MIT
2019-20 · AI/ML Researcher
Context
I was an AI/ML researcher at MIT's Poggio Lab within the Center for Brains, Minds, and Machines. My research focused on using DNNs to detect language patterns in real and fake news from articles published directly before, during, and after the 2016 United States Presidential Election.
As part of Professor Poggio’s research team, we sought to answer two questions: (1) Can a convolutional neural network detect fake news based on the language of an article? and (2) How is the convolutional neural network classifying the language patterns it detects in real and fake news articles? To eliminate bias towards subject-specific content, the machine’s access to meta-data (i.e., source and author) was limited.
Our results showed that the emergent DNNs' representations capture subtle but consistent differences in the language of fake and real news: signatures of exaggeration and other forms of rhetoric. Our best model had an accuracy of 88%. As a result, we developed an online fake news detector, allowing users to see the algorithm’s detection of real or fake language in an article.
Contributions
I trained and tested several deep neural networks with five different network iterations, increasing the machine learning accuracy from 81% to 88%.
Our dataset consisted of 244 different fake news websites, while our real news dataset was an augmentation of articles from The New York Times and The Guardian. Both datasets contained a significant amount of patterns and advertisements embedded within the text of the articles, which would cause overfitting, so I automated the task of cleaning our datasets, which included 24,000 articles of both real and fake news.
After cleaning the datasets, I created an additional algorithm which separated our neural network’s findings by parts of speech (i.e., noun, verb, adjective, or adverb). In addition, I restructured the research code for open source availability on GitHub.
Through this work, I co-authored a research paper that earned publication at the NeurIPS 2018 Conference.
Awarded at NeurIPS 2018
On behalf of our team, I presented our research at this conference, and our research received the ‘Best Poster Presentation Award' at the “AI for Social Good” workshop.
MIT News Research Interview
Interview with Professor Xaiver Boix and myself by MIT News.
Relevant Links
Credits
Professor Poggio, Professor Boix, and Nicole.