Classified marking method and system based on AI text analysis
By constructing a multi-level topic tagging system and a small sample dataset, and combining it with a large model for semantic analysis, the problem of high cost and insufficient accuracy of manual annotation in the sentiment and topic recognition of unstructured text is solved, achieving efficient and accurate annotation result display and business adaptability.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SUZHOU ZHONGYAN NETWORK TECH CO LTD
- Filing Date
- 2026-01-27
- Publication Date
- 2026-05-08
AI Technical Summary
Existing technologies for sentiment and topic recognition in unstructured text suffer from high costs, low efficiency, and significant subjective differences due to traditional manual annotation. Machine learning-based methods, on the other hand, rely on high-quality labeled data and have limited ability to handle complex semantics, while also risking overfitting, resulting in insufficient model generalization ability and annotation accuracy.
We employ an AI-based text analysis-based classification and labeling method. By constructing a multi-level topic tagging system, configuring topic annotations and small sample datasets, and combining it with a large model for semantic analysis, we output the labeling results and improve labeling efficiency and accuracy through visualization.
It enables efficient processing of unstructured text, reduces the cost of manual annotation, improves annotation accuracy and interpretability, enhances the system's adaptability to business changes, simplifies operation processes, and improves decision support capabilities.
Smart Images
Figure CN121997936A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of natural language processing and artificial intelligence applications, and in particular to a classification and labeling method and system based on AI text analysis. Background Technology
[0002] In the field of natural language processing, automatic text annotation technology has gradually evolved from the early stage of relying on rule systems and traditional neural network models to the current stage of widely using large pre-trained language models (LLMs). Leveraging the powerful semantic understanding and generalization capabilities of large models, basic annotation tasks such as sentiment and topic annotation are achieved through fine-tuning or cue word engineering, which has become mainstream practice in the industry.
[0003] However, existing technologies still face significant bottlenecks in sentiment and topic recognition for unstructured text: traditional manual annotation methods are costly, inefficient, and subject to large subjective differences; while machine learning-based methods rely heavily on high-quality labeled data, have limited ability to handle complex semantics, new domain knowledge, and linguistic phenomena (such as metaphors and irony), and are at risk of overfitting, resulting in insufficient model generalization ability and annotation accuracy.
[0004] Therefore, there is an urgent need for a new method and system that integrates the semantic understanding capabilities of large models, which can improve the recognition accuracy of unstructured text while ensuring annotation efficiency and scalability, and reduce the dependence on large-scale labeled data and continuous manual maintenance. Summary of the Invention
[0005] The purpose of this invention is to provide a classification and labeling method and system based on AI text analysis, which aims to improve the efficiency, accuracy and interpretability of sentiment and topic labeling of unstructured text, reduce reliance on manual labeling, and lower the cost of manual labeling.
[0006] To achieve the above objectives, this invention provides a classification and labeling method based on AI text analysis, comprising the following steps: Step S1: Create a configuration annotation model; Step S2: Create a text analysis project and associate the unstructured text dataset with the annotation model; Step S3: Based on the configuration information of the annotation model, call the large model to perform semantic analysis on the associated unstructured text dataset and output the annotation results; Step S4: Perform statistical analysis and visualization of the annotation results.
[0007] Preferably, step S1 specifically includes: Step S101: Construct a topic tag system with at least two levels of structure; Step S102: Configure tag information for topic tags in the topic tag system; tag information includes topic name, topic alias, and annotations; annotations include the definition of the topic and tagging rules; Step S103: Add a small sample dataset to the labeled model, where each small sample includes text information, whole sentence sentiment, topic tags, and topic reasons.
[0008] Preferably, step S2 specifically includes: Step S201: Create a text analysis project and configure the project name; Step S202: Select at least one data source for the text analysis project from the unstructured text dataset, and specify the text field to be analyzed from the data source; Step S203: Select at least one annotation model from the annotation models configured in step S1 to associate with the text analysis project and establish a binding relationship between the data source and the annotation model.
[0009] Preferably, step S202 further includes: specifying a field containing time information from the data source as the time field, and selecting one or more fields as selection filter fields.
