WeFEND-AAAI20
Weak Supervision for Fake News Detection via Reinforcement Learning
Yaqing Wang,
Weifeng Yang,
Fenglong Ma,
Jin Xu, Bin Zhong, Qiang Deng,
Jing Gao
SUNY Buffalo & WeChat. AAAI, 2020.
Dataset
[2020-07-26] We collected more data and make it public.
└── data/
└── train/
└── test/
└── unlabeled data/
Data Statistics
# of data | # of fake news | |
---|---|---|
train | 10,587 |
2,743 |
test | 10,141 |
1,482 |
unlabeled news | 67,748 |
- |
Data Description
Columns | Description |
---|---|
Official Account Name | The name of official account, news publisher |
Title | News Title |
News Url | The url of news |
Image Url | The url of cover image |
Report Content | The reports from reader, split by ## |
label | label of news, 0 is real and 1 is fake |
Main Idea
Challenge:
Due to the dynamic nature of news, annotated samples may become outdated quickly and cannot represent the news articles on newly emerged events. Therefore, how to obtain fresh and high-quality labeled samples is the major challenge in employing deep learning models for fake news detection.
Solution:
We propose a reinforced weakly supervised fake news detection framework, i.e., WeFEND, which can leverage users’ reports as weak supervision source to enlarge the amount of training data for fake news detection.
Experiment
We aim to answer two important questions:
- Does the distribution of news change with time?
- why should we use the reports to annotate the fake news?