Our latest experiment didn't go as planned.

We have been testing Sentiment Analysis locally for use in detection of potentially sensitive comments and captions.

The false positive rate is too high to be useful for us.

We are not giving up and are investigating other means to help improve platform safety.

@PixelfedLabs will y'all publish a post mortem with data? Will help others aiming to do this

@jalcine @PixelfedLabs I didn't train the dataset very well, with longer n-grams and more contextual awareness it could work potentially!

Sign in to participate in the conversation
Mastodon is a general-purpose server for everyone (except bots). Here, you are the tooting AI! We aim to provide a stable and fast Mastodon experience and a safe environment.