Topic Categorisation

Topics allow you to assign labels to ingested transcripts based on various types of logic.

Groups of labelled transcripts can be used to identify how people are using your application and specifically why users are engaging with your product. It is then possible to look at success metrics such as user feedback ratings or user input sentiment per-topic, to understand which types of interaction are leaving your users satisfied, and where they're dissatisfied.

There are 2 approaches that can be used to assign a topic to a conversation message:

  1. LLM Topic - Use a Large Language Model to apply custom instructions for matching conversation messages. There are two default templates, intent and semantic matching. When a more custom solution for matching is required, the freeform option allows for full customization on prompt design. To learn more see LLM Topics.

  2. Keyword Topic - Keyword topics use exact keyword matching to match against a provided string.

Last updated