Datasets:
language:
- en
license: cc-by-4.0
task_categories:
- token-classification
dataset_info:
features:
- name: context
sequence: string
- name: label
sequence: float64
- name: domain
dtype: string
splits:
- name: test
num_bytes: 46345
num_examples: 50
download_size: 15462
dataset_size: 46345
configs:
- config_name: default
data_files:
- split: test
path: data/test-*
tags:
- word importance
Word Importance
Dataset Description
- Repository: https://github.com/adam-osusky/predicting-word-importance
- Paper: TODO
Dataset Summary
The Word Importance dataset consists of short contexts, approximately 50 words in length, along with annotations indicating the importance of words within these contexts. Annotators were tasked with ranking the top 10% of important words within each context.
Any words left unranked by the user received the same last rank. For instance, if a user selected 5 words, the remaining words were assigned a rank of 6.
Moreover, multiple users contributed rankings for each context, and the final ranking for a context was computed by averaging these contributions.
The dataset is designed to facilitate research in word importance prediction and token classification tasks. For further details on annotation instructions and methodology, refer to the associated paper (link to be provided when available).
Supported Tasks
Given its small size, the dataset is primarily intended for evaluating models that predict word importance scores. For code related to evaluation, please refer to [https://github.com/adam-osusky/predicting-word-importance].
Languages
All the text is in english. And it consists of 5 domains: news, beletry, poetry, jokes, and transcribed spoken language
Additional Information
Licensing Information
This work is licensed under CC BY 4.0
Citation Information
@article{wordimp-osus,
author = {Adam Osuský},
title = {Predicting Word Importance Using Pre-Trained Language Models},
school = {Charles University, Faculty of Mathematics and Physics},
year = {2024},
type = {Bachelor's Thesis},
}