[0010] Preferably, in step S4, the visualization includes: Generate a sentiment analysis dashboard, including the total number of texts, positive ratings, negative ratings, sentiment index, overall sentiment trend map, and sentiment distribution; Generate topic analysis dashboards, including the top 5 topics by mention rate, the top 5 topics by positive review rate, the top 5 topics by negative review rate, topic statistics, topic statistics bubble charts, and topic relevance.
[0011] Preferably, topic statistics include the topic and its corresponding mentions, sentiment index, negative review rate, and positive review rate; Topic statistics bubble charts use bubble charts to display topic performance, where the size of the bubble represents the number of topics, the pie chart inside the bubble represents the ratio of positive reviews to negative reviews, and the color represents the sentiment of the topic.
[0012] Preferably, the topic relevance is obtained by calculating the correlation coefficient between topics, with a value range of -1 to 1; the topic analysis dashboard classifies the relevance between topics according to the numerical range of the topic relevance, specifically: below 0.3 is weakly correlated; 0.3 to 0.6 is moderately correlated; and above 0.6 is strongly correlated.
[0013] This invention also provides a classification and labeling system based on AI text analysis, which employs the above method and includes a model management module, a text analysis project management module, a large model module, and a visualization dashboard module; The model management module provides a model management configuration interface for users to create and configure labeled models. The text analysis project management module provides a text analysis project management interface, allowing users to create text analysis projects and associate unstructured text datasets with labeled models from the model management module. The large model module is used to receive the associated information from the text analysis project management module, call the large model to process the unstructured text dataset based on the configuration information of the annotation model, and obtain the annotation results output by the large model. The visualization dashboard module is used to receive annotation results from the large model module and perform statistical and visualization displays.
[0014] Preferably, the model management module includes: The topic configuration unit is used to build and manage a topic tag system with at least two levels of structure, and to configure tag information for topic tags in the topic tag system. The sample learning unit is used to add and manage a small sample dataset for the labeled model.
[0015] Preferably, the visual dashboard module includes: The sentiment analysis dashboard unit is used to generate and display sentiment statistics and charts; The topic analysis dashboard unit is used to generate and display topic statistics and charts; The original text unit is used to view the text's tagging status.
[0016] Therefore, the present invention employs the above-mentioned classification and labeling method and system based on AI text analysis, and the beneficial technical effects are as follows: (1) This invention achieves efficient processing of unstructured text through a large model-based annotation method. Users only need to configure the model in three steps (i.e., constructing a multi-level topic tag system, adding topic annotations, and adding small sample datasets to guide the large model to understand and learn) to call the large model for analysis, which reduces the large amount of manually annotated data, long training cycle and manpower and time costs required for traditional modeling methods.
[0017] (2) This invention guides large models to understand the annotation rules more accurately by designing a configuration process consisting of a multi-level topic tag system, semantically enhanced topic annotations and small sample datasets, thereby improving the annotation accuracy and semantic understanding depth of complex semantic, metaphor, irony and other linguistic phenomena and new domain texts.
[0018] (3) This invention provides a flexible and scalable topic tag management system. Users can add, modify or adjust the topic tag hierarchy at any time according to business needs without retraining the model, thereby enhancing the system's adaptability to business changes and new topic types, and avoiding the continuous manual maintenance required by traditional methods.
[0019] (4) This invention reduces the user threshold by simplifying the three-step operation process and the intuitive visual management interface. It can quickly complete the model configuration and text analysis project management, and can intuitively obtain multi-dimensional analysis results such as topic statistics and topic relevance through the visual dashboard, thereby improving the operation efficiency and decision support capabilities. Attached Figure Description
[0020] Figure 1 This is a flowchart of a classification and labeling method based on AI text analysis according to the present invention; Figure 2 This is a structural diagram of a classification and labeling system based on AI text analysis according to the present invention. Detailed Implementation
[0021] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.
[0022] Unless otherwise defined, the technical or scientific terms used in this invention shall have the ordinary meaning as understood by one of ordinary skill in the art to which this invention pertains.
[0023] Example 1 like Figure 1 As shown, a classification and labeling method based on AI text analysis includes the following steps: Step S1: Create a configuration annotation model. This specifically includes: Step S101: Construct a topic tag system with at least two levels.
[0024] Taking the two-level topic tag system as an example, the first-level topic is "product", which includes two second-level topics: "product quality" and "product appearance".
[0025] Step S102: Configure tag information for topic tags in the topic tag system.
