Intelligent data labeling method and system

Abstract

The invention discloses an intelligent data annotation method and system, and relates to the technical field of text processing, and the system comprises a data collection module which obtains a to-be-annotated text document set; and a data preprocessing module. A distributed cluster module, an intelligent engine processing module and an intelligent engine analysis module are arranged, a computer system is used for processing a to-be-labeled text document set, the processing speed is increased, manpower is reduced, meanwhile, a data analysis package and a to-be-labeled text sign are established, a text is analyzed according to the data analysis package, and the annotation efficiency is improved. The method comprises the following steps of: firstly, matching a text label, judging information needing to be labeled of the text, matching the text label with the information needing to be labeled of the text, performing high-precision extension explanation on the uncommon words, the terminologies, the idiom vocabularies and the like, and in addition, visually presenting the text label by a visualization module so as to be convenient for a reader to memorize.

Description

Technical Field

[0001] The present invention relates to the field of text processing technology, and in particular to an intelligent data annotation method and system. Background Art

[0002] Natural Language Processing (NLP) technology has always had many problems in actual implementation, such as insufficient labeled data, multiple business scenarios, and frequent adjustments to business scenarios. Therefore, how to solve the accumulation of business knowledge in the actual implementation and application process, and how to continuously use this knowledge accumulation to improve the natural language processing model based on the existing small amount of prior knowledge has become the focus of research.

[0003] When the existing content annotation is extended for uncommon words, professional terms, idioms, etc. in the text, the text is sometimes disorganized, requiring staff to spend a lot of time to organize it, wasting manpower and material resources. In addition, the existing technology lacks matching degree for text data processing, and its processing effect needs to be improved. Summary of the invention

[0004] In order to solve the above technical problems, an intelligent data annotation method and system are provided. The technical solution solves the problem raised in the above background technology that when extending the existing content annotation for uncommon words, professional terms, idioms, etc. in the text, the text is sometimes disorganized, and the staff needs to spend a lot of time to organize it, which wastes manpower and material resources, and the existing technology also lacks matching degree for text data processing.

[0005] In order to achieve the above purpose, the technical solution adopted by the present invention is:

[0006] An intelligent data labeling method, comprising:

[0007] Get the set of text documents waiting to be annotated;

[0008] Preprocess the text document set waiting for annotation, supplement the missing data in the text document set waiting for annotation, and remove the abnormal data in the text document set waiting for annotation;

[0009] Slice the preprocessed text document set to obtain at least one slice node, perform slice node correlation analysis using judgment regression analysis, establish a text to be annotated schematic diagram based on the correlation of the data, and store the text to be annotated schematic diagram in a pivot table;

[0010] Obtain historical annotated data, integrate and analyze the historical annotated data using artificial intelligence and pattern recognition technology, extract features of the historical annotated data, and build a data analysis package based on the historical annotated data;

[0011] Get the text to be annotated in the pivot table, analyze the data according to the data analysis package, and obtain the annotation information;

[0012] Visualize the annotation information.

[0013] Preferably, the preprocessing of the text document set waiting to be annotated comprises the following steps:

[0014] Detect missing data in the set of text documents waiting to be annotated, and use the non-missing data as the training set;

[0015] Train the prediction model based on the data in the training set as the training environment;

[0016] Extract the features of the part where the data is missing, use the prediction model to predict the alternative data of the part where the data is missing, and use the alternative data to compensate for the part where the data is missing;

[0017] Make a scatter plot corresponding to the text document set waiting to be annotated, and obtain the upper quartile and lower quartile in the scatter plot. The upper quartile is the data at one quarter of the position from the top to the bottom of the scatter plot, and the lower quartile is the data at one quarter of the position from the bottom to the top of the scatter plot;

[0018] Calculate the absolute value of the difference between the upper quartile and the lower quartile to obtain the benchmark difference;

[0019] The sum of the preset multiples of the benchmark difference and the upper quartile is the upper edge value, and the difference between the lower quartile and the preset multiples of the benchmark difference is the lower edge value;

[0020] Data outside the range between the upper edge value and the lower edge value are abnormal data;

[0021] Extract the features of abnormal data, use the prediction model to predict alternative data for the abnormal data, and use the alternative data to compensate for the abnormal data.

