Symposium event illustration

Arjun Sharma presented at The Sixth Polygence Symposium of Rising Scholars. Interested in the next Symposium? Fill out the interest form here for more information.

Go to Polygence Scholars page
Arjun Sharma's cover illustration
Polygence Scholar2021
Arjun Sharma's profile

Arjun Sharma

The Shri Ram School MoulsariClass of 2024

About

I'm an 10th grade student in Delhi NCR in India. I have a strong interest in computer science and its application in the real world using emerging technologies. In my free time, I like to read science fiction and historical fiction novels, and write on my technology-focused blog.

Projects

  • Identifying Fake News Articles Using Deep Learning with mentor Maria (Oct. 26, 2021)

Project Portfolio

Identifying Fake News Articles Using Deep Learning

Abstract or project description

Fake news articles rapidly spread online and pose an array of serious consequences on society at large. Manually verifying the veracity of the millions of articles that exist online, however, is an impractical task. This paper therefore attempts to adopt a machine learning-based approach to identify fake news articles by attempting to identify common trends in the style of the articles. A publicly available dataset consisting of 20,000 articles flagged as reliable and unreliable articles, with 4202 unique authors, is obtained and used for the purposes of this paper. Three features are extracted for the titles and bodies for the articles in the dataset. These are topics, sentiment, and length. Models to identify fake articles are developed, where each network excludes one or more of the six potential features. Each network is optimised using the Keras Hypertuner to attain the maximum potential performance, and is evaluated in terms of a number of different metrics. This paper finds title and body topics, body length, and body sentiment to be effective features for this binary classification problem. Title sentiment and title length are found to be ineffective features. The most successful model developed, in terms of test accuracy, attained a test accuracy of 89.87%, in addition to a low test loss value of 0.2615.