This blog post aims to provide commercial and academic understanding towards popular recommendation system. Commercially, the recommendation system contributes greatly to the number of views of webpages, it helps viewers to find contents that they’re interested in. Academicly, it is a good study case for understanding how contents in a large repository are discovered and the importance for content discovery tools, how information is stored and retrieved.
Summerized from paper I read, the recommendation system contributes the repository in the following ways:
1. Recommendation is one of the most imporant traffic sources of a large repository system. Take Youtube for example, there are several ways to access youtube. People can access through direct visit, Google Search, Mobile applications and Desktop applications. Research shows that over 50% of views are contributed through recommendation. For a large portion of the videos, they’re viewed many times before it is indexed to search engine.
Commercially, this adds the stickiness of websites therefore users stay on YouTube longer. It results in more advertisement apperance chances and higher advertisement click through times.
2. The more source is viewed, the more recommendation will be viewed. This means there’s a strong bond between source and recommendation. If the source is popular, the recommendation has a higher chance of becoming popular. This can be used for promotion and control of trend.
3. Recommendation helps user discover what they like rather than what is popular. Information systems without recommendation usually pop up what is believed to be popular or what they want to show to the users. They plant information to users rather than let them accept it actively. Recommendation induces users thru giving users options to choose, users could find what is provided useful for them therefore users are likely to revisit the system again.
Main paper:
The Impact of YouTube Recommendation System on Video Views
http://conferences.sigcomm.org/imc/2010/papers/p404.pdf
Flickr Tag Recommendation based on Collective Knowledge
http://www2008.org/papers/pdf/p327-sigurbjornssonA.pdf
Next blog post will introduce the algorithm for tagging system.
