A Result Paper On CrowdPulse: From Trends to Insights

Abstract

The rapid advancement of Natural Language Pro- cessing (NLP) and transformer-based deep learning has created new opportunities to analyze and compare information nar- ratives across different media platforms. This paper presents CrowdPulse, a cross-platform intelligence system designed to compare news media coverage with public conversations on Reddit. The system collects real-time data from both sources and applies advanced NLP techniques including RoBERTa- based sentiment classification, achieving a validation accuracy of 94.02%, FASTopic topic modeling with an optimal coher- ence score of 0.485 at 15 topics, and Sentence-Transformer- based semantic similarity computation. The system further integrates an AI-powered daily digest using the Gemini 2.0 Flash model, narrative drift alerts, subreddit-level sentiment breakdown, and narrative framing tag generation. Results are presented through an interactive React-based web dashboard. The importance of understanding differences between media framing and public opinion is a major takeaway of this project.