PolitScope: An Unsupervised Natural Language Processing Tool for Political Bias Detection

Smith, Titus

PolitScope: An Unsupervised Natural Language Processing Tool for Political Bias Detection

Persistent URL

https://hdl.handle.net/10456/59264

Author(s)

Smith, Titus

Date Issued

May 5, 2025

Abstract

PolitScope is a web-based browser extension designed to address the growing issue of political bias present in social media, a challenge that contributes to the increasing polarization in society today. As bias escalates on social media, it's more important than ever to have reliable methods for detecting and analyzing it. This study explores the use of natural language processing (NLP) to create an automated system for identifying political bias. The methodology involves cleaning and embedding text using the RoBERTa model, then clustering similar text to create a baseline model. This model is saved and used to compare and dynamically detect bias in newly collected web text. Through a series of experiments, we evaluate PolitScope's effectiveness, examining how various parts of speech influence bias detection. While the results show that PolitScope is not yet perfect, the tool demonstrates the potential of NLP in highlighting political bias in online social media platforms, and contributing to the broader conversation about bias in digital media.

Major

Computer Science

Honors

Computer and Information Science, 2025

First Reader(s)

Graber, Emily