Paper | Code RANKVIDEO is a reasoning-based reranker for video retrieval that explicitly reasons over query-video pairs using video content to assess relevance. This repository contains the data files used for training and evaluating RANKVIDEO, primarily based on the MultiVENT 2.0 benchmark.

Dataset Structure

The repository includes the following key data components:

training_data.json: Training examples… See the full description on the dataset page: https://huggingface.co/datasets/hltcoe/RankVideo-Dataset.