Baychi – Beyond Search: Social and Personal Ways of Finding Information

Here are my notes from the session:

Beyond Search: Social and Personal Ways of Finding Information

April 11, 2006 – PARC, Palo Alto – Neil Hunt, Netflix; David Porter, Live365; Tom Conrad, Pandora; Kevin Rose, Digg; Joshua Schachter, del.icio.us; Rashmi Sinha, Moderator

Rashmi gave a really nice intro, using del.icio.us tags! (see links below, I will be sure to steal this sometime soon). She also presented a 3-way theoretical framework for discussing the "beyond search space"

1. Input from individual (implicit, explicit, level of expertise, what is the motivation?, has to be really simple)
2. How to keep them motivated?- Input across people (trust, transparency)
3. Present info to users

Speakers plus Rashmi's del.icio.us intros

• Neil Hunt – Netflix
• David Porter – live365
• Tom Conrad – Pandora
• Joshua Schacter – del.icio.us
• Kevin Rose – digg

Neil Hunt – NetFlix- http://www.netflix.com

– A Brit! Yay.

– Online/postal movie rental
– 60000 titles, all but p0rn
– 50 iterations and tests on the ratings stars
– Reached 1 billion ratings a month ago
– 100 movies a year arrive with large marketing budget, 4-500 movies a month have no budget
– Challenge: how do you hook in a brand new movie? Need 1000 users at least
– Challenge: balancing nicheness versus popularity (stepping down the long tail)
– Focus on the Long Tail (Chris Anderson), demand creation (Blockbuster is demand fulfillment)
– 3 types of connections
– 1. "Official" Recommendations on your homepage
– 2. decision support: ratings from other users
– 3. friends: u2u – watch the same movies which your friends watched
– spam: not an issue, as people have to pay 20/month! also filter on curve (flag if too many ratings)

2. David Porter – Live365 – http://live365.com

– largest independent internet radio network – Question: are people using Pandora now? (-;
– 5th after MS, Yahoo, AOL, ClearChannel
– web1.0 and proud of it!
– anyone can use it to create their own radio station based on the MP3s they have downloaded
– Charge broadcasters to broadcast (with ads/without ads)
– Sell audio ads, placed within stream
– 2 types of recommendation
– 1. DJs (10000 of them)
– 2. recommendations based on similar presets
– Find stations via playlist search, browse genre, favorites
– last.fm: When asked about last.fm, admits that a user (-;

3. Tom Conrad – Pandora (Music Genome Project) – http://pandora.com
–
– Music Genome Project -> 40 musicians analyze fundamental nature of music ("unfundable approach!")
– Uses objective classification method – musicological basis, breathiness of vociaal, timbre, rhythm
– 4 distinct genomes: pop/country (200 attributes), hiphop/electronica (400 attributes), jazz (400 attributes), latin
– Early focus was recommender system -> licensed by AOL, Warner
– Then moved onto mass market – focus on low overhead, people who don't want to spend sunny saturday afternoons
– last.fm: Supporting a kid who is doing pandora/last.fm mashup
– problem with last.fm: all roads lead to radiohead

4. Joshua Schacter – del.icio.us – http://del.icio.us
– how do you spell delicious? y a h o o
– a way to remember bookmarks
– previous project: single-user, 2001
– 15000 readers for just my daily bookmarks
– todo: check out <bundled tags>
– browses by URL constructs
– http://del.icio.us/popular/x
– http://del.icio.us/tags/x

5. Kevin Rose – digg – http://digg.com
– stories submitted by the masses
– but unlike slashdot, rated by masses
– digg = individual vote for a story, and also bookmarked in your profile -> rss feed
– can see friends who dug story, my friends' activities (recommendations)
– http://digg.com/diggall/cloud
– http://digg.com/spy – see digg's live submissions
– Can bury diggs if you don't agree
– The dream: add recommendations -> find me other people into debian linux and oolong tea

===

Discussion
Input from individual
– GIGO? Garbage in, garbage out (-> Wisdom of Crowds, guessing number of beans in a jar)
– Interesting vs. True
– Danger of false stories, follow-up refutations of bad stories on digg (e.g. Oracle buying JBoss)
– implicit, explicit
– level of expertise
– what is the motivation?
– has to be really easy, make as implicit as possible
– Approaches
pandora – focus on expert
netflix – seeded by expert
digg/del.icio.us – masses
– producer/consumer split

Input across people
– aspects of trust
– transparency: where ratings come from?
– different facets: rating, rater, trust ratings, rating the raters
– easy in ebay: trust is transaction-based, aggregate
– del.icio.us not convinced that trust is important

Exploitation
1. People design movies to match expectations
2. DoJ subpoena …

Most services seem to issue feeds via RSS

Present info to users
– Too many services? We'll all get confused, but kids master it all

Business Models
– digg: serve enough pages you can survive off ads, same with myspace
– netflix: massive benefits of database, cf. huge broadvast tv ad (very inefficient)
– pandora: advertising, subscriptions, purchasing tx, hard to make money in music
– live365: cost penny/hour to play in terms of royalty per user, $10 CPM

Advertisements