Itepee — Indonesia Tweet Prediction

Hate speech detection in Indonesian tweets using 4 TensorFlow/Keras deep learning models.

Table of Contents

• Overview
• Tech Stack
• Project Structure
• Dataset
• Getting Started
• Author

---

Overview

This system classifies Indonesian tweets into 4 aspects of hate speech using an ensemble of 4 deep learning models.

Pipeline:

1. Preprocessing — Text cleaning, slang normalization, stopword removal
2. Detection — Binary classification: hate speech or non-hate speech
3. Multi-label Classification — Target (Individual/Group), Type (Religion/Race/Physical/Gender/Other), Level (Weak/Moderate/Strong)
4. Web Interface — Django with Bootstrap 5

Tech Stack

Category	Tools
Language	Python 3.11
Web Framework	Django 4.1
Deep Learning	TensorFlow / Keras (4 models)
Frontend	Bootstrap 5, Inter font
Database	SQLite (dev) / MySQL
Data Processing	Pandas, NumPy

Project Structure

Itepee/
├── Frontend/
│   ├── Layout/main.html       # Main layout (navbar, footer)
│   ├── index.html             # Landing page
│   ├── predict.html           # Tweet input form
│   ├── hate_speech.html       # Result: HS detected
│   ├── non_hate_speech.html   # Result: safe
│   └── detail.html            # Probability detail + progress bar
├── Itepee/
│   ├── settings.py            # Django configuration
│   ├── urls.py                # URL routing
│   ├── asgi.py / wsgi.py
├── ItepeeApp/
│   ├── views.py               # Prediction & rendering logic
│   ├── preprocessor.py        # Text preprocessing (8 steps)
│   ├── models.py              # Database model (TweetModel)
│   ├── admin.py               # Django admin
│   ├── model/                 # 4 .h5 models + tokenizer pickle
│   └── data/                  # Slang dictionary, stopwords, HTML entities
├── static/
│   └── style.css              # Custom styling
├── .env.example
├── manage.py
└── README.md

Dataset

• Source: id-multi-label-hate-speech-and-abusive-language-detection by Okky Ibrohim & Indra Budi
• License: CC BY-NC-SA 4.0

Labels:

#	Label	Description
1	HS	Hate speech
2	Abusive	Abusive language
3	HS_Individual	Targeted at individual
4	HS_Group	Targeted at group
5	HS_Religion	Religion
6	HS_Race	Race/ethnicity
7	HS_Physical	Physical/disability
8	HS_Gender	Gender/sexual orientation
9	HS_Weak	Weak level
10	HS_Moderate	Moderate level
11	HS_Strong	Strong level

Paper: Multi-label Hate Speech and Abusive Language Detection in Indonesian Twitter — ALW3 2019

Getting Started

1. Clone

git clone https://github.com/sintiasnn/Itepee.git
cd Itepee

2. Setup Environment

Using conda (recommended)

conda activate satelit

If satelit environment doesn't exist:

conda create -n satelit python=3.11 -y
conda activate satelit
pip install tensorflow django pandas numpy python-decouple
pip install tf_keras keras-preprocessing

Using venv

python -m venv venv
source venv/bin/activate    # macOS/Linux
# venv\Scripts\activate     # Windows
pip install -r requirements.txt

3. Migrate Database

python manage.py migrate

4. Run Server

python manage.py runserver

Open http://localhost:8000

Author

Ni Putu Sintia Wati

• GitHub: @sintiasnn

Contributors:

• Kaenova Mahendra
• Kelvin Mulyawan
• Adrianus Wicaksono