|
|
|
|
|
---
|
|
|
title: Falcon
|
|
|
emoji: 🔥
|
|
|
colorFrom: indigo
|
|
|
colorTo: yellow
|
|
|
sdk: docker
|
|
|
pinned: false
|
|
|
---
|
|
|
|
|
|
# Falcon: Fake News Analysis and Language Comprehension for Online Neutrality
|
|
|
|
|
|
**Falcon** is an advanced Gen AI-powered system designed to analyze, verify, and classify online claims to promote digital content neutrality. Leveraging state-of-the-art language models, it classifies user-submitted text, detects tone and intent, verifies facts using multiple data sources, and generates an informed verdict about claim credibility.
|
|
|
|
|
|
---
|
|
|
|
|
|
## Features
|
|
|
|
|
|
* **Claim Classification**: Classifies input text into categories like Factual Claim, Opinion, Irrelevant Talk, or Vague/Incomplete.
|
|
|
* **Tone and Intent Detection**: Analyzes the emotional tone and intent behind the claim (e.g., neutral, persuasive, humorous).
|
|
|
* **Fact Verification**: Verifies factual claims by searching reliable sources like Google (via Serper API) and Wikipedia.
|
|
|
* **Verdict Generation**: Produces a clear verdict on the claim’s truthfulness based on gathered evidence and reasoning.
|
|
|
* **Fallback Language Model**: Uses OpenAI’s GPT-4 as the primary LLM and automatically switches to **Deepseek** as a fallback model to ensure high availability and robustness.
|
|
|
* **Simple Web Interface**: Easy-to-use frontend using HTML, CSS, and JavaScript for seamless user interaction.
|
|
|
* **FastAPI Backend**: Fast, lightweight API server handling all processing with asynchronous endpoints.
|
|
|
|
|
|
---
|
|
|
|
|
|
## Tech Stack
|
|
|
|
|
|
* **Backend**: FastAPI (Python)
|
|
|
* **Frontend**: HTML, CSS, JavaScript
|
|
|
* **Language Models**: OpenAI GPT-4 (primary), Deepseek (fallback)
|
|
|
* **APIs**: Serper API (Google Search), Wikipedia API
|
|
|
* **Prompt Engineering**: Carefully crafted templates for accurate and reliable results
|
|
|
* **Deployment**: Docker containerization (optional)
|
|
|
|
|
|
---
|
|
|
|
|
|
## Project Structure
|
|
|
|
|
|
```
|
|
|
falcon/
|
|
|
├── backend/
|
|
|
│ ├── api/
|
|
|
│ │ ├── claims.py # Claim classification logic
|
|
|
│ │ ├── fact_check.py # Fact verification logic
|
|
|
│ │ ├── tone_intent.py # Tone and intent detection logic
|
|
|
│ ├── langchain_tools.py # LangChain prompt templates and LLM wrappers
|
|
|
│ ├── config.py # API keys and config variables
|
|
|
│ └── main.py # FastAPI app entry point
|
|
|
├── frontend/
|
|
|
│ ├── index.html # Main UI page
|
|
|
│ ├── assets/ # CSS, JS, images
|
|
|
├── Dockerfile # Docker container definition (optional)
|
|
|
└── README.md # This file
|
|
|
```
|
|
|
|
|
|
---
|
|
|
|
|
|
## Setup Instructions
|
|
|
|
|
|
### Prerequisites
|
|
|
|
|
|
* Python 3.9+
|
|
|
* FastAPI
|
|
|
* Uvicorn (ASGI server)
|
|
|
* OpenAI API key
|
|
|
* Serper API key
|
|
|
* Deepseek API key (for fallback)
|
|
|
|
|
|
### Installation
|
|
|
|
|
|
1. Clone the repository:
|
|
|
|
|
|
```bash
|
|
|
git clone https://github.com/yourusername/falcon.git
|
|
|
cd falcon
|
|
|
```
|
|
|
|
|
|
2. Create and activate a virtual environment:
|
|
|
|
|
|
```bash
|
|
|
python -m venv venv
|
|
|
source venv/bin/activate # On Windows: venv\Scripts\activate
|
|
|
```
|
|
|
|
|
|
3. Install dependencies:
|
|
|
|
|
|
```bash
|
|
|
pip install -r requirements.txt
|
|
|
```
|
|
|
|
|
|
4. Configure API keys in `backend/config.py`:
|
|
|
|
|
|
```python
|
|
|
OPENAI_API_KEY = "your_openai_api_key"
|
|
|
SERPER_API_KEY = "your_serper_api_key"
|
|
|
DEEPSEEK_API_KEY = "your_deepseek_api_key"
|
|
|
```
|
|
|
|
|
|
5. Run the FastAPI server:
|
|
|
|
|
|
```bash
|
|
|
uvicorn backend.main:app --port 8000
|
|
|
```
|
|
|
|
|
|
6. Open your browser and visit `http://localhost:8000` to use the Falcon app.
|
|
|
|
|
|
---
|
|
|
|
|
|
## Usage
|
|
|
|
|
|
* Enter a claim or statement in the input form.
|
|
|
* The system classifies the claim, detects tone and intent, performs fact verification if applicable, and returns a detailed verdict with supporting evidence.
|
|
|
* The fallback model Deepseek is automatically used if GPT-4 is unavailable, ensuring consistent performance.
|
|
|
|
|
|
|
