NewsVideoDataset

Dataset Overview

Dataset Name

• Name: NewsVideoDataset

Dataset Statistics

• Total Videos: 2,883
• Total Duration: 151,474 seconds
• Number of Labels: 13,431
• Number of Sentences: 3,302
• Number of Unique Tokens: 9,179
• Average Duration per Video: 52.5 seconds
• Average Number of Labels per Video: 4.7
• Average Number of Sentences per Video: 1.2
• Average Number of Unique Tokens per Video: 3.2

Dataset Acquisition Method

• Video Download: Use the youtube-dl program to batch download videos and their metadata via the urls.txt file.
• Metadata Processing: Use the python pack_data.py script to clean and pack the metadata.

Citation Information

• Citation Format:

@inproceedings{whitehead2018KaVD, Author = {Whitehead, Spencer and Ji, Heng and Bansal, Mohit and Chang, Shih-Fu and Voss, Clare R.}, title={Incorporating Background Knowledge into Video Description Generation}, booktitle={Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing (EMNLP)}, year={2018}, month={November}, publisher={Association for Computational Linguistics}, location={Brussels, Belgium} }