This document describes the necessary steps to identify and delete a secondary topic in the Conversational Insights topic modeling feature.
Topic modeling is typically used to identify primary call drivers, but it can also recognize secondary topics that might be less interesting from an analysis perspective. These secondary topics are frequently related to regular process steps that happen in a conversation, including authentication, confirmation, and collecting feedback.
Conversational Insights provides the important matching topics for your conversations. Secondary topics can sometimes crowd out the more interesting primary topics, which makes it harder to spot the primary ones. This page provides guidance for how to identify and remove such secondary topics in your data, so it is easier to spot interesting trends.
Identify secondary topics
This section outlines how to identify secondary topics.
Examine the topic description and sample topic utterances
After you train a Conversational Insights model, you can examine both the names and sample snippets for each topic. If necessary, you can edit the topic names to match the wording your business uses. Be sure to inspect the list and delete any secondary or other topics that are less relevant to your needs.
This feature is available in both the Conversational Insights console and by API.
To check each topic , call the get method of the issue resource.
(Optional) Deploy and test a topic model on a sample of conversations
To further test the topic model's performance, you can deploy it and analyze a
small sample of conversations. After completion, check the topic distribution
in the Topic Model Deployed data page. Secondary topics might be the
dominant topic in deployed results because they might be common and have
stronger matches. Topics that match to a high proportion (more than 30%) of the sample
conversations are likely to be secondary topics. Carefully examine these topics
and delete them if they aren't interesting.
Whether uninteresting secondary topics exist highly depends on the input data.
If all the major topics on Deployed data have a relatively even distribution,
and each topic only matches to a small proportion (less than 20%) of conversations, then
it is very likely there are no such secondary topics that need to be deleted.
For more information on how to deploy a topic model and analyze conversations, see the topic inference documentation.
Delete a secondary topic
This feature is available in both the CCAI Insights console and by API.
To delete a topic , call the delete method of the issue
resource.
After deleting the topic, you need to reanalyze previously-labeled conversations to overwrite the previous analysis result.