content, connections, and context
DESCRIPTION
Content, Connections, and Context Daniel Tunkelang, LinkedIn Keynote at Workshop on Recommender Systems and the Social Web At 6th ACM International Conference on Recommender Systems (RecSys 2012) Recommender systems for the social web combine three kinds of signals to relate the subject and object of recommendations: content, connections, and context. Content comes first - we need to understand what we are recommending and to whom we are recommending it in order to decide whether the recommendation is relevant. Connections supply a social dimension, both as inputs to improve relevance and as social proof to explain the recommendations. Finally, context determines where and when a recommendation is appropriate. I'll talk about how we use these three kinds of signals in LinkedIn's recommender systems, as well as the challenges we see in delivering social recommendations and measuring their relevance.TRANSCRIPT
Recruiting Solutions Recruiting Solutions Recruiting Solutions
Content, Connections, and Context Daniel Tunkelang Principal Data Scientist at LinkedIn
Daniel
1
Recommendation Products at LinkedIn
2
Similar Profiles
Events You May Be Interested In
News
Network updates
Connections
More than 50%
3
Recommendations drive:
> 50% of connections > 50% of job applications > 50% of group joins
Inputs for Recommender Systems
4
Content Social Graph
…
Behavior
Page Views Actions
Queries
Take-Aways
5
Content is king.
Connections provide social dimension.
Context determines where and when a recommendation is appropriate.
What is the goal of recommendations?
O Growth
O Engagement
O Revenue
6
What is the goal of recommendations?
O Growth
O Engagement
O Revenue
7
Provide relevant content
and establish social connections
in appropriate context.
8
Users First!
Content is King
9
Goal-Seeking is about Content
10
Content (Ir)relevance
§ No right answer, but many wrong answers.
11
%no [Voorhees, 2004]
WTF! @k
http://bit.ly/wtfatk http://bit.ly/percentno
Example: Related Searches
12
Collaborative Filtering as Content Signal
§ Use temporal locality within sessions.
§ Find queries with clicks on similar results.
§ Look for query overlap.
§ Learn more at CIKM! [Reda et al, 2012]
13
Example: Jobs You Might Be Interested In
14
Content Signals Dominate Social Signals
15
Corpus Stats
Job
User Base
Filtered
title geo company
industry description functional area
…
Candidate
General expertise specialties education headline geo experience
Current Position title summary tenure length industry functional area …
Similarity (candidate expertise, job description)
0.56 Similarity
(candidate specialties, job description)
0.2 Transition probability
(candidate industry, job industry)
0.43
Title Similarity
0.8
Similarity (headline, title)
0.7 . . .
derived
Matching Binary Exact matches: geo, industry, … Soft transition probabilities, similarity, … Text
Transition probabilities Connectivity yrs of experience to reach title education needed for this title …
Provide relevant content
and establish social connections
in appropriate context.
16
Summary
Connections
17
Connections are Social Dimensions
18
Example: People You May Know
19
Relationships are Social
20
Shared Connections as a Signal
21
The Power of Social Proof
22
Beyond Triadic Closure
§ Triads suggest and affect relationships. [Simmel, 1908], [Granovetter, 1973]
§ Triangle closing is a Big Data problem. [Shah, 2011]
§ Use machine learning to rank candidates.
23
Provide relevant content
and establish social connections
in appropriate context.
24
Summary
Context
25
"to every thing there is a season” [Ecclesiastes 3:1]
One Platform, Many Users, Many Needs
26
http://blog.lab42.com/the-linkedin-profile
No Weekend Meetings!
27
navigational queries
exploratory queries
Weekdays
Weekends
Different Devices for Different Needs
28
Time and Place for a Career Change
29
Active vs. Passive Job Seekers
30
>10x more likely to reply if active
[Rodriguez et al, 2012]
Provide relevant content
and establish social connections
in appropriate context.
31
Summary
Recap
32
Content is king.
Connections provide social dimension.
Context determines where and when a recommendation is appropriate.
2 4 8
17
32
55
90
2004 2005 2006 2007 2008 2009 2010 2011 LinkedIn Members (Millions)
175M+
25th Most visit website worldwide (Comscore 6-12)
Company pages
>2M
62% non U.S.
2/sec
85% Fortune 500 Companies use LinkedIn to hire
Thank You!
33
We’re
Hiring!
Learn more at http://data.linkedin.com/