Post++

Project Structure

• MMHS150K: Contains curated datasets and related files.
• Webscrape_result: Directory for storing web scraping results (images and annotations).
• Scripts:
  • Core Functionality:
    • gpt_editor.py: Uses OpenAI's GPT API to revise text based on guideline violations explained by the moderator.
    • gpt_moderator.py: Moderates text against community guidelines using OpenAI's GPT API.
  • Data Handling:
    • process_dataset.py: Cleans and preprocesses datasets, including label adjustments.
    • webscrape.py: Script for collecting and saving data via web scraping.
  • Pipeline Management:
    • main.py: Central script to manage dataset processing and moderation tasks.
  • Evaluation:
    • compute_metrics.py: Calculates performance metrics for text analysis or moderation tasks.
  • Support Files:
    • prompts.py: Stores reusable templates for GPT-based moderation and editing.
    • conversation_templates.py: Structures predefined conversations for moderation and editing flows.

Requirements

This project requires Python 3.10 and specific library dependencies. Ensure a proper environment setup using the instructions below.

Package Requirements

Windows Environment

pandas~=2.2.3
matplotlib~=3.9.2
numpy~=1.26.4
torch~=2.5.1+cu124
transformers~=4.46.2
torchvision~=0.20.1+cu124
openai~=0.28.0
python-dotenv~=1.0.0

macOS Environment

pandas~=2.2.3
matplotlib~=3.9.2
numpy~=1.26.4
torch~=2.5.1
transformers~=4.46.2
torchvision~=0.20.1
openai~=0.28.0
python-dotenv~=1.0.0

---

Installation

1. Clone this repository:

   git clone https://github.com/ece1786-2024/Post.git

2. Navigate to the project directory:

   cd Post

3. Create and activate a Conda environment:

   conda create -n post python=3.10
   conda activate post

4. Install dependencies:

   pip install -r requirements.txt

---

Usage

Run the Project Pipeline

1. To process datasets and run text moderation or editing tasks, execute:

   python main.py

Individual Script Execution

• Moderate a single text sample:

  python gpt_moderator.py

• Edit a text based on guideline violations:

  python gpt_editor.py

• Web scraping:

  python webscrape.py

• Compute evaluation metrics:

  python compute_metrics.py

---

Features

GPT Moderator

• Evaluates text against defined community guidelines, including:
  • Violent Content
  • Hateful Conduct
  • Abuse/Harassment
  • Child Safety
• Outputs detailed analysis, specifying:
  • Whether the text is compliant.
  • List of guideline violations.
  • Explanation for each violation.

Example Usage

Input:

moderator = GPTModerator(api_key="your_api_key")
result = moderator.moderate_text("This retard doesn't know how to play.")
print(result)

Output:

{
  "compliant": False,
  "violations": ["Hateful Conduct"],
  "explanations": "The term 'retard' is used as a derogatory insult, violating guidelines against attacking individuals based on disability."
}

---

GPT Editor

• Revises text based on moderator-provided explanations.
• Ensures compliance with guidelines while preserving the original message's intent.

Example Usage

Input:

editor = GPTEditor(api_key="your_api_key")
edited_text = editor.edit_text(
    "This retard doesn't know how to play.",
    "The term 'retard' is used as a derogatory insult, violating guidelines against attacking individuals based on disability."
)
print(edited_text)

Output:

{
  "original_text": "This retard doesn't know how to play.",
  "revised_text": "This person doesn't know how to play.",
  "explanation": "Revised the derogatory term 'retard' to 'person' to remove offensive language while maintaining the original sentence's intent."
}