A newer version of the Streamlit SDK is available:
1.52.1
title: EU Legal Recommender System
emoji: 📚
colorFrom: blue
colorTo: indigo
sdk: streamlit
sdk_version: 1.24.0
app_file: recommender/streamlit_app/app.py
pinned: false
EU Legal Recommender System
A recommender system for EU legal documents that provides personalized recommendations based on semantic similarity and categorical features. The system includes a complete pipeline from data scraping to interactive recommendation delivery via a Streamlit web application.
Thesis Context
Student:
Alexander Benady
University:
IE University - Bachelor in Data and Business Analytics (BDBA)
Thesis Advisor:
Prof. Borja González del Regueral
Partner Organisation:
Vinces Consulting
About This Demo
This Hugging Face Space provides a demonstration of the recommender component of the full EU Legal Recommender system. The system features:
- Legal-BERT Embeddings: Specialized embeddings for legal text understanding
- Hybrid Recommendation: Combines semantic similarity with categorical preferences
- Personalized Profiles: Customizable client profiles with industry-specific preferences
- Interactive Web Interface: Explore recommendations through a user-friendly Streamlit application
Using This Demo
You can:
- Enter queries to find relevant EU legal documents
- Select from pre-configured client profiles for industry-specific recommendations
- View document details and summaries
- Explore similar documents
Client Profiles
The system includes pre-configured client profiles for various industries. These profiles demonstrate how the recommender can be tailored to specific business needs.
Each profile contains:
- An expert description of business interests
- Historical documents the client has engaged with
- Categorical preferences for document types, subject matters, etc.
- Component weights for the recommendation algorithm
Full Project
The complete project, including the scraper and summarization components, is available on GitHub:
https://github.com/alexanderbenadyieu/eu-legal-recommender
Acknowledgements
This project was developed by Alexander Benady as part of an undergraduate thesis at IE University in collaboration with Vinces Consulting.