What if technology could help with moderating online content?

At the Discourse Processing Lab, we have two focused projects on organizing and moderating online content: fake news and news comment moderation.

In these projects, we apply general methods of text classification ranging from classic machine learning algorithms (e.g., Naive Bayes, Decision Trees, SVMs) to deep learning architectures (e.g., feed-forward, recurrent and convolutional neural networks). Our work is informed by research in linguistics, media analysis and natural language processing.

Fake news

How can news readers identify bias and misinformation in news articles?

We have collected a large corpus of news articles from online publishers with different levels of veracity and built classification models based on linguistic cues of deception. The resulting system can be used to score the likelihood of an unseen article to be deceptive.

Comment moderation

We want to encourage constructive discussion online. This can be done, for example, by promoting the most informative reader comments.

As part of this project, we have collected and curated a corpus of about 10,000 opinion articles, together with their reader comments, from the national newspaper in Canada, The Globe and Mail.

We have annotated subsets of this corpus for constructiveness and toxicity. We have also developed computational methods to identify constructive comments.

Our methods can assist in moderation tasks, typically performed by humans, such as organizing and summarizing reader comments.