[0026] Tag information includes topic name, topic alias, and comments.
[0027] The topic name must be unique. Topic aliases are edited versions of the original topic name, making them more easily understood and facilitating model recognition. Annotations include the topic definition and labeling rules; appropriate annotations can improve the accuracy of topic recognition.
[0028] In addition, topic tags and their information can be uploaded, canceled, and saved in batches, and the hierarchy of topic tags can be adjusted by dragging.
[0029] Step S103: Add a small sample dataset to the labeled model, where each small sample includes text information, whole sentence sentiment, topic tags, and topic reasons.
[0030] Small sample datasets are provided as examples for large models to guide them on how to perform sentiment recognition and topic labeling for specific tasks based on annotation models. This helps them to understand more clearly the classification rules and labeling logic corresponding to different semantics, thereby improving the accuracy and interpretability of the annotation results and avoiding errors in labeling semantically similar texts.
[0031] It should be noted that the number of small samples in a small sample dataset can be increased to a maximum of 10.
[0032] Step S2: Create a text analysis project, linking the unstructured text dataset to be processed with the annotation model. Specifically, this includes: Step S201: Create a text analysis project and configure the project name.
[0033] Step S202: Select at least one data source for the text analysis project from the unstructured text dataset, and specify one or more text fields to be analyzed from the data source.
[0034] Only text fields from the data source can be selected for analysis to ensure the effectiveness of large model analysis.
[0035] Specify a field containing time information from the data source as the time field, i.e., a date-formatted field in the data source, which will be used to generate trend changes in sentiment or topic over time.
[0036] Select one or more fields from the data source as selection filter fields to dynamically filter and group the analysis results in the visualization dashboard. Up to three filter fields can be selected.
[0037] If the data source selected for the text analysis project is pre-linked with hierarchical data, the organizational hierarchy filtering function can be enabled. This allows for drill-up, drill-down, or peer-to-peer comparison of the analysis results in the visualization dashboard based on the organizational hierarchy, enabling multi-granularity performance insights from the group level down to the department level.
[0038] In addition, you can manage each text analysis project, such as viewing the analysis progress, modifying the configuration of the text analysis project, and deleting, updating, or rerunning the analysis task.
[0039] Step S203: Select at least one annotation model from the annotation models configured in step S1 to associate with the text analysis project and establish a binding relationship between the data source and the annotation model.
[0040] Step S3: Based on the configuration information of the annotation model, call the large model to perform semantic analysis on the associated unstructured text dataset, identify sentiment and topic, and output annotation results.
[0041] The annotation results include the overall sentiment of the text (positive / neutral / mixed / negative), the sentiment of the mentioned topics and the sentiment of the topics (positive / neutral / negative).
[0042] Step S4: Perform statistical analysis and visualization of the annotation results.
[0043] The visualization includes: (1) Generate a sentiment analysis dashboard to display the overall sentiment distribution and trends of the unstructured text dataset.
[0044] The sentiment analysis dashboard includes the total number of texts in the unstructured text dataset, positive ratings (the number of texts with positive sentiment and their percentage), negative ratings (the number of texts with negative sentiment and their percentage), sentiment index (calculated based on the percentage of positive and negative sentiments), overall sentiment trend map, and sentiment distribution.
[0045] The overall sentiment trend chart is used to show the overall sentiment trend of the quantity of each sentiment over time. This chart supports switching the date dimension; in addition, the sentiment distribution also shows the distribution of the quantity of different sentiments through graphics.
[0046] (2) Generate topic analysis dashboards to display the mentions and evaluation statistics of each topic.
[0047] The topic analysis dashboard includes the top 5 topics by mention rate, the top 5 topics by positive review rate, the top 5 topics by negative review rate, topic statistics, a topic statistics bubble chart, and topic relevance.
[0048] The top 5 topics by mention rate, positive review rate, and negative review rate refer to the top 5 topics displayed in order of mention rate, positive review rate, and negative review rate, respectively.
[0049] Topic statistics include the topic and its corresponding mentions, sentiment index, negative review rate, and positive review rate. Following the hierarchical structure of the topic tagging system, the mentions, sentiment index, negative review rate, and positive review rate for each level of topic are displayed sequentially. The mention count refers to the number of topics; a single text can be tagged with multiple topics.
