You need to agree to share your contact information to access this dataset

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

Log in or Sign Up to review the conditions and access this dataset content.

Dataset Card for [np20ng]

Dataset Summary

This is a multi-class Nepali text classification dataset. Text are the news documents and labels are the news categories. It consists over 200,000 documents categorized into 20 different Nepali news groups. News documents from 10 different news sources are compiled into this dataset. Labeling is done using the category-specific news from the respective news portals.

Supported Tasks and Leaderboards

  • Multi-class text classification from news document
  • Multi-class text classification from news headings
  • News heading generation from news document

Languages

  • Nepali

Dataset Structure

Data Instances

The dataset consists over 200,000 Nepali news documents categorized into 20 different news categories.

Data Fields

  • category: News category
  • content: News document (main text)
  • headline: News headline
  • source: News source from where the news is taken from

Data Splits

The dataset is a whole dataset and is not splitted.

Dataset Creation

Curation Rationale

To develop and create a large-scale Nepali text classification dataset and releasing it to the public for further research and developments

Source Data

Initial Data Collection and Normalization

Data are scraped from popular Nepali news portals such as Onlinekhabar, Nepalkhabar, Ekantipur, Ratopati, Gorkhapatra, Nepalipatra, Educationpati, Crimenews, etc.

Who are the source language producers?

News portals

Annotations

Annotation process

Category labeling of news documents are automatically done as the documents are scraped from category-specific URLs of particular news source

Who are the annotators?

News portals

Personal and Sensitive Information

This dataset does not possess any personal and sensitive information. However, the news content can possess some biasness and irregular information which might be sensitive and not quite related with the original author of the dataset

Considerations for Using the Data

Social Impact of Dataset

No issues.

Discussion of Biases

Categories can be depended on how news portals have categorized them. Otherwise could cause some bias between them.

Other Known Limitations

News summary are not included

Additional Information

Dataset Curators

Me myself.

Licensing Information

Apache-2.0

Citation Information

To be updated later (Paper submission in process)

Contributions

Thanks to @Suyogyart for adding this dataset.

Downloads last month
38