/beHappy-Ai4Youth

Ajajajajaj

Primary LanguageJupyter Notebook

Ai4Youth - beHappy

General Information

  • Project name: beHappy
  • Authors: Jakub Stanula-Kaczka,Wiktor Koźlik,Piotr Chajec
  • Project form: Google Chrome extension with predicting model exposed on localhost moderating reddit.com comments.

Installation & Dependences

  • python
    • tensorflow
    • keras
    • numpy
    • sklearn
    • nltk
    • pandas
    • matplotlib
    • flask
    • flask_cors

Install all required dependencies with:

pip install -r requirements.txt

Dataset

Name Download
Twitter Toxicity Sentiment Analysis link

Use

  • Train

    Run all cells in .ipynb notebook model and tokenizer will be saved in it's directory

  • Test & Run

    In order to load extension go to Extensions -> Set Developer Mode in upper right cornet to On -> On Upper left corner select this repository directory by clicking on Load unpacked

    Generally speaking you don't need to train model from scratch. You could use pretrained models contained in /python_api directory. Python script uses relative path so it does need to be run in folder with model and tokenizers files. You can test different models by changing it's names in main.py file.

    cd python_api
python main.py

Important Information

  • By default all paragraphs containing offensive language are blured. To see moderated content simply hower mouse on it.
  • It is possible to disable blurring completely. See extension menu for futher details.

Directory Hierarchy

|—— .DS_Store
|—— .gitignore
|—— background.js
|—— Ai4Youth_beHappy.ipynb
|—— beHappy in English.pptx
|—— beHappy po Polsku.pptx
|—— blur.css
|—— contentScript.js
|—— content_img
|    |—— 0.png
|    \—— 1.png
|—— images
|    |—— ai_icon128.png
|    |—— ai_icon16.png
|    |—— ai_icon32.png
|    |—— ai_icon48.png
|    \—— ai_icon64.png
|—— manifest.json
|—— popup.css
|—— popup.html
|—— popup.js
\—— python_api
     |—— Dense128.h5
     |—— Dense32-new-tfidf.h5
     |—— Dense32.h5
     |—— Dense64.h5
     |—— Dense8-4.h5
     |—— main.py
     |—— model.h5
     |—— model2.h5
     |—— tfidf-2.dat
     |—— tfidf-3.dat
     |—— tfidf.dat
     \—— tokenizer_twitter_save.dat

Code Details

Tested Platform

  • software 
    OS: macOS Monterey 12.5, Windows 10
Python: 3.9+
Tensorflow: latest compiled version as for 16.10.2022, 2.8.2
  • hardware 
    CPUs: Apple Silicon M1 8 cores, Intel i5-9300h, Intel i5-7400
GPUs: Apple Silicon M1 8 cores, Nvidia GeForce GTX 1650, Nvidia GTX 1660 Super
  • RAM 
    Trained on platforms with at least 12GB of RAM

References