[0050] Topic statistics bubble charts use bubble charts to display topic performance, where the size of the bubble represents the number of topics, the pie chart inside the bubble represents the ratio of positive reviews to negative reviews, and the color represents the sentiment of the topic.
[0051] Topic relevance is obtained by calculating the correlation coefficient between topics, with a value ranging from -1 to 1. The higher the value, the stronger the correlation, indicating that the two topics are more likely to be mentioned simultaneously. The topic analysis dashboard classifies the relevance between topics according to the numerical range of the topic relevance coefficient: below 0.3 is weakly correlated; 0.3 to 0.6 is moderately correlated; and above 0.6 is strongly correlated.
[0052] like Figure 2 As shown, a classification and labeling system based on AI text analysis employs the above-mentioned method and includes a model management module, a text analysis project management module, a large model module, and a visualization dashboard module.
[0053] The model management module provides a model management configuration interface for users to create and configure labeled models, including: The topic configuration unit is used to build and manage a topic tag system with at least two levels of structure, and to configure tag information for the topic tags in the topic tag system.
[0054] The sample learning unit is used to add and manage a small sample dataset for the labeled model.
[0055] The text analysis project management module provides a text analysis project management interface, allowing users to create text analysis projects and associate unstructured text datasets with labeled models from the model management module.
[0056] The large model module is used to receive associated information from the text analysis project management module, call the large model based on the configuration information of the annotation model to process the unstructured text dataset, and obtain the annotation results output by the large model.
[0057] The visualization dashboard module receives annotation results from the large model module and performs statistical and visualization displays, including: The sentiment analysis dashboard unit is used to generate and display sentiment statistics and charts.
[0058] The topic analysis dashboard unit is used to generate and display topic statistics and charts; The original text unit is used to view the text's tagging status.
[0059] The following section will use a specific application scenario, "online user review analysis in the catering industry," as an example to illustrate the implementation process of this invention in detail.
[0060] Suppose a restaurant chain wants to use the system provided by this invention to analyze the massive amount of online user reviews it collects (from a food delivery platform, a review app, etc.) to understand customers' positive and negative feedback on aspects such as "taste," "service," and "environment."
[0061] Users first need to create a labeling model specifically for the "restaurant reviews" field to define the sentiment dimensions and specific topics that the system needs to identify.
[0062] After logging into the system, users can click the "Model Management" button to enter the model management interface. This interface displays a list of all created models and their usage statistics.
[0063] Users click the "Create Model" button to enter the model creation process. On the "Basic Information" screen, users need to fill in: Model name: "Restaurant Reputation Analysis Model_v1".
[0064] Business Description: "Used to analyze the sentiment and specific topics of user reviews in the catering industry to assist in product and service optimization."
[0065] After filling in the above information, click "Next".
[0066] Enter the "Add Topic and Keywords" interface. On this interface, users can build a two-level topic tag system and add detailed tag information. Specifically: Click "Add Topic", enter "Food" in the topic name field, and create a first-level topic "Food".
[0067] Add a second-level topic to the primary topic "Food": (1) Topic name: "Taste", topic alias: "Flavor", annotation: "Evaluation of the sour, sweet, bitter, spicy, salty and other flavors of the dishes, as well as the texture (such as soft, hard, crisp) and freshness".
[0068] (2) Topic name: "portion", topic alias: "portion size", note: "evaluate whether the portion size of the dish is sufficient and cost-effective".
[0069] (3) Topic name: "Appearance", topic alias: "Selling quality", annotation: "Evaluate the color, plating and visual effect of the dish".
[0070] Click "Add Topic" again to create a first-level topic "Service".
[0071] Add a second-level topic to the first-level topic "Service": (1) Topic name: "Service attitude", topic alias: "Attitude", annotation: "Evaluate whether the waiter is enthusiastic, polite and patient".
[0072] (2) Topic name: "Speed of serving food", topic alias: "Speed of serving dishes", note: "Evaluate the time from ordering to serving the food".
[0073] Users can adjust the hierarchy and order of topics by dragging and dropping. After configuring all topics, click "Save".
[0074] To improve the annotation accuracy of large models in complex contexts, users added small sample datasets to guide model learning.
