• Goals: Recent and Fresh, Diverse and Relevant
• 2 Phase
  • Candidate generations
    • Co-visitation RecommendationCo-visitation Recommendation
      
      Customer who viewed this also viewed
      
      
      
      
      
      
      
      
      
      
      
      
      Related work
      
      [[The Youtube video recommendation system 2010]]
      
      There are other interesting work which is Fake Co-visitation Injection Attacks to... (association rule mining)
  • Ranking
    • Variety of signals for relevance and diversity
• Evaluations
  • CTR (Click Through RateClick Through Rate
    Click-Through-Rate (CTR) is the number of clicks that the item receives divided by the number of times the item is shown.
    
    clicks / impressions = CTR
    
    
    If we have 5 clicks and 100 impressions, our ...)
  • CVR — long CTR (only counting clicks that led to watches of a substantial fraction of the video) Adoption and Conversion RatesAdoption and Conversion Rates
    Adoption and Conversion Rates often require a domain- and application-specific definition.
    
    Supports
    
    [[The Youtube video recommendation system 2010]]
    
    Long CTR - only counting clicks t...
  • Session length
  • Time until first long watch
  • Recommendation coverage (the fraction of logged in users with recommendations)

Results

They run an experiment as follows.

Period: 21 days

Metrics: CTR (Click Through RateClick Through Rate
Click-Through-Rate (CTR) is the number of clicks that the item receives divided by the number of times the item is shown.

clicks / impressions = CTR


If we have 5 clicks and 100 impressions, our ...)

Variants:

1. Most viewed - Most popular videos within 1 day
2. Top Favorited - Most favorited videos
3. Top Rated - Most rated (liked) videos
4. Recommended - Co-visitation along with users behaviors

Results: CTR increased by 207% compared to most viewed.

Reference

James Davidson et al., “The YouTube video recommendation system,” in Proceedings of the fourth ACM conference on Recommender systems - RecSys ’10, Barcelona, Spain, 2010, p. 293. doi: 10.1145/1864708.1864770.

#recommender-system/videos #recommender-system