Catching and Forecasting Popular Videos on Youtube
Teammate WANTED!
Have you ever considered why some videos could attract millions of views within just a few days?
Recently, research community begin to study the characteristics of the popular videos hoping the discovery would benefit the marketing and advertisement. Tom
Team
Datasets
In Google's paper they adopt their internal datset which is unavailable for research study outside of Google. There exists a public dataset[1] consistsing of 1,761 videos metadata. However, several limitations renders it inappropriate for this study.
- Most videos are NOT popular videos(Almost no viral videos).
- Information is incomplete. For example the comments and the information about the uploaded user are missing.
- Clone videos, which refers to the copy of videos uploaded by different users, are manually labeled. Therefore the dataset may not be easily scaled.
A dataset of popular YouTube videos (initially 2000~3000 videos) will be collected in this project. For each video the following information will be included:
- Matedata
- Comments
- Video
- Public statistics
The clone videos will be identified by an automatically approach named Near Duplicate Videos Detection [2].
Baseline Method
- Modeling the view growth distribution based on BA model
- Popularity prediction by Naive SVM and features.
Advantages
Challenges
- First public dataset on YouTube Popular videos of comprehensive information.
- Video and Thumbnail content analysis (Semantic Object Detection)
- Growth model for viral videos
Related Work
[3]Tom Broxton and Yannet Interian and Jon Vaver and Mirjam Wattenhofer: Catching a viral video. Journal of Intelligent Information Systems 2011: 1-19.
[4]Youmna Borghol, Siddharth Mitra, Sebastien Ardon, Niklas Carlsson, Derek L. Eager, Anirban Mahanti: Characterizing and modelling popularity of user-generated videos. Perform. Eval. 68(11): 1037-1055 (2011)
[5]Y. Borghol, S. Ardon, N. Carlsson, D. Eager, and A. Mahanti, Proceedings of the 18th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD 2012), Beijing, China, August 2012, to appear.
[6]Gábor Szabó, Bernardo A. Huberman: Predicting the popularity of online content. Commun. ACM 53(8): 80-88 (2010)