[0075] In the sample learning area, click "Add". In the pop-up window, enter the following: Text message: "The waiter was too slow in serving the food; I waited for almost an hour, but his attitude was quite good."
[0076] Overall sentiment: negative (because the overall experience was poor).
[0077] Click "Add" to tag the topic: Select the topic: "Service speed" and choose "Negative" for the sentiment under this topic.
[0078] Select the topic "Service Attitude" and choose "Positive" as the emotion for this topic.
[0079] Reasons for filling in the topic: "The text message 'the food was served too slowly' clearly mentions the speed of food preparation, which is a negative evaluation; 'the attitude was quite good' clearly mentions the service attitude, which is a positive evaluation."
[0080] Click "Confirm" to save the sample. Users can add up to 10 representative samples in this way to cover complex review scenarios such as "good taste but small portion" and "poor environment but excellent service".
[0081] At this point, a "Restaurant Reputation Analysis Model_v1" containing a customized topic tagging system, clear tagging rules, and guiding samples has been created and put into use.
[0082] Once the model is configured, users can create specific analysis tasks.
[0083] Enter the text analysis project management interface, click the "Add Analysis Project" button to create a new text analysis project.
[0084] In the pop-up configuration window, the user makes the following configuration: Project Name: "Store Review Analysis in a Certain City in Q3 2024"
[0085] Data source: Select the uploaded unstructured text dataset "a food delivery platform & an online review software_2024Q3_a certain city" from the drop-down list.
[0086] Text field: Select the field in the dataset that contains user comments, i.e., "comment content".
[0087] Time field: Select the field where the comment was recorded, i.e., "Comment Time".
[0088] Select filter fields: Select the "City" and "Store ID" fields for subsequent analysis result filtering.
[0089] Select Model: Select the newly created "Restaurant Reputation Analysis Model_v1" from the model drop-down list.
[0090] After configuration, click "Confirm" to create a text analysis project. The system will automatically call the large model in the background and label each review in the selected dataset with sentiment and topic according to the rules defined in "Restaurant Reputation Analysis Model_v1".
[0091] Once the system has finished processing, users can view the annotation results through a visual dashboard.
[0092] Access the sentiment analysis dashboard for this text analysis project to view the sentiment analysis. The top of the sentiment analysis dashboard displays key sentiment metrics: Total number of texts: 26,476.
[0093] Positive reviews: 22,592, accounting for 85%.
[0094] Negative reviews: 2942, accounting for 11%.
[0095] Emotional Index: 74% (positive percentage vs. negative percentage).
[0096] Meanwhile, the overall sentiment trend chart shows the fluctuations in the number of positive, negative, neutral, mixed, and total ratings each day within a given month. The sentiment distribution is visually displayed using a pie chart, showing the proportion of positive, negative, neutral, and mixed ratings. Users can use the time filter to view performance across different time periods.
[0097] In the topic analysis dashboard, switch to the "Topic Analysis" view for in-depth analysis of topic performance. The system automatically generates multiple leaderboards: Top 5 most frequently mentioned topics: "Food taste" (28.4%) ranked first and was the most discussed topic among customers.
[0098] Top 5 in terms of positive reviews: "Environmental Hardware" (97.6%) had the highest positive review rate.
[0099] The top 5 issues with the highest negative review rates are "delivery fees" (100% negative review rate) and "food not fresh" (100% negative review rate) (although the number of mentions may be low).
[0100] Below the topic analysis dashboard, a table is used to group each primary topic such as "Food," "Service," and "Environment," displaying the number of mentions, sentiment index, and positive / negative review rate for each secondary topic. A progress bar is used to visualize the ratio of positive to negative reviews.
[0101] Click the "bubble" icon in the upper right corner of the page to switch to the topic statistics bubble chart. The chart focuses on "food," displaying the strength of its association with topics such as "service," "price," and "packaging," as well as their sentiment (green indicates positive association, red indicates negative association). Clicking the "Food Taste" bubble allows you to drill down and view specific comment examples mentioning this topic, such as "The pickled fish tastes good, but it's a bit expensive," thus understanding the reasons behind the data.
[0102] With the above content, users can quickly obtain in-depth, multi-dimensional semantic analysis results of massive amounts of unstructured text without writing any code or performing complex machine learning training, providing efficient and intuitive data support for business decisions.
