README.md

metadata

language:
  - tl
pretty_name: >-
  Filipino multi-modal NLP dataset. Consists of 350k+ Filipino news articles and
  associated images
size_categories:
  - 100K<n<1M
source_datasets:
  - original
tags:
  - news
task_categories:
  - text-to-image
  - image-to-text
  - text-generation
  - summarization
  - text-classification
  - text2text-generation
task_ids:
  - news-articles-headline-generation
  - news-articles-summarization
dataset_info:
  - config_name: default
    features:
      - name: title
        dtype: string
      - name: body
        sequence: string
      - name: image
        dtype: image
      - name: website
        dtype: string
      - name: category_group
        dtype: string
      - name: category
        dtype: string
      - name: title_choice_first_paragraph
        dtype: string
      - name: title_choices
        sequence: string
      - name: title_choice_gold_idx
        dtype: int32
      - name: date
        dtype: string
      - name: author
        dtype: string
      - name: url
        dtype: string
      - name: img_url
        dtype: string
    splits:
      - name: train
        num_bytes: 39449948960.917
        num_examples: 281403
      - name: validation
        num_bytes: 5093856283.5
        num_examples: 35175
      - name: test
        num_bytes: 4923596011.806
        num_examples: 35177
    download_size: 39815624261
    dataset_size: 49467401256.223
  - config_name: no-image
    features:
      - name: title
        dtype: string
      - name: body
        sequence: string
      - name: category_group
        dtype: string
      - name: category
        dtype: string
      - name: website
        dtype: string
      - name: title_choice_first_paragraph
        dtype: string
      - name: title_choices
        sequence: string
      - name: title_choice_gold_idx
        dtype: int32
      - name: date
        dtype: string
      - name: author
        dtype: string
      - name: url
        dtype: string
      - name: img_url
        dtype: string
    splits:
      - name: train
        num_bytes: 578462672
        num_examples: 281403
      - name: validation
        num_bytes: 74036069
        num_examples: 35175
      - name: test
        num_bytes: 73488921
        num_examples: 35177
    download_size: 427915786
    dataset_size: 725987662
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
      - split: validation
        path: data/validation-*
      - split: test
        path: data/test-*
  - config_name: no-image
    data_files:
      - split: train
        path: no-image/train-*
      - split: validation
        path: no-image/validation-*
      - split: test
        path: no-image/test-*

A Filipino multi-modal language dataset for text+visual tasks. Consists of 351,755 Filipino news articles (w/ associated images) gathered from Filipino news outlets.

Description

Total # of articles: 351,755

80-10-10 split for training, validation, and testing.

Dataset field descriptions:

title                        - Article title  
image                        - Article image  
website                      - Name of the news outlet  
category_group               - Category grouped into 5 distinct classes. News, Sports, Entertainment, Crime, and Other  
category                     - News category name given by the news outlet  
date                         - Date published  
author                       - Article author  
url                          - URL of the article  
img_url                      - URL of the article image  
title_choice_first_paragraph - Opening paragraph of the article
title_choices                - 4 possible titles, one of them being the true one
title_choice_gold_idx        - Idx of the true title among the choices

title_choice_* fields are included to support the task of textual entailment — taking advantage of the "inverted pyramid" structure of news articles.

Dataset Usage

Two dataset configurations: default (includes images) and no-image (excludes images)

Using datasets library

default

from datasets import load_dataset
dset = load_dataset('LanceBunag/BalitaNLP', streaming=True) # streaming recommended due to size of dataset w/ images

no-image

from datasets import load_dataset
dset = load_dataset('LanceBunag/BalitaNLP', 'no-image')

Citation

Published in Buñag & Esquivel, 2023. If you are using BalitaNLP in your work, please cite the following:

@inproceedings{bunagtransformer,
  author={Bunag, Kenrick Lance T and Esquivel, Rosanna A}
  title={Transformer-Based Conditional Language Models to Generate Filipino News Articles},
  year = {2023},
  publisher = {IEOM Society International},
  url = {https://ieomsociety.org/proceedings/2023manila/595.pdf},
  booktitle = {Proceedings of the International Conference on Industrial Engineering and Operations Management},
  pages = {2231–2237},
  numpages = {7},
  location = {Manila, Philippines},
}