A deep learning system that predicts political bias in news articles by combining textual content with visual cues. This project demonstrated that integrating image features improves classification accuracy.
Project outline
This project combines over 20,000 headlines and accompanying images to classify the political bias of articles. The model architecture is multimodal, combining: a Convolutional Network (ConvNet) to extract features from images, and a Bag-of-bigrams to learn textual patterns in headlines.
The project builds on methodologies from François Chollet’s Deep Learning with Python, along with relevant academic literature on bias detection and multimodal learning.
Key features:
- Multi-input model combining unstructured data
- Focus on political bias detection
- Built using Keras 3 and TensorFlow 2.17
Media
Example predictions on the test data: showing the model’s predictions, the ground truth, and the textual and image features.
