A better online music store is built on search and recommendation. It brings more money for rights holders, publishers and retailers, more visibility and promotion for artists, and more music and fun for consumers.
3. Outline
1. Why we need better online music stores
2. Power of The Long Tail
3. Role of recommender systems in e-commerce
4. Expert, social and content-based
recommendations
5. Demo
6. Additional thoughts and conclusion
8. Better... how?
● Rights Holders, Publishers and Retailers
– «Make more money»
● Artists
– «Visibility and promotion»
● Consumers
– «Broaden my horizon»
– «Something new that will impress my friends»
– «I'm in the mood for some soft rock ballads»
9. Long Tail Economics
● The cost of shelf space online is ZERO
● Therefore:
1. Make everything available
2. Help me find it
● «Recommender systems expose consumers to
a larger selection of interesting and relevant
music»
10. The Shape of The Long Tail
The Hits
Popularity
The Long Tail
Items
12. Source: «I Tube, You Tube, Everybody Tubes: Analyzing the World’s Largest User Generated Content Video System»,
Cha et.al., ACM Internet Measurement Conference 2007.
13. What is a Recommender
● System that connects (relevant) items to items,
items to users, and users to users
● Way to navigate large data collections
● Content relevance filter
● Important characteristics include:
– Transparency
– Familiarity vs. novelty
– Completeness
14. Orders of Information Management
1st order 2nd order 3rd order
Structuring Classification Tagging and
other
metadata
16. Expert Recommendations
● «I'm telling you that you will like this, because I
know a lot about music»
● Pros
– Transparency of the recommendations
– Can differentiate between “good and bad” music,
according to the expert
● Cons
– Not personalized
– Limited coverage
– No scaling
Source: Celma & Lamere, Music Recommendation Tutorial, ISMIR 2007
18. Social recommenders
● «You will like this, because it's popular with
people like you»
● Pros
– Works for and between everything
● Cons
– Lack of transparency
– Already popular items stay popular (the rich get
richer effect)
– Cold start, new items enter at the bottom
19. Artists Similar to U2
Source: Celma & Lamere, Music Recommendation Tutorial, ISMIR 2007
23. Social Tagging
Pros Cons
● Order emerges from chaos ● Polysemy, synonyms, spelling
(folksonomies) ● Idiosyncracity
● Layers of metadata ● Sparsity
24. Content-based Recommendations
● «You will like this, because it sounds like something
you already like»
● Objective musical similarity
– Timbre, instrumentation, rhythm, tempo, intensity
● Pros
– No popularity bias
– No cold-start
– No manual effort required
● Cons
– Not so transparent
– Can't tell «good» from «bad»
26. The Effect of CB Recommendations
Source: Celma & Lamere, Music Recommendation Tutorial, ISMIR 2007
27. User Ratings – Yes and No
● Very effective, but highly suggestive
– We trust other people
– We tend to like what others like
● Can counteract Long Tail effects
● This is viral marketing!
29. Conclusion
● A Better Online Music Store is built on search
and recommendation
● More money for Rights Holders, Publishers and
Retailers
● Visibility and promotion for Artists
● More music and fun for Consumers
30. Thank you!
● Check out «Widgets, Viral Marketing and
Findability» by Andrew Dubber
– 14:15 in this auditorium
Vegard Sandvold
+47 48 23 92 32
vsandvold@gmail.com