first cleaning rule is adjusted, a second cleaning rule is generated, and a second data cleaning is performed; this facilitates improved data cleaning efficiency; the data obtained from the first and second data cleaning are used as the target data; the first cleaned data also includes data that has been cleaned before the second data cleaning is performed; this facilitates obtaining data after data cleaning to remove "dirty" data, and using it as effective target data.

[0096] The beneficial effects of the above technical solution are as follows: by determining the type of data, quantitative feature processing is performed on unstructured data, which facilitates the acquisition of the first cleaning rule for cleaning unstructured data; by judging and adjusting the first cleaning rule, a second cleaning rule is generated, which helps to improve the efficiency of data cleaning and also improves the efficiency of data classification.

[0097] According to some embodiments of the present invention, when the data is determined to be of the unstructured type, quantization processing is performed, and a first cleaning rule is determined based on the quantized data to be processed, including:

[0098] When the data type is determined to be unstructured, quantization processing is performed, and quantization features are extracted from the quantized data to obtain corresponding quantization feature information. The quantization feature information is compared with the quantization feature information of historical processed data in the data cleaning model included in the processing process. When the comparison is successful, the data cleaning rule of the historical processed data that was successfully compared is called as the first cleaning rule. When the comparison fails, a corresponding data cleaning rule is generated based on the quantization feature information, and it is used as the first cleaning rule and added to the data cleaning model library of the data cleaning model.

[0099] The working principle of the above technical solution is as follows: When the data type is determined to be unstructured, quantization processing is performed, and quantization features are extracted from the quantized data to obtain the corresponding quantization feature information; this facilitates the acquisition of the first cleaning rule for data cleaning of unstructured data; for structured data, data features are directly extracted; the quantization feature information is compared with the quantization feature information of historical processed data in the data cleaning model included in the processing process. When the comparison is successful, the data cleaning rule of the successfully compared historical processed data is called as the first cleaning rule; this facilitates the rapid determination of the data cleaning rule; when the comparison fails, the quantization feature information is used as the first cleaning rule. The process involves: determining the attribute category information of the quantified feature information; establishing a corresponding Bayesian network based on the dependency relationship between the quantified feature information, the attribute category information, and the preset cleaning type of unstructured data; training the Bayesian network to obtain a standard cloud distribution; generating a cloud distribution of the unstructured data based on the unstructured data and the Bayesian network; comparing the standard cloud distribution with the cloud distribution of the unstructured data to determine the cleaning category of the unstructured data; and adding the preset cleaning rule corresponding to the cleaning category as the first cleaning rule to the data cleaning model library of the data cleaning model.

[0100] The beneficial effects of the above technical solution are: it facilitates the rapid determination of data cleaning rules based on data type and data cleaning model, thereby improving the efficiency of data cleaning and shortening the time required for data classification.

[0101] According to some embodiments of the present invention, it further includes:

[0102] The final classification results are stored on a data server;

[0103] Based on the data servers currently running in the network, obtain historical traffic logs; parse the packets in the first traffic data included in the historical traffic logs and perform sensitive data detection;

[0104] Based on the detection results, generate sensitive data behavior standards;

[0105] The system detects the access status of preset traffic transmission nodes in the data server and obtains the corresponding sessions; records the traffic behavior data of the sessions and obtains the current traffic log; performs sensitivity analysis on the second traffic data included in the current traffic log based on the sensitive data behavior standard; and executes preset countermeasures and issues warning information when it is determined that there is a risk of data leakage.

[0106] The working principle of the above technical solution is as follows: The final classification result is stored in a data server, facilitating user access to the data; historical traffic logs are obtained based on the data server currently running in the network; packets in the first traffic data included in the historical traffic logs are parsed, and sensitive data detection is performed; this facilitates determining the status of sensitive data in packets of traffic data under historical normal conditions by detecting sensitive data; the method for determining sensitive data includes: determining the industry type to which the classification result data belongs based on the classification result data; matching the industry type with industry types in a preset industry database; and determining the corresponding sensitive data based on the matched industry type and the correspondence between the industry types in the industry database and sensitive industry data.

[0107] Based on the detection results, sensitive data behavior standards are generated. These standards define the basic user operations on sensitive data under normal historical conditions. The system detects access to preset traffic transmission nodes in the data server and obtains corresponding sessions. This facilitates tracking current traffic behavior data. The system records the traffic behavior data of each session to obtain the current traffic log. Based on the sensitive data behavior standards, a sensitivity analysis is performed on the second traffic data included in the current traffic log. This facilitates judgment by comparing the current user's basic operations on sensitive data with those performed under normal historical conditions. When the number of times the current user accesses sensitive data exceeds the maximum access limit for sensitive data under normal historical conditions, a data leakage risk is identified, preset countermeasures are executed, and a warning message is issued. The warning message includes the sensitive data at risk of leakage and the corresponding leakage pathways.

[0108] The beneficial effects of the above technical solution are as follows: by storing the final classification results on a data server and then detecting the risk of data leakage, it is easier to improve data security, ensure the absolute privacy of the user's data, and at the same time improve the practical value of the data classification results.

[0109] According to some embodiments of the present invention, a method for determining sensitive data includes:

[0110] The classification results data are divided into terms, and the weight Q of each term is calculated. (x) :

[0111]

[0112] Among them, Q (x) t represents the weight coefficient of term x; x c(W) represents the number of times term x appears in the classification result data; m ) refers to the preset sensitive file W included in the classification result data. m Quantity; c(W) n(x) ) refers to the preset non-sensitive file W included in the classification result data. n The number of occurrences of term x in the middle; ∑ X∈W c(X) represents the total number of all terms in the classification result data; X represents all terms in the classification result data.