[0022] Preferably, the data slicing of the preprocessed text document set comprises the following steps:

[0023] Get the preprocessed text document set, determine whether the data is empty, if the data is an empty set, delete the data, if the data is not an empty set, generate a string corresponding to the data;

[0024] Generate segmentation criteria, and divide the preprocessed text document set into automatic layer and manual layer by category. The automatic layer segmentation is completed by the database system, and the manual layer segmentation is completed manually.

[0025] According to the segmentation standard, a segmentation tag is inserted into the text, and the segmentation tag segments the text into at least one segmentation character;

[0026] Re-identify and segment text into data to obtain data slices.

[0027] Preferably, the slice node correlation analysis using judgment regression analysis includes the following steps:

[0028] Determine the type of the slice node, which is divided into a continuous type and a categorical type. The continuous type slice node and the categorical type slice node are not associated;

[0029] For any two continuous slicing nodes 1 and 2, draw a scatter plot with the continuous slicing node 1 as the independent variable and the continuous slicing node 2 as the dependent variable;

[0030] Determine the fitting model according to the scatter plot, and calculate the fitting function according to the fitting model;

[0031] The sum of the distances from the points in the scatter plot to the fitting function is calculated as a judgment value. If the judgment value is greater than a first preset value, the continuous type slicing node 1 and the continuous type slicing node 2 are not related. If the judgment value does not exceed the first preset value, the continuous type slicing node 1 and the continuous type slicing node 2 are related.

[0032] Among them, the distance from a point in the scatter plot to the fitting function is the minimum value of the distance from a point in the scatter plot to a point on the fitting function;

[0033] For any two category-type slice nodes one and category-type slice nodes two;

[0034] The data in the category-type slice node 1 is classified according to the categories in the category-type slice node 1, the number of data in each category is counted, and the proportion of data in each category is calculated, and the data is arranged from small to large to obtain a1, a2…, a n ;

[0035] The data in the category-type slice node 2 is classified according to the categories in the category-type slice node 2, the number of data in each category is counted, and the proportion of data in each category is calculated, and arranged from small to large to obtain b1, b2…, b n ;

[0036] calculate

[0037] Preferably, if A is greater than a second preset value, category type slice node one and category type slice node two are not associated; if A does not exceed the second preset value, category type slice node one and category type slice node two are associated.

[0038] Preferably, the extracting of historical annotation data features comprises the following steps:

[0039] Obtain sample features and training sample sets extracted from sample data, adjust classifier parameters, input the training sample set into the classifier after parameter adjustment, and output the corresponding sample features;

[0040] Represent historical annotation data using symbols that can be operated by computers;

[0041] De-noise the historical annotation data, extract useful information, and restore the degradation phenomenon existing in the historical annotation data;

[0042] The historical annotated data is transformed and input into the classifier, and the classifier outputs features that reflect the essence of the historical annotated data.

[0043] Preferably, the step of constructing a data analysis package based on historical annotation data includes the following steps:

[0044] Obtain all features of historical annotation data and establish an analysis branch for each feature;

[0045] The analysis branch collects, organizes, processes and models feature data. The analysis branch performs exploratory data analysis on the modeled feature data to obtain a preliminary analysis model of the feature;

[0046] Customized design and testing of the preliminary analysis model. If the preliminary analysis model fails the test, the analysis branch will re-perform exploratory data analysis on the feature data and update the results in the preliminary analysis model until the preliminary analysis model passes the test.

[0047] The preliminary analysis model is matched with the features, the preliminary analysis model is integrated, and a data analysis package is obtained.

