Polarity Pulse: Modelling Social Influence from Positive and Negative Opinions in Online Networks

Abstract

The research presents a Tkinter-based application designed for the classification of tweets as fake or 
authentic using natural language processing (NLP) and polarity score-based supervised learning. The 
system facilitates the upload of a tweet dataset in CSV format, preprocesses the text by removing 
punctuation, stopwords, and non-alphabetic characters, and employs TextBlob to compute polarity 
scores for sentiment analysis. Tweets are classified as "AUTHENTICATED" (polarity > 0.51) or 
"FAKE" (polarity ≤ 0.5), with results displayed in a user-friendly graphical interface and visualized 
through a pie chart using matplotlib. The application leverages pandas for data handling and NLTK for 
text preprocessing, demonstrating a streamlined workflow from data input to visualization. While 
effective as a prototype for detecting potentially misleading content, the system’s reliance on a fixed 
polarity threshold and limited error handling suggests opportunities for enhancement. Future 
improvements include integrating advanced machine learning models, supporting multilingual datasets, 
and enabling real-time data analysis, making the application a promising tool for combating 
misinformation on social media platforms.