Toxicity detection and moderation assistance bot
Arm of State is designed to be a modular, intelligent moderation bot. At the bot's core sits a GRU network, tasked with processing every non-bot post and determining if it is toxic. The bot keeps a record of every user's toxicity score, changing it accordingly.
The GRU network in the current bot version (V0.5 Alpha) is heavily flawed due to data corruption which went undetected during training. It is scheduled to be replaced by Google's BERT network.
Due to the flaws, we have not implemented any auto-moderation, and have limited ourselves to merely tracking toxicity. However, in the future, the bot is planned to receive the ability to automatically quarantine extremely toxic users in restricted channels, or autoban if certain types of toxicity (for example, racially targeted harassment) are detected repeatedly.