[0048] Preferably, analyzing the data to obtain labeling decision information includes the following steps:

[0049] Acquire data, slice the data, and obtain at least one slice node;

[0050] Retrieving at least one text annotation schematic that has an intersection with at least one slice node 1 in the pivot table, and selecting the one with the largest intersection with at least one slice node 1 from the at least one text annotation schematic to be annotated as the best text annotation schematic;

[0051] At least one feature is extracted according to the optimal text annotation schematic, a preliminary analysis model corresponding to the feature is called from a data analysis package, and at least one preliminary analysis model is integrated to form a total analysis to obtain a text annotation.

[0052] Preferably, the visual presentation of the annotation information comprises the following steps:

[0053] Obtain the focus of presentation requirements, extract text annotation data, and clean text annotation data;

[0054] Integrate text annotation data to determine the dimensions of visualization;

[0055] Select a visualization model and generate visualization text annotations.

[0056] An intelligent data annotation system, comprising:

[0057] A data collection module, wherein the data collection module obtains a set of text documents to be annotated;

[0058] A data preprocessing module, wherein the data preprocessing module preprocesses the text document set waiting for annotation, supplements the missing data in the text document set waiting for annotation, and removes abnormal data in the text document set waiting for annotation;

[0059] A distributed cluster module, wherein the distributed cluster module performs data slicing on the preprocessed text document set to obtain at least one slicing node, performs a slice node correlation analysis using a judgment regression analysis, establishes a text to be annotated schematic according to the correlation of the data, and stores the text to be annotated schematic in a pivot table;

[0060] An intelligent engine processing module, wherein the intelligent engine analysis module obtains historical annotation data, integrates and analyzes the historical annotation data using artificial intelligence and pattern recognition technology, extracts features of the historical annotation data, and constructs a data analysis package based on the historical annotation data;

[0061] An intelligent engine analysis module, wherein the intelligent engine analysis module obtains a text to be annotated indication in a pivot table, analyzes the data according to a data analysis package, and obtains annotation information;

[0062] A visualization module is used to visualize the annotation information.

[0063] Compared with the prior art, the present invention has the following beneficial effects:

[0064] By setting up a distributed cluster module, an intelligent engine processing module and an intelligent engine analysis module, a computer system is used to process the text document set waiting for annotation, thereby improving the processing speed and reducing manpower. At the same time, a data analysis package and a text to be annotated indication are established. According to the data analysis package, the text is analyzed to determine the information that needs to be annotated in the text. The information that needs to be annotated in the text includes uncommon words, professional terms, idioms, etc., and then the text annotations are matched to make high-precision extended explanations of the uncommon words, professional terms, idioms, etc. In addition, the visualization module visualizes the text annotations to facilitate readers' memory. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Figure 1 It is a schematic diagram of the method flow of the present invention;

[0066] Figure 2 It is a schematic flow chart of a method for performing data slicing on a preprocessed text document set in the present invention;

[0067] Figure 3 A schematic diagram of the process of extracting historical annotation data features in the present invention;

[0068] Figure 4 A schematic diagram of the process flow of the method for constructing a data analysis package based on historical annotated data in the present invention;

[0069] Figure 5 A schematic diagram of a method flow for obtaining labeling decision information in the present invention;

[0070] Figure 6 It is a flowchart of the method for visually presenting the annotation information in the present invention. DETAILED DESCRIPTION

[0071] The following description is used to disclose the present invention so that those skilled in the art can implement the present invention. The preferred embodiments described below are only examples, and those skilled in the art may think of other obvious variations.

