You think you're a good at storytelling... ehh??? Come and fight me you ....! 🦀 🦀 🦀
crabby is a literature critic of some kind. It does so by first recognising the named entities and extracting their relations from a raw text. By doing so it extracts them one by one which enables the creation of a storygraph data structure. Once that is done we have an ontology of the named entities and their relations! Isn't this great? Furthermore it employs an easy to train model for multi-relation data called Transe to create a GNN-like model for link-prediction. Since the embeddings are in a latent vector space then distances between entities could be calculated to obtain a truthfulness score for a relation. This score is then used for criticising a text regarding the story.
# Create a python venv like so.
python3 -m venv .venv
# Install the requirements like so.
(source .venv/bin/activate && pip install -r requirements.txt)
make fetch-lang-mdl
# To test relation extraction run.
# It is being thaught over the dataset from semval2010-task8.
make test-relex
# To test transe run.
make test-transe
# To run unit tests run.
make test-unit