|
|
--- |
|
|
title: Recommendation LLM Explainer |
|
|
emoji: 🎬 |
|
|
colorFrom: blue |
|
|
colorTo: purple |
|
|
sdk: gradio |
|
|
sdk_version: 6.3.0 |
|
|
app_file: app.py |
|
|
pinned: false |
|
|
--- |
|
|
|
|
|
# AI-Powered Explainable Recommendations |
|
|
|
|
|
This project is a high-fidelity demonstration of how Large Language Models (LLMs) can bridge the "explainability gap" in recommendation systems. It transforms opaque collaborative filtering data into transparent, persona-aligned narratives. |
|
|
|
|
|
 |
|
|
|
|
|
## 🚀 Overview |
|
|
|
|
|
Instead of providing a "black box" list of suggestions, this system uses LLMs to synthesize natural language explanations based on user personas and collaborative filtering logic. |
|
|
|
|
|
### Key Features |
|
|
- **Specialist Centroids**: Personas built by aggregating top "specialists" for specific genres. |
|
|
- **Layered Re-ranking**: Incorporates Genre Affinity Boosts and Recency Bias. |
|
|
- **Narrative Synthesis**: Uses LiteLLM to generate 3-sentence explanations in "Logic-Driven" or "Social-Behavioral" styles. |
|
|
|
|
|
## 🛠 Tech Stack |
|
|
|
|
|
- **Interface**: [Gradio](https://gradio.app/) |
|
|
- **Recommendation Engine**: Python (`pandas`, `numpy`, `scikit-learn`) |
|
|
- **LLM Integration**: [LiteLLM](https://github.com/BerriAI/litellm) |
|
|
- **Data Source**: [MovieLens Latest Small](https://grouplens.org/datasets/movielens/latest/) |
|
|
- **Visuals**: [TMDB API](https://www.themoviedb.org/documentation/api) for movie posters. |
|
|
|
|
|
## 📁 Quick Start |
|
|
|
|
|
1. **Environment Setup**: |
|
|
Create a `.env` file: |
|
|
```env |
|
|
HF_TOKEN=your_huggingface_token |
|
|
TMDB_API_KEY=your_tmdb_api_key |
|
|
``` |
|
|
2. **Install Dependencies**: |
|
|
```bash |
|
|
pip install -r requirements.txt |
|
|
``` |
|
|
3. **Data Preparation**: |
|
|
```bash |
|
|
python data_prep.py |
|
|
``` |
|
|
4. **Run the Application**: |
|
|
```bash |
|
|
python app.py |
|
|
``` |
|
|
|
|
|
## 📜 Acknowledgments & Data Sources |
|
|
|
|
|
- **Dataset**: This project uses the **MovieLens Latest Dataset - Small** provided by [GroupLens Research](https://grouplens.org/datasets/movielens/). |
|
|
- **Images**: Movie posters are retrieved using the **TMDB API**. *Note: This product uses the TMDB API but is not endorsed or certified by TMDB.* |
|
|
|
