KurtMica's picture
Merge one-hot labels into single feature.
6bb0321 verified
metadata
language:
  - mt
license: cc-by-nc-sa-4.0
size_categories:
  - 10K<n<100K
task_categories:
  - text-classification
pretty_name: Maltese News Categories
dataset_info:
  features:
    - name: url
      dtype: string
    - name: title
      dtype: string
    - name: base_url
      dtype: string
    - name: text
      dtype: string
    - name: labels
      sequence:
        class_label:
          names:
            '0': court
            '1': covid
            '2': culture
            '3': eu
            '4': economy
            '5': education
            '6': entertainment
            '7': environment
            '8': health
            '9': immigration
            '10': international
            '11': opinion
            '12': politics
            '13': religion
            '14': social
            '15': sports
            '16': transport
  splits:
    - name: train
      num_bytes: 19700614
      num_examples: 10784
    - name: validation
      num_bytes: 4286743
      num_examples: 2293
    - name: test
      num_bytes: 4560168
      num_examples: 2297
  download_size: 16511339
  dataset_size: 28547525
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
      - split: validation
        path: data/validation-*
      - split: test
        path: data/test-*

Maltese News Categories

A multi-label topic classification dataset for Maltese News Articles.

Data Collection

The data was collected from the press_mt subset from Korpus Malti v4.0. Article contents were cleaned to filter out JavaScript, CSS, & repeated non-Maltese sub-headings. The labels are based on the category field from this corpus. Additional filtering & cleaning was performed as follows:

  1. Documents with generic categories (News, Local, Headlines, Uncategorised, Archived) were ignored.
  2. Some categories were merged as deemed necessary by native Maltese speakers to standardise tags assigned by different news portals (for example merging European Union & Unjoni Ewropea under EU).
  3. Articles with categories having less than 100 articles were ignored.
  4. Categories having a >75% co-occurence with another category are dropped.

The different tags & their distribution is shown in the following table.

Tag Token Count Article Count Article Percentage
Court 404,329 860 5.59
Covid 458,120 1,735 11.29
Culture 750,406 2,186 14.22
EU 93,227 240 1.56
Economy 112,972 321 2.09
Education 56,084 191 1.24
Entertainment 837,248 3,147 20.47
Environment 38,522 147 0.96
Health 81,630 290 1.89
Immigration 29,665 120 0.78
International 784,878 3,957 25.74
Opinion 231,266 321 2.09
Politics 682,007 1,294 8.42
Religion 186,300 465 3.02
Social 98,127 203 1.32
Sports 835,484 3,066 19.94
Transport 74,959 241 1.57

Additional Information

License

This work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License. Permissions beyond the scope of this license may be available at https://mlrs.research.um.edu.mt/.

CC BY-NC-SA 4.0

Citation

This work was first presented in Topic Classification and Headline Generation for Maltese using a Public News Corpus. Cite it as follows:

@inproceedings{maltese-news-datasets,
    title = "Topic Classification and Headline Generation for {M}altese using a Public News Corpus",
    author = "Chaudhary, Amit Kumar  and
              Micallef, Kurt  and
              Borg, Claudia",
    booktitle = "Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation",
    month = may,
    year = "2024",
    publisher = "Association for Computational Linguistics",
}