Efficient Nearest Neighbor based Uncertainty Estimation for Natural Language Processing Tasks

Code for "Efficient Nearest Neighbor based Uncertainty Estimation for Natural Language Processing Tasks" in Findings of NAACL 2025.

Directory Structure

.
├── classification/  # classification tasks (sentiment analysis and natural language inference)
└── ner/  # named entity recognition

Usages

Build environments

We used Singularity environment.

module load singularity
singularity build --fakeroot research-dev.sif research-dev.def

Prepare datasets

Please see data/README.md.

Run for classification Tasks

cd classification
sbatch batch_run_custom_classification.sh

Run for NER

cd ner
sbatch batch_run_custom_ner.sh