[0113] The preset sensitivity value J is combined with the weight Q of the term. (x) Comparison, when Q (x) When the value is greater than J, the corresponding term is identified as sensitive data.

[0114] The working principle of the above technical solution is as follows: The classification result data is divided into terms, and the weight Q of each term is calculated. (x) The preset sensitivity value J is combined with the weight Q of the term. (x) Comparison, when Q (x) When the value is greater than J, the corresponding term is identified as sensitive data.

[0115] The beneficial effects of the above technical solution are as follows: by calculating the weight of terms to determine sensitive data, the accuracy of identifying sensitive data is improved, thereby ensuring the accuracy of data leakage risk assessment.

[0116] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A data classification method based on a fitted classification model and user profiles, comprising: Step 1: Obtain the user's target data and categorize it into numerical data and analog data based on data type; Step 2: Based on the fitted classification model, classify the numerical data to obtain the first classification result; Step 3: Based on user profiles, classify the simulated data to obtain the second classification results; Step 4: Use the first classification result and the second classification result as the final classification result for the target data; In step 1, the target data of the user is obtained, including: Based on several data acquisition interfaces, several sets of data to be processed are obtained, and a data cleaning task is generated. Construct several processing processes and dynamically allocate data cleaning tasks based on processing load; The processing process handles the assigned data cleaning tasks to obtain the target data; When the data type is determined to be unstructured, quantization is performed. Based on the quantized data to be processed, the first cleaning rule is determined, including: When the data type is determined to be unstructured, quantization processing is performed, and quantization features are extracted from the quantized data to obtain corresponding quantization feature information. The quantization feature information is compared with the quantization feature information of historical processed data in the data cleaning model included in the processing process. When the comparison is successful, the data cleaning rule of the historical processed data that was successfully compared is called as the first cleaning rule. When the comparison fails, a corresponding data cleaning rule is generated based on the quantization feature information, and it is used as the first cleaning rule and added to the data cleaning model library of the data cleaning model. Also includes: The final classification results are stored on a data server; Based on the data servers currently running in the network, obtain historical traffic logs; parse the packets in the first traffic data included in the historical traffic logs and perform sensitive data detection; Based on the detection results, sensitive data behavior standards are generated; the sensitive data behavior standards are the basic operations of users on sensitive data under historical normal conditions. The system detects the access status of preset traffic transmission nodes in the data server and obtains the corresponding sessions; records the traffic behavior data of the sessions and obtains the current traffic log; performs sensitivity analysis on the second traffic data included in the current traffic log based on the sensitive data behavior standard; and executes preset countermeasures and issues warning information when a data leakage risk is determined. Methods for identifying sensitive data include: The classification results data are divided into terms, and the weights of the terms are calculated. : ； in, For terms Weighting coefficients; For terms The number of times it appears in the classification result data; The pre-defined sensitive files included in the classification result data. Quantity; The pre-defined non-sensitive files included in the classification result data Chinese term The number of occurrences; The total number of all terms in the classification result data; For all terms in the classification result data; The preset sensitivity value J is compared with the weight of the term. When comparing, When the value is greater than J, the corresponding term is identified as sensitive data.

2. The data classification method based on a fitted classification model and user profile as described in claim 1, characterized in that, The simulation data includes graphic data, text data, symbol data, and image data.

3. The data classification method based on a fitted classification model and user profile as described in claim 1, characterized in that, Step 2 includes: Input the numerical data into the fitted classification model; The fitting classification model is used to fit digital data using a nonlinear least squares fitting method to obtain several fitting classification curves. The numerical data corresponding to each fitted classification curve is used as sub-classification data; Based on several sub-classification data, the first classification result is obtained.

4. The data classification method based on a fitted classification model and user profile as described in claim 1, characterized in that, Step 3 includes: Extract the feature information from the simulation data; The feature information is matched with several tags in the user profile; The simulated data corresponding to the successfully matched feature information is used as the tag class data for the matched tags; The aforementioned labeled data are used as the second classification result.

5. The data classification method based on a fitted classification model and user profile as described in claim 1, characterized in that, The training method for the fitted classification model includes: Construct an initial fitted classification model and obtain several sets of sample numerical data and corresponding data classification results; A set of sample numerical data is input into the initial fitted classification model, and the actual data classification result is output. The actual data classification result is compared with the numerical classification result corresponding to the set of sample numerical data. The model parameters of the initial fitted classification model are adjusted according to the comparison result to obtain the corrected fitted classification model. Input another set of sample numerical data into the corrected fitted classification model, and continue iterative training until the output actual data classification result is consistent with the corresponding data classification result, thus obtaining the trained fitted classification model.

6. The data classification method based on a fitted classification model and user profile as described in claim 1, characterized in that, The method for generating the user profile includes: Obtain the user's data source; Based on the data source, several category labels are determined; User profiles are built based on tags from several categories.

7. The data classification method based on a fitted classification model and user profile as described in claim 1, characterized in that, When the data type is determined to be unstructured, quantization is performed. Based on the quantized data to be processed, the first cleaning rule is determined, including: When the data type is determined to be unstructured, quantization processing is performed, and quantization features are extracted from the quantized data to obtain corresponding quantization feature information. The quantization feature information is compared with the quantization feature information of historical processed data in the data cleaning model included in the processing process. When the comparison is successful, the data cleaning rule of the historical processed data that was successfully compared is called as the first cleaning rule. When the comparison fails, a corresponding data cleaning rule is generated based on the quantization feature information, and it is used as the first cleaning rule and added to the data cleaning model library of the data cleaning model.

Images