YT Shorts Project

This repository contains the full codebase for our full paper submission on YouTube Shorts from four international news outlets.

Repository Structure

• transcriptions/ Whisper transcription wrapper + environment checks

• parsing/ Biaffine dependency parsing (SuPar)

• absa/ ABSA model training + inference + fine-Tuning configurations (BERT, RoBERTa, DeBERTa, Qwen-QLoRA)

• CV/ Scene-type taxonomy + computer-vision implementation

• data/ Publicly available dataset on YouTube Shorts & Snippets for GroundTruth (APC) + VLM-Model output

• docs/ Compute resources, dataset documententation, known artifacts & biases

• synthetic_data/ Metadata, parses, ABSA predictions