This repository contains the code for our state-of-the-art sarcasm explanation model, TURBO, as proposed in our paper titled "Target-Augmented Shared Fusion-based Multimodal Sarcasm Explanation Generation".
This model is trained on the MORE+ dataset which is an extension of the MORE dataset proposed by Desai et al. (2022). All of the data files for MORE+ are available in the Dataset/ directory except for the image files, which can be downloaded from the original Github repo for the MORE dataset. Put all of the images into a directory named images/ inside the Dataset/ directory of this repository.
TURBO can be trained by running the following in a terminal:
python train.py