[0072] Please refer to Figure 1-6 As shown, an intelligent data labeling method includes:

[0073] Get the set of text documents waiting to be annotated;

[0074] Preprocess the text document set waiting for annotation, supplement the missing data in the text document set waiting for annotation, and remove the abnormal data in the text document set waiting for annotation;

[0075] Slice the preprocessed text document set to obtain at least one slice node, perform slice node correlation analysis using judgment regression analysis, establish a text to be annotated schematic diagram based on the correlation of the data, and store the text to be annotated schematic diagram in a pivot table;

[0076] Obtain historical annotated data, integrate and analyze the historical annotated data using artificial intelligence and pattern recognition technology, extract features of the historical annotated data, and build a data analysis package based on the historical annotated data;

[0077] Get the text to be annotated in the pivot table, analyze the data according to the data analysis package, and obtain the annotation information;

[0078] Visualize the annotation information.

[0079] Preprocessing the text document set waiting for annotation includes the following steps:

[0080] Detect missing data in the set of text documents waiting to be annotated, and use the non-missing data as the training set;

[0081] Train the prediction model based on the data in the training set as the training environment;

[0082] Extract the features of the part where the data is missing, use the prediction model to predict the alternative data of the part where the data is missing, and use the alternative data to compensate for the part where the data is missing;

[0083] Make a scatter plot corresponding to the text document set waiting to be annotated, and obtain the upper quartile and lower quartile in the scatter plot. The upper quartile is the data at one quarter of the position from the top to the bottom of the scatter plot, and the lower quartile is the data at one quarter of the position from the bottom to the top of the scatter plot;

[0084] Calculate the absolute value of the difference between the upper quartile and the lower quartile to obtain the benchmark difference;

[0085] The sum of the preset multiples of the benchmark difference and the upper quartile is the upper edge value, and the difference between the lower quartile and the preset multiples of the benchmark difference is the lower edge value;

[0086] Data outside the range between the upper edge value and the lower edge value are abnormal data;

[0087] Extract the features of abnormal data, use the prediction model to predict alternative data for the abnormal data, and use the alternative data to compensate for the abnormal data.

[0088] Data slicing of the preprocessed text document set includes the following steps:

[0089] Get the preprocessed text document set, determine whether the data is empty, if the data is an empty set, delete the data, if the data is not an empty set, generate a string corresponding to the data;

[0090] Generate segmentation criteria, and divide the preprocessed text document set into automatic layer and manual layer by category. The automatic layer segmentation is completed by the database system, and the manual layer segmentation is completed manually.

[0091] According to the segmentation standard, a segmentation tag is inserted into the text, and the segmentation tag segments the text into at least one segmentation character;

[0092] Re-identify and segment text into data to obtain data slices.

[0093] In this solution, data slicing is performed to divide the data into basic parts, complete the analysis and processing of the basic parts, use the combination of the basic parts to form the data to be processed, and the analysis and processing of the combination of the basic parts forms the analysis and processing of the data.

[0094] Using judgment regression analysis to perform slice node correlation analysis includes the following steps:

[0095] Determine the type of slice node, which is divided into continuous type and category type. Continuous slice nodes and category slice nodes are not related;

[0096] For any two continuous slicing nodes 1 and 2, draw a scatter plot with the continuous slicing node 1 as the independent variable and the continuous slicing node 2 as the dependent variable;

[0097] Determine the fitting model according to the scatter plot, and calculate the fitting function according to the fitting model;

[0098] The sum of the distances from the points in the scatter plot to the fitting function is calculated as a judgment value. If the judgment value is greater than a first preset value, the continuous type slicing node 1 and the continuous type slicing node 2 are not related. If the judgment value does not exceed the first preset value, the continuous type slicing node 1 and the continuous type slicing node 2 are related.

