Contribute

Machine learning algorithms, such as the one ApriBot uses, need to be trained on labelled data: that is, posts which have been manually classified (by experts—yes, that's you!) as being either Aprimon-related or not.

So far:

• ApriBot has seen a total of 57037 posts.
• 31135 of these were marked as not needing manual review (as they are almost certainly not Aprimon-related); and
• 24763 of these have been labelled by some amazing people, of which 14487 were used to train ApriBot's current algorithm.

ApriBot's current algorithm is based on XGBoost, and has quite respectable performance: it achieves an accuracy of approximately 96% on unseen posts, or an F1 score of 93%. Generally, this accuracy doesn't seem to improve with more data: so, there's no urgent need to label more posts. However, if you want to, you can still do so—I will be very grateful, and there is a possibility that a different algorithm may be able to make use of the extra data!

Please log in with Reddit (in the top-right corner) to continue.