YouTube is so familiar to our life, it is a platform for users to watch and share videos. The service, headquartered in San Bruno, California, was created by three former PayPal employees—Chad Hurley, Steve Chen, and Jawed Karim—in February 2005.( Wikipedia)
Founded in 2005, YouTube started its service by “embedding videos onto other websites”(Newton, 2017). In early days, the system transformed from simply “asking users to subscribe to channels” to adopting the number of clicks per video as the main factor in implementing its machine-learning-based recommendation system in 2011(Newton, 2017) . However, the number of clicks for each video was far from satisfactory; the development team soon realized they should find a better factor to boost the performance of the recommendation system, as users started to get video recommendations that are irrelevant to their browsing history (Business Insider).
The YouTube recommendation system as we see today was invented on March 15, 2012, by YouTube engineer Christos Goodrow and his team. Ever since Google bought YouTube in 2006, Google changed the YouTube recommendation system with its Artificial Intelligence technology, deep learning technology, which is being used in the fourth version of the YouTube recommendation system, to recommend users personalized and high-quality videos. The first version of recommendation system is based on the analysis of entire user-video graph to generate recommendations (Baluja, 2008), the second version of algorithm is based on user’s activity on the site, to recommend videos that similar to the previously viewed one to users (Davidson, 2010), the third version is also based on user activity to generate similar topic videos to users, but with improved algorithm for recommendation system (Bendersky, 2014). And the fourth version, which is the current version we have, added new technology of neural network, to further solve the problem of large-scale data, improve efficiency, and eliminate noise. (Covington, 2016)
Since the first video being uploaded to the website in 2005, after 13 years now the number of users visits the website has reached 800 million times per month, and the total watching time on YouTube is 4 billion hours per month. How could YouTube attract so many people to watch videos on their website? Team vector is here to deblack-box what’s so magical in their recommendation system!
Baluja, S. et al. (2008). Video suggestion and discovery for youtube. Proceeding of the 17th International Conference on World Wide Web – WWW 08. doi:10.1145/1367497.1367618
https://dl.acm.org/citation.cfm?id=1367618
Bendersky, M., Garcia-Pueyo, L., Harmsen, J., Josifovski, V., & Lepikhin, D. (2014). Up next: Retrieval method for Large Scale Related Video Suggestion. Proceedings of the 20th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining – KDD 14. doi:10.1145/2623330.2623344
http://static.googleusercontent.com/media/research.google.com/zh-CN//pubs/archive/42623.pdf
Covington, P., Adams, J., & Sargin, E. (2016). Deep Neural Networks for YouTube Recommendations. Proceedings of the 10th ACM Conference on Recommender Systems – RecSys 16. doi:10.1145/2959100.2959190
https://static.googleusercontent.com/media/research.google.com/zh-CN//pubs/archive/45530.pdf
Davidson et al. (2010) The YouTube Video Recommendation System. Proceedings of the Fourth ACM Conference on Recommender Systems, doi:10.1145/1864708.1864770
D’Onfro, Jilian. (2015, July) “The terrifying moment in 2012 when YouTube changed its entire philosophy.” Business Insider, http://www.businessinsider.com/youtube-watch- time-vs-views-2015- 7 . Accessed 26 January 2018.
Newton, C. (2017, August 30). How YouTube perfected the feed. The Verge. Retrieved from https://www.theverge.com/2017/8/30/16222850/youtube-google-brain-algorithm-video-recommendation-personalized-feed Accessed 25 January
YouTube. (2018, May 08). Retrieved from https://en.wikipedia.org/wiki/YouTube
