mlcolab/learning-dependencies

Prompting Large Language Models for Concept Dependency Graph Extraction

This research project by Dominik Glandorf and Anastasiia Alekseeva tries to extract knowledge from LLMs and creates baseline for the evaluation from textbooks as well as Wikipedia.

Report

The paper about the project can be found here.

Setup

The project is based on Python. You can install the dependencies conveniently using conda by running within the cloned repository:

conda create
conda activate LD-inference

The analysis can be reproduced using the experiment files in the folder /exp. To get Wikipedia article annotations for a textbook fulltime, copy .env.example to .env, obtain a user key for Wikifier API.

The interactive dashboard is located in the subfolder /dash and can be started within the conda enviroment executing:

python app.py