annotations_creators:
- other
language:
- en
language_creators:
- found
license:
- afl-3.0
multilinguality:
- monolingual
pretty_name: News Dataset
size_categories:
- 1K<n<10K
source_datasets:
- original
tags: []
task_categories:
- text-classification
task_ids:
- topic-classification
- multi-class-classification
Dataset Card for news-data
Table of Contents
Dataset Summary
The News Dataset is an English-language dataset containing just over 4k unique news articles scrapped from AriseTv- One of the most popular news television in Nigeria.
Supported Tasks and Leaderboards
It supports news article classification into different categories.
Languages
English
Dataset Structure
Data Instances
''' {'Title': 'Nigeria: APC Yet to Zone Party Positions Ahead of Convention' 'Excerpt': 'The leadership of the All Progressives Congress (APC), has denied reports that it had zoned some party positions ahead of' 'Category': 'politics' 'labels': 2} '''
Data Fields
- Title: a string containing the title of a news title as shown
- Excerpt: a string containing a short extract from the body of the news
- Category: a string that tells the category of an example (string label)
- labels: integer telling the class of an example (label)
Data Splits
| Dataset Split | Number of instances in split |
|---|---|
| Train | 4,594 |
| Paragraph | 811 |
Dataset Creation
Source Data
Initial Data Collection and Normalization
The code for the dataset creation at https://github.com/chimaobi-okite/NLP-Projects-Competitions/blob/main/NewsCategorization/Data/NewsDataScraping.ipynb. The examples were scrapped from https://www.arise.tv/
Annotations
Annotation process
The annotation is based on the news category in the arisetv website
Who are the annotators?
Journalists at arisetv
Considerations for Using the Data
Social Impact of Dataset
The purpose of this dataset is to help develop models that can classify news articles into categories.
This task is useful for efficiently presenting information given a large quantity of text. It should be made clear that any summarizations produced by models trained on this dataset are reflective of the language used in the articles, but are in fact automatically generated.
Discussion of Biases
This data is biased towards news happenings in Nigeria but the model built using it can as well classify news from other parts of the world with a slight degradation in performance.
Dataset Curators
The dataset is created by people at arise but was scrapped by @github-chimaobi-okite