[0099] Among them, the distance from a point in the scatter plot to the fitting function is the minimum value of the distance from a point in the scatter plot to a point on the fitting function;

[0100] For any two category-type slice nodes one and category-type slice nodes two;

[0101] The data in the category-type slice node 1 is classified according to the categories in the category-type slice node 1, the number of data in each category is counted, and the proportion of data in each category is calculated, and the data is arranged from small to large to obtain a1, a2…, a n ;

[0102] The data in the category-type slice node 2 is classified according to the categories in the category-type slice node 2, the number of data in each category is counted, and the proportion of data in each category is calculated, and arranged from small to large to obtain b1, b2…, b n ;

[0103] calculate If A is greater than the second preset value, there is no correlation between category type slice node one and category type slice node two. If A does not exceed the second preset value, there is a correlation between category type slice node one and category type slice node two.

[0104] In this scheme, the purpose of performing slice node correlation analysis is to find out the nodes in the data that play an analytical role, because there is a certain correlation between the nodes that play a role in the analysis. Therefore, nodes that are not related to other nodes in the rest of the data will inevitably have little effect on the analysis.

[0105] Extracting historical annotation data features includes the following steps:

[0106] Obtain sample features and training sample sets extracted from sample data, adjust classifier parameters, input the training sample set into the classifier after parameter adjustment, and output the corresponding sample features;

[0107] Represent historical annotation data using symbols that can be operated by computers;

[0108] De-noise the historical annotation data, extract useful information, and restore the degradation phenomenon existing in the historical annotation data;

[0109] The historical annotated data is transformed and input into the classifier, and the classifier outputs features that reflect the essence of the historical annotated data.

[0110] In this solution, the features of historical annotation data are extracted to separate the valid parts of the historical annotation data, so as to avoid the interference of the invalid parts of the historical annotation data on data analysis.

[0111] Building a data analysis package based on historical annotated data includes the following steps:

[0112] Obtain all features of historical annotation data and establish an analysis branch for each feature;

[0113] The analysis branch collects, organizes, processes and models feature data. The analysis branch performs exploratory data analysis on the modeled feature data to obtain a preliminary analysis model of the features.

[0114] The preliminary analysis model is customized and tested. If the preliminary analysis model fails the test, the analysis branch will re-perform exploratory data analysis on the feature data and update the results in the preliminary analysis model until the preliminary analysis model passes the test. The preliminary analysis model is matched with the features, and the preliminary analysis model is integrated to obtain a data analysis package. Constructing the data analysis package can provide an analysis basis for the data to be analyzed, extract features from the data to be analyzed, and merge and summarize the processing and analysis of the corresponding features to obtain the analysis results of the data to be analyzed.

[0115] Analyzing the data and obtaining labeling decision information includes the following steps:

[0116] Acquire data, slice the data, and obtain at least one slice node;

[0117] Retrieving at least one text annotation schematic that has an intersection with at least one slice node 1 in the pivot table, and selecting the one with the largest intersection with at least one slice node 1 from the at least one text annotation schematic to be annotated as the best text annotation schematic;

[0118] At least one feature is extracted according to the optimal text annotation schematic, a preliminary analysis model corresponding to the feature is called from a data analysis package, and at least one preliminary analysis model is integrated to form a total analysis to obtain a text annotation.

[0119] Visualizing the annotation information includes the following steps:

[0120] Obtain the focus of presentation requirements, extract text annotation data, and clean text annotation data;

[0121] Integrate text annotation data to determine the dimensions of visualization;

[0122] Select a visualization model and generate visualization text annotations.

[0123] In summary, the advantages of the present invention are: by setting up a distributed cluster module, an intelligent engine processing module and an intelligent engine analysis module, a computer system is used to process a set of text documents waiting to be marked, thereby improving processing speed and reducing manpower. At the same time, a data analysis package and a text to be marked schematic are established. According to the data analysis package, the text is analyzed to determine the information that needs to be marked in the text. The information that needs to be marked in the text includes uncommon words, professional terms, idioms, etc., and then the text annotations are matched to make high-precision extended explanations of the uncommon words, professional terms, idioms, etc. In addition, the visualization module visualizes the text annotations to facilitate readers' memory.

