AI Agents - Text Autotagger Agent

Introduction

The Text Autotagger Agent autonomously categorizes and tags text data with customizable tags. You can define tags you wish to use based on your inputs and/or let the AI Agent organically discover tags based on the content of the data. The agent will iterate through your data, applying tags to each row.

The agent supports tagging in all major languages and considers emoticons. It offers advanced tagging functionalities, including sentiment analysis, and various tagging format options.

Defining the Tags

There are four ways to input or generate tags for the AI to use:

Pure Organic: Provide tagging instructions without mandatory keywords or reference files. The AI will generate tags based on your instructions and the data content.
Comma-Separated List: Input mandatory tags in a comma-separated format, suitable for a single hierarchy level. This can be combined with organic tagging to find additional tags.
Structured Reference Dataset: Define tags in a structured format either listed in one column or with different hierarchy levels across multiple columns if required (e.g., Level 1: Automotive, Level 2: Parts Manufacturer, Retailer, etc.). These columns will then appear in your output tagged dataset.
Open-Ended/Unstructured Documents: Connect documents with data in any format (e.g., Word, PDF). The AI will generate tags based on the unstructured information provided.

Output Formats

There are different ways in which the tagged output dataset can be formatted:

Expand into multiple rows: Rows with multiple tags will be duplicated, with one tag applied per row.
Create a column for each unique tag: Instead of a single column containing all tags, the agent will create a new column for each unique tag across the dataset. Each column will contain a True/False value indicating whether the tag applies to that row. This is similar to one-hot encoding.
Force only one tag per element: The agent will select the most appropriate or dominant tag for each row and discard any additional tags.
Use a comma-separated list within one column: All tags will be preserved as a comma-separated list within a single column.

Agent Requirements

Below are the requirements (both mandatory and optional) for the agent to run successfully. These will either appear as input fields when using this agent through AI Agent Run.

Requirement	Description
Dataset to be Tagged	Source dataset for applying tags. Ensure it's added to the canvas and connected to the AI DT, AI Agent Run or AI Chat node. For more info on collecting data into the platform see here.
Column to be Tagged	The column that contains the content you wish to tag.
Tagging Instructions	Provide instructions to guide the agent on the desired tagging outcomes, focusing on higher-level themes (for example, analyzing customer reviews for product defects or customer service issues). List any rules or guidance you want the agent to follow during auto-tagging. Optionally, specify topics you do not want tagged (for example, excluding price-related issues). You can also add context about the data being tagged, such as whether it comes from CSAT surveys, support tickets, or social media posts.
Output Column Name	The name of the tagger results column in the generated dataset
Reference Dataset (Optional)	Provide a dataset with a list of tags or a tagging hierarchy for the agent to use. Or you can provide unstructured data in any format which the AI will interpret and use to create the tagging list.
Number of Tags	Maximum number of unique tags to use.
Mandatory Tags (Optional)	Tags that must be used. The agent will generate additional tags up to the specified limit.
Run Sentiment Analysis (Optional)	Tag data with Positive, Negative, or Neutral sentiment in a new column.
Build Summaries (Optional)	Select to also produce summaries of each element.
Reuse Previous Tags (Optional)	Option to reuse tags from a previous run.
Multiple Tag Expansion (Optional)	Options for handling multiple tags per data row: expand into multiple rows, create a column for each unique tag, force one tag per element, or use a comma-separated list within one column. (See more details in the section above if needed).