[0103] Therefore, the present invention adopts the above-mentioned classification and labeling method and system based on AI text analysis, which improves the labeling efficiency, accuracy and interpretability of sentiment and topic labeling of unstructured text, reduces the dependence on manual labeling and lowers the cost of manual labeling.
[0104] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.
Claims
1. A classification and labeling method based on AI text analysis, characterized in that, Includes the following steps: Step S1: Create a configuration annotation model; Step S2: Create a text analysis project and associate the unstructured text dataset with the annotation model; Step S3: Based on the configuration information of the annotation model, call the large model to perform semantic analysis on the associated unstructured text dataset and output the annotation results; Step S4: Perform statistical analysis and visualization of the annotation results.
2. The classification and labeling method based on AI text analysis according to claim 1, characterized in that, Step S1 specifically includes: Step S101: Construct a topic tag system with at least two levels of structure; Step S102: Configure tag information for topic tags in the topic tag system; tag information includes topic name, topic alias, and annotations; annotations include the definition of the topic and tagging rules; Step S103: Add a small sample dataset to the labeled model, where each small sample includes text information, whole sentence sentiment, topic tags, and topic reasons.
3. The classification and labeling method based on AI text analysis according to claim 1, characterized in that, Step S2 specifically includes: Step S201: Create a text analysis project and configure the project name; Step S202: Select at least one data source for the text analysis project from the unstructured text dataset, and specify the text field to be analyzed from the data source; Step S203: Select at least one annotation model from the annotation models configured in step S1 to associate with the text analysis project and establish a binding relationship between the data source and the annotation model.
4. The classification and labeling method based on AI text analysis according to claim 3, characterized in that, Step S202 further includes: specifying a field containing time information from the data source as the time field, and selecting one or more fields as selection filter fields.
5. The classification and labeling method based on AI text analysis according to claim 1, characterized in that, In step S4, the visualization includes: Generate a sentiment analysis dashboard, including the total number of texts, positive ratings, negative ratings, sentiment index, overall sentiment trend map, and sentiment distribution; Generate topic analysis dashboards, including the top 5 topics by mention rate, the top 5 topics by positive review rate, the top 5 topics by negative review rate, topic statistics, topic statistics bubble charts, and topic relevance.
6. The classification and labeling method based on AI text analysis according to claim 5, characterized in that, Topic statistics include the topic and its corresponding mentions, sentiment index, negative review rate, and positive review rate; Topic statistics bubble charts use bubble charts to display topic performance, where the size of the bubble represents the number of topics, the pie chart inside the bubble represents the ratio of positive reviews to negative reviews, and the color represents the sentiment of the topic.
7. The classification and labeling method based on AI text analysis according to claim 5, characterized in that, Topic relevance is obtained by calculating the correlation coefficient between topics, with a value range of -1 to 1. The topic analysis dashboard classifies the relevance between topics according to the numerical range of the topic relevance: below 0.3 is weakly correlated; 0.3 to 0.6 is moderately correlated; and above 0.6 is strongly correlated.
8. A classification and labeling system based on AI text analysis, employing the method described in any one of claims 1 to 7, characterized in that, It includes a model management module, a text analysis project management module, a large model module, and a visualization dashboard module; The model management module provides a model management configuration interface for users to create and configure labeled models. The text analysis project management module provides a text analysis project management interface, allowing users to create text analysis projects and associate unstructured text datasets with labeled models from the model management module. The large model module is used to receive the associated information from the text analysis project management module, call the large model to process the unstructured text dataset based on the configuration information of the annotation model, and obtain the annotation results output by the large model. The visualization dashboard module is used to receive annotation results from the large model module and perform statistical and visualization displays.
9. A classification and labeling system based on AI text analysis according to claim 8, characterized in that, The model management module includes: The topic configuration unit is used to build and manage a topic tag system with at least two levels of structure, and to configure tag information for topic tags in the topic tag system. The sample learning unit is used to add and manage a small sample dataset for the labeled model.
10. A classification and labeling system based on AI text analysis according to claim 9, characterized in that, The visual dashboard module includes: The sentiment analysis dashboard unit is used to generate and display sentiment statistics and charts; The topic analysis dashboard unit is used to generate and display topic statistics and charts; The original text unit is used to view the text's tagging status.