Skip to content

Latest commit

 

History

History
62 lines (45 loc) · 2.09 KB

CONTRIBUTING.md

File metadata and controls

62 lines (45 loc) · 2.09 KB
Raw

Contributing to AI Mod

I encourage anyone interested in this project to contribute. There are many ways large and small to help out; from creating content classifiers to adding new swear words to the filter. This document provides some guidance.

Development Guidelines

The general structure of AI Mod's content classification is to bring together multiple separate 'classifiers' that determine if a text string is nsfw / spam / etc and a master method to combine all of those reports into one.

On top of that we have a web-facing API and a Reddit interface that both make use of those reports. If you want to contribute to AI Mod here are a few suggestions:

  • Adding Classifiers
  • Extending the Reddit interface
  • Improving the ai-mod.com website
  • Integrating third party APIs and libraries

Adding New Classifiers

A classifier takes an arbitrary string and returns returns some data about it, specifically the confidence / probability (0.0 to 1.0) that it falls into one or more categories, and maybe a few notes.

So in order to easily integrate many classifiers, I'm requiring that they all return a string like the following:

{
   "nsfw":0.25,
   "spam":0.92,
   "notes":["Contains Swear Words"]
}

The proposed categories are notes, nsfw, spam, troll, doxx, rules, bot, repost, other

A string to be analyzed will be run through every classifier and combined. If multiple classifiers output info in the same category, their notes will be combined and confidence taken from the highest one.

Let's look at a very basic classifier, this will classify your text as NSFW if it contains the word "shit":

def detect_nsfw_language(text):
    report = {'nsfw': 0.0, 'notes':[]}
    if "shit" in text:
        report['nsfw'] = 1.0
        report['notes'].append("NSFW Language Detected!")
return report

Now I'll add the method to the analysis_methods array in text_analysis.py and run this string:

analyze_text("You're full of shit")

{'notes': ['NSFW Language Detected!'], 'nsfw': 1.0}

And there you have it. It's integrated with the collective.