All Collections
The Management Page
How do I customize Topic or Opinion Groups?
How do I customize Topic or Opinion Groups?

Refine the system’s grouping of synonyms and closely related words in your analysis.

Amanda Robinson avatar
Written by Amanda Robinson
Updated this week

Why do we need groups?

When many different people provide customer feedback, their comments will contain a wide variety of words and phrases to express the same idea. The AI technology behind the Keatext text analytics platform groups all these variations automatically to allow the user to save time. A group is a container for one or several Topics and Opinions that are synonymous, nearly synonymous or should be treated as such by Keatext.

The above example shows variations in the Opinions. Here, “dirty”, “filthy”, “stained”, “grubby”, “gross”, “not been cleaned” and even the misspelled “unclen” are all grouped together. Each group has a representative label, which in this case is “dirty”. The labels are chosen automatically based on frequency and other parameters, such as length.

The user has control over both Topic groups and Opinion groups, which in turn affect how Comments are grouped.

Topic and Opinion groups are ranked in order of comment count, showing a clear prioritization based on how widespread a specific Problem or Praise is.

Why customize groups?

In some instances, users may want to customize the analysis to further adapt and refine the text mining results to suit their needs. There are several reasons why a user would choose to do this, including a need to reflect domain knowledge in the results, change the focus of the analysis, or tweak some visible errors.

Groups can also reflect organizational knowledge—for example, a “unit” and an “apartment” may mean the same thing for an apartment leasing company. Your organization may also use abbreviations (such as CSR for customer service representative in call center transcripts or other documents).

Can I use grouping to improve accuracy?

For small tweaks, modifying groups can be quite powerful. However, for recurring issues with accuracy, it’s best to discuss this with your Keatext account manager to see if retraining is a better option than tweaking groups.

Who will these changes affect?

It’s important to note that groupings are organization-wide. If several users are part of the same organization, changes made to groups by one user will be reflected for all users.

Are the changes permanent?

Changes made by the user to groups can be reset. The reset button returns all of the groups, for all of the users in the organization, to default, so it’s best avoided if your organization has spent hours customizing its groups. You cannot undo this reset once it has been done. It’s good practice to export your grouping file once modifications are complete. In the case that someone resets your groups, you could always import the grouping file and you will not have lost your work.

How do I manage groups?

First, head to the Management page by clicking on the checklist icon in the navigation menu on the left of the screen. This page is used to manage Topic and Opinion Groups and Filtersets. Click on either Topics or Opinions.

This page has two panels. The Topic groups panel (on the left) contains all groups with more than one member. To see the members of a group, click on the arrow to the right of the group name.

The Topics panel (on the right) contains all the topics in that dataset, both those that are in groups— indicated by "Member of”—and single words (groups with only one member). Use the toggle to switch between viewing all topics or only viewing single words. The Topics panel is useful when searching for members contained within a group.

The Topic groups panel can be sorted by number of Members in the group as well as alphabetically by Group Name. The Topics panel can be sorted alphabetically by Topic, by the frequency of its occurrence within the data, and by the date created - meaning the date that Topic was imported in a source. Once you enter a search term, the sort will default to Similarity to the search term.

Groups can be managed either by modifying the settings for a group individually, or in bulk by selecting the checkboxes to the left of the groups and modifying multiple groups at the same time. The options are:

Topic groups panel - Group-level actions:

Dissolve group - (Topic groups panel only) Permanently breaks a group down into individual words. This is a permanent change, so make sure everyone in the organization approves this change as it will affect all users.

Merge groups - Combine two or more groups into one larger group.

Topic groups panel - Member-level actions:

Move - Select a member and move it to a new or another existing group.

Remove - If you remove a member from a group it will become a single word.

Topics panel:

Move - Move a topic to a new or another existing group. You can move a single word or a word that is already part of another group.

Resetting, importing and exporting groups, and smart grouping

Resetting groups

In the upper right corner, you’ll also see the “Reset Groups” button. Please be very very careful with this button! This will reset all the groups that you have manually modified back to the groups that were formed automatically with Keatext’s AI technology. The reset is organization-wide, meaning if you reset, you reset for all the users in the organization. Once the reset is performed you will lose the manually modified groups permanently.

Exporting groups

Once you are done modifying your groups they can be exported in JSON format to be used in another organization, or to be saved in the case that you need to reimport your groups to this organization.

Importing groups

You can import groups by uploading a JSON file. If you make changes to your JSON file before uploading, please follow the guidelines provided.

Checking your modified file

If you make any modifications to a JSON file, it is a good idea to check to make sure it is valid prior to importing. This link should help you with that.

Smart Grouping

Smart grouping differ slightly from the default groups. A different algorithm is used.

Default groups gather members that share the same context together into a group (apple, pear, food).

Smart groupings gathers members based on semantic based synonymy (awesome, incredible, terrific, good) and on their morphological similarity (works great, works very well; able to process, ability to process).

If you prefer to use smart grouping simply press the button and they will be applied. If you have previously modified you groups the changes you have made will not be altered by pressing this button.

Did this answer your question?