Using chatbots to solve open world knowledge graph completion. This repository is used to document the state of the research code used for the publication of the paper Re-ranking with LLMs for Open-World Knowledge Graph Completion.
We recommend using pyenv for installation.
pyenv install 3.11
pyenv global 3.11
poetry install
poetry run ilpYou can find the source code in ./ilp and all configurations (including legacy) in conf/.
$ poetry run ilp
Usage: ilp [OPTIONS] COMMAND [ARGS]...
Use ilp from the command line.
╭─ Options ─────────────────────────────────────────────────────╮
│ --quiet -q suppress console output │
│ --debug -d activate debug mode (drop into pudb on error) │
│ --help Show this message and exit. │
╰───────────────────────────────────────────────────────────────╯To reproduce the experiments of the paper, see the scripts/ directory. Calls to ilp to run the experiments which have been conducted for the final paper results are documented in scripts/all-experiments.sh. If you want to have the flexibility and automation we worked with during development, you can execute the *.fish files. The fish shell is required to execute these scripts. There, for the IRT2 and BLP datasets, experiments are started using scripts/exp-irt-full.fish and scripts/exp-blp-full.fish. These entry points source the outher files in the directory to configure the experiment runs. However, these are only used to properly configure the ilp entry point shown above like documented in scripts/all-experiments.sh.
If you find our work useful, please consider giving us a cite. You can also always contact Felix Hamann for any comments or questions!
TBA