The KnightsWatch

ML moderation bot for Reddit.

Uses machine learning to flag down comments that are possibly toxic.

Based from python chatbot by Tech with Tim.

Installation

1. Requires Python 3.6. Virtual environment recommended.
2. Use pip3 install -r requirements.txt to install dependencies.
3. Configure the praw.ini file, located at env/lib/python3.6/site-packages/praw, with your bot tokens. PRAW Documentation
4. Edit the settings.json to match your PRAW configuration. Copy the sample configuration from the sample-config folder.
5. Create the intents.json file in the training folder. You can copy the sample configuration from the sample-config folder or modify the categories and labels.

Training

Method 1

Use this method if there is no model trained yet. This method will gather the latest comments and output them. Sort the comments and it will automatically populate the intents.json file for you.

1. Run the collecter.py script.
2. Comments will be collected automatically and will await user input.
3. Type:
   • a: If the comment is acceptable.
   • n: If the comment is neutral.
   • w: If the comment is considered to be a warning.
   • Type any other character to skip the entry.
4. Press Enter to submit.
5. Rebuild the model as needed.

Method 2

This method will gather the latest comments and output them. It will display what it currently thinks a comment is categorized as. Sort the comments and it will automatically populate the intents.json file for you.

1. Run the collecter_trainer.py script.
2. Comments will be collected automatically and will await user input.
3. Type:
   • a: If the comment is acceptable.
   • n: If the comment is neutral.
   • w: If the comment is considered to be a warning.
   • Type any other character to skip the entry.
4. Press Enter to submit.
5. Rebuild the model as needed.

Method 3

Manual entry method.

1. Run the self_assign.py script.
2. Enter a comment, it will be sanitized and re-outputted.
3. Type:
   • a: If the comment is acceptable.
   • n: If the comment is neutral.
   • w: If the comment is considered to be a warning.
   • Type any other character to skip the entry.
4. Press Enter to submit.

Notes

• Comments are sanitized of all punctuation and potential offending characters.
• Training data and models have been removed for public release.