[0124] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions only describe the principles of the present invention. The present invention may be subject to various changes and improvements without departing from the spirit and scope of the present invention. These changes and improvements fall within the scope of the present invention. The scope of protection claimed by the present invention is defined by the attached claims and their equivalents.

Claims

1. An intelligent data labeling method, characterized in that: include: Get the set of text documents waiting to be annotated; Preprocess the text document set waiting for annotation, supplement the missing data in the text document set waiting for annotation, and remove the abnormal data in the text document set waiting for annotation; Slice the preprocessed text document set to obtain at least one slice node, perform slice node correlation analysis using judgment regression analysis, establish a text to be annotated schematic diagram based on the correlation of the data, and store the text to be annotated schematic diagram in a pivot table; Obtain historical annotated data, integrate and analyze the historical annotated data using artificial intelligence and pattern recognition technology, extract features of the historical annotated data, and build a data analysis package based on the historical annotated data; Get the text to be annotated in the pivot table, analyze the data according to the data analysis package, and obtain the annotation information; Visualize the annotation information.

2. The intelligent data labeling method according to claim 1, characterized in that: The preprocessing of the text document set waiting to be annotated comprises the following steps: Detect missing data in the set of text documents waiting to be annotated, and use the non-missing data as the training set; Train the prediction model based on the data in the training set as the training environment; Extract the features of the part where the data is missing, use the prediction model to predict the alternative data of the part where the data is missing, and use the alternative data to compensate for the part where the data is missing; Make a scatter plot corresponding to the text document set waiting to be annotated, and obtain the upper quartile and lower quartile in the scatter plot. The upper quartile is the data at one quarter of the position from the top to the bottom of the scatter plot, and the lower quartile is the data at one quarter of the position from the bottom to the top of the scatter plot; Calculate the absolute value of the difference between the upper quartile and the lower quartile to obtain the benchmark difference; The sum of the preset multiples of the benchmark difference and the upper quartile is the upper edge value, and the difference between the lower quartile and the preset multiples of the benchmark difference is the lower edge value; Data outside the range between the upper edge value and the lower edge value are abnormal data; Extract the features of abnormal data, use the prediction model to predict alternative data for the abnormal data, and use the alternative data to compensate for the abnormal data.

3. The intelligent data labeling method according to claim 2, characterized in that: The data slicing of the preprocessed text document set comprises the following steps: Get the preprocessed text document set, determine whether the data is empty, if the data is an empty set, delete the data, if the data is not an empty set, generate a string corresponding to the data; Generate segmentation criteria, and divide the preprocessed text document set into automatic layer and manual layer by category. The automatic layer segmentation is completed by the database system, and the manual layer segmentation is completed manually. According to the segmentation standard, a segmentation tag is inserted into the text, and the segmentation tag segments the text into at least one segmentation character; Re-identify and segment text into data to obtain data slices.

4. The intelligent data labeling method according to claim 3, characterized in that: The slice node correlation analysis using judgment regression analysis includes the following steps: Determine the type of the slice node, which is divided into a continuous type and a categorical type. The continuous type slice node and the categorical type slice node are not associated; For any two continuous slicing nodes 1 and 2, draw a scatter plot with the continuous slicing node 1 as the independent variable and the continuous slicing node 2 as the dependent variable; Determine the fitting model according to the scatter plot, and calculate the fitting function according to the fitting model; The sum of the distances from the points in the scatter plot to the fitting function is calculated as a judgment value. If the judgment value is greater than a first preset value, the continuous type slicing node 1 and the continuous type slicing node 2 are not related. If the judgment value does not exceed the first preset value, the continuous type slicing node 1 and the continuous type slicing node 2 are related. Among them, the distance from a point in the scatter plot to the fitting function is the minimum value of the distance from a point in the scatter plot to a point on the fitting function; For any two category-type slice nodes one and category-type slice nodes two; The data in the category-type slice node 1 is classified according to the categories in the category-type slice node 1, the number of data in each category is counted, and the proportion of data in each category is calculated, and the data is arranged from small to large to obtain a1, a2…, a n ; The data in the category-type slice node 2 is classified according to the categories in the category-type slice node 2, the number of data in each category is counted, and the proportion of data in each category is calculated, and arranged from small to large to obtain b1, b2…, b n ; calculate 5. The intelligent data labeling method according to claim 4, characterized in that: If A is greater than the second preset value, there is no correlation between category type slice node one and category type slice node two. If A does not exceed the second preset value, there is a correlation between category type slice node one and category type slice node two.

