Most humor detection models only focus on language and the text itself, but humor is often contextual and dependent on visual cues. This repository contains analysis and a report on a model for detecting humor using multisensory input - built for a world of comic strips and memes.
Original source for New Yorker Caption Contest data found at https://github.com/nextml/caption-contest-data?fbclid=IwAR2B3pfC6NX5iBDZ0p91KX5UcSjLTp5Iij02PDBYQhbsj5QYnK_4vpYbtkg
Original source for News Headline data found at https://www.kaggle.com/therohk/million-headlines
Original source for one-liner data found at https://github.com/CrowdTruth/Short-Text-Corpus-For-Humor-Detection
Contextual_Humor_Detection.ipynb is an executable Jupyter Notebook containing all models and results for the project