Varun Srinivasan
@v
We're making the Warpcast spam dataset public. Over 400,000 accounts have been processed by our model, which determines the accounts that are most likely to generate inauthentic content or unwanted notifications. https://github.com/warpcast/labels
36 replies
80 recasts
283 reactions
Varun Srinivasan
@v
Developers can use this to protect their apps from spammy users. Spam labels are provided as a JSONL file which follows the FIP: Labels specification (still in review). Data will be updated weekly with the latest labels.
2 replies
0 recast
14 reactions
Varun Srinivasan
@v
While we've taken a lot of care to correct mistakes, its possible that a small number of legitimate accounts are misclassified. If you notice this, please reply to this thread or DM me. We will use these reports to improve the model.
2 replies
0 recast
15 reactions
mcilroyc
@mcilroyc
I’ve been unfairly maligned (IMHO). I sent a DM too.
0 reply
0 recast
0 reaction
mcilroyc
@mcilroyc
Cc @pichi
0 reply
0 recast
0 reaction