REACT_RU_BOT
A chat bot for @react_ru
Features
| Feature |
|---|
| Spam removal (educated guess) |
Spam removal
❤️ Approach
I collected 11 (for now) examples of spam and trained LogisticRegressionClassifier (from natural)
If bot sees something that looks like spam it will:
- Remove message if it's sure that's spam
- Ask if that's actually spam if it's unsure
When receive answer, bot update internal classifier to not pass that message again
Motivation
Current antispam solutions offers some kind of CAPTCHA to every new member.
It's not very fair for most of the memers, so described approach are implemented.
Rationale
Evidence-based approach affects on really spammy messages
In false negative scenario bot reassuring itself by members of a group
False negative scenario are mitigated by confidence rate: if bot not sure if message is spam or not, it will ask from community