6. The intelligent data labeling method according to claim 5, characterized in that: The extracting of historical annotation data features comprises the following steps: Obtain sample features and training sample sets extracted from sample data, adjust classifier parameters, input the training sample set into the classifier after parameter adjustment, and output the corresponding sample features; Represent historical annotation data using symbols that can be operated by computers; De-noise the historical annotation data, extract useful information, and restore the degradation phenomenon existing in the historical annotation data; The historical annotation data is transformed and input into the classifier, and the classifier outputs features that reflect the essence of the historical annotation data.

7. The intelligent data labeling method according to claim 6, characterized in that: The construction of a data analysis package based on historical annotation data includes the following steps: Obtain all features of historical annotation data and establish an analysis branch for each feature; The analysis branch collects, organizes, processes and models feature data. The analysis branch performs exploratory data analysis on the modeled feature data to obtain a preliminary analysis model of the feature; Customized design and testing of the preliminary analysis model. If the preliminary analysis model fails the test, the analysis branch will re-perform exploratory data analysis on the feature data and update the results in the preliminary analysis model until the preliminary analysis model passes the test. The preliminary analysis model is matched with the features, the preliminary analysis model is integrated, and a data analysis package is obtained.

8. The intelligent data labeling method according to claim 7, characterized in that: Analyzing the data to obtain labeling decision information includes the following steps: Acquire data, slice the data, and obtain at least one slice node; Retrieving at least one text annotation schematic that has an intersection with at least one slice node 1 in the pivot table, and selecting the one with the largest intersection with at least one slice node 1 from the at least one text annotation schematic to be annotated as the best text annotation schematic; At least one feature is extracted according to the optimal text annotation schematic, a preliminary analysis model corresponding to the feature is called from a data analysis package, and at least one preliminary analysis model is integrated to form a total analysis to obtain a text annotation.

9. The intelligent data labeling method according to claim 8, characterized in that: The visualization module visually presents the annotation information, including the following steps: Obtain the focus of presentation requirements, extract text annotation data, and clean text annotation data; Integrate text annotation data to determine the dimensions of visualization; Select a visualization model and generate visualization text annotations.

10. An intelligent data labeling system, used to implement an intelligent data labeling method according to claims 1-9, characterized in that: include: A data collection module, wherein the data collection module obtains a set of text documents to be annotated; A data preprocessing module, wherein the data preprocessing module preprocesses the text document set waiting for annotation, supplements the missing data in the text document set waiting for annotation, and removes abnormal data in the text document set waiting for annotation; A distributed cluster module, wherein the distributed cluster module performs data slicing on the preprocessed text document set to obtain at least one slicing node, performs a slice node correlation analysis using a judgment regression analysis, establishes a text to be annotated schematic according to the correlation of the data, and stores the text to be annotated schematic in a pivot table; An intelligent engine processing module, wherein the intelligent engine analysis module obtains historical annotation data, integrates and analyzes the historical annotation data using artificial intelligence and pattern recognition technology, extracts features of the historical annotation data, and constructs a data analysis package based on the historical annotation data; An intelligent engine analysis module, wherein the intelligent engine analysis module obtains a text to be annotated indication in a pivot table, analyzes the data according to a data analysis package, and obtains annotation information; A visualization module is used to visualize the annotation information.

Images