ryannli/cs152bots

Starter code for a Slack Bot.

CS 152 - Trust and Safety Engineering

Discord Bot Framework Code

This is the base framework for students to complete Milestone 2 of the CS 152 final project. Please follow the instructions you were provided to fork this repository into your own repository and make all of your additions there.

Install libraries

python3 -m pip install requests
python3 -m pip install discord.py
python3 -m pip install alt-profanity-check
python3 -m pip install scipy
python3 -m pip install editdistance
python3 -m pip install openai
python3 -m pip install argparse

If there are errors from scipy, try uninstall it first python3 -m pip uninstall scipy and then reinstall python3 -m pip install scipy.

Run the Discord bot by

python3 bot.py

Run the Discord bot with OpenAI detection

python3 bot.py --openai=true

Run the Discord bot with OpenAI detection and debugging mode

python3 bot.py --openai=true --debug=true

Example for using formatter library:

>>> import formatter
>>> dict = {'A': 5, 'B': "3", 'C': "DD"}
>>> str = formatter.format_dict_to_str(dict)
>>> str
'`A`: 5\n`B`: 3\n`C`: DD\n'
>>> formatter.unformat_str_to_dict(str)
OrderedDict([('A', '5'), ('B', '3'), ('C', 'DD')])

Current implementated functions:

1. Basic user reporting flow. Follow Milestone 2 guidelines to test on Discord.
2. Automatically detects offensive messages by using profanity-check library.
   • If score > 0.95, deletes the message directly (like "fuck you")
   • Else, if score > 0.4, forward the message to moderator (like "I hate you"). Example message:
     • reporter: SYSTEM AUTOMATIC
     • author: My Name
     • message: i'll kill you
     • link: https://discord.com/channels/1103033282779676743/1103033284834902076/1108947923355574352
     • metadata: profanity_check score is 0.83
3. Automatically detects offensive messages by using OpenAI. Detect for 9 different abusive categories and rate in scale 1-5 (from negligible to severe)
   • If at least 2 categories has score>=4, deletes the message directly (like "fuck you")
   • Else, if at least 1 category has score>=3, forward the message to moderator (like "I hate you"). Example message:
     • reporter: SYSTEM AUTOMATIC
     • author: My Name
     • message: i'll kill you
     • link: https://discord.com/channels/1103033282779676743/1103033284834902076/1108947923355574352
     • metadata: OpenAI detected harmful score (scale 1-5) is ...
4. 2 adversarial strategies:
   • After message is edited, detects the offensive contents again.
5. User reports go to mod channel. Example message:
   • reporter: 81149813901741260812
   • author: My Name
   • message: i'll kill you
   • link: https://discord.com/channels/1103033282779676743/1103033284834902076/1108947923355574352
   • metadata: profanity_check score is 0.83