Overview of our visualization system: The panels (A–C) display companies and temporal journal reports—(A) Company Graph, (B) Journal Bias Timeline, (C) Journal Bias Breakdown.
Abstract
Identifying unreliable sources is crucial for preventing misinformation and making informed decisions. CatchNet, the Oceanus Knowledge Graph, contains biased perspectives that threaten its credibility. We use Large Language Models (LLMs) and interactive visualization systems to identify these biases. By analyzing police reports and using GPT-3.5 to extract information from articles, we establish the ground truth for our analysis. Our visual analytics system detects anomalies, revealing unreliable news sources such as The News Buoy and biased analysts such as Harvey Janus and Junior Shurdlu
BibTeX
@inproceedings{2024-FishBiasLens,
title = {Integrating Large Language Models and Visual Analytics for Bias Detection},
author = {Mauro Diaz AND Felipe Moreno-Vera AND Juanpablo Heredia AND Fabrício Dalvi Venturim AND Jorge Poco},
booktitle = {IEEE Visual Analytics Science \& Technology (VAST) Challenge},
year = {2024},
url = {http://www.visualdslab.com/papers/FishBiasLens},
}