SPACE-KBP: A Dataset for LLM-Based Knowledge Base Population in the Space Mission Domain

This repository contains necessary code to reproduce the experiments in the paper. The structure of the repo is:

1. missionCorpus: scripts to gather the text corpus used in the experiments.
2. eohandbook: Contains the excel files from CEOS database with structured information that we use for evaluation.
3. syntheticDataset: Jupyter notebooks to generate and validate the training dataset. The dataset contains mission descriptions and turtle code modeling the ontology instances.
4. SFTPythiaModels: Jupyter notebooks to train pythia models on the KBP task and evaluate the generated turtle code using CEOS database as reference data.

In addition, we release the turtle serialization of the mission ontology in the spaceOnt.ttl file. The rests of the files in the root folder are used to build the prompts that we use in the generation of the dataset and when training the models.