---
language:
- en
license: apache-2.0
task_categories:
- text-classification
dataset_info:
  features:
  - name: text
    dtype: string
  - name: label
    dtype: int64
  - name: bert_embeddings
    sequence: float32
  - name: roberta_embeddings
    sequence: float32
  splits:
  - name: train
    num_bytes: 45378307
    num_examples: 4487
  - name: test
    num_bytes: 5147293
    num_examples: 499
  download_size: 49990113
  dataset_size: 50525600
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
  - split: test
    path: data/test-*
tags:
- fake-news
---

Dataset Source: [Fake News Detection Challenge KDD 2020](https://www.kaggle.com/competitions/fakenewskdd2020/data)

This is a copied and reformatted version of the Fake News Detection Challenge KDD 2020.

We use the raw `train.csv` from the official Kaggle Dataset and split the data into train and test sets.

- text: text of the article (str)
- embeddings: BERT embeddings (768, )
- label: (int)
  - 1: fake
  - 0: true

Datasets Distribution:
- Train: 4487
- Test: 499