LLM as a Guide project

This repository contains data used during our experiments to identify unknown relations in an administrative corpus.

Data

• You will find multiple versions of .txt files, corresponding to the administrative corpus used; you will find the files corresponding to a random selection of sentences, and the version of this corpus based on entity templates to form clusters.
• You will also find the snapshot of the FewRel validation set we used for our experiments.
• If you need access to the dataset BizRel, please fill out the access form required by the authors available in this repo: https://github.com/Geotrend-research/business-relation-dataset.

Information

• Cluster_Nb_random corresponds to the random selection of sentences from administrative data.
• Cluster_Ent_Ent_small corresponds to the entity pairs clusters based on entity templates.
• Snapshot_FewRel_rd corresponds to the snapshot randomly extracted from the FewRel validation set.
• Administrative_relations correspond to the list of the maximum number of relations generated by LLM and validated by domain experts. We provide the original labels in French, along with the English version below.