Classification in the Discourse Sentiment Dashboard

Hi everyone,

I’m exploring the sentiment analysis feature available in the Discourse admin dashboard (Admin > Dashboard > Sentiment) and I have questions about how it works:

  1. How is sentiment assigned to posts? Is it based on a machine learning model, keyword matching, or a combination of both?
  2. Are there specific keywords or phrases that trigger a particular sentiment? If so, is there a way to view, edit or export the full list?

Thanks in advance.
Laura

Sentiment is part of the ai plugin. You can read more about it here:

Thank you. Is there a way to adjust this?

We’ve noticed that sentiment is only assigned to a small number of posts. Since we find this feature very helpful, we were wondering whether there is a way to expand or configure the sentiment classification so that it applies to more posts.

Alternatively, is it possible to have sentiment determined by users (for example, members of our community) rather than by AI? If that’s not currently supported, would there be a way for us to manually assign or manage sentiment classifications ourselves?

Backfill for older postsare controlled by this setting:

As for customizing this… the agent used for Sentiment can be found in the AI > Agents tab, under the name ‘Sentiment classifier’.

You can also edit Sentiment-related settings at /admin/plugins/discourse-ai/settings?filter=sentiment.