1.
Social Network Analysis at
Linkedin
Mitul Tiwari
Search, Network, and Analytics (SNA)
LinkedIn
1

2.
Who am I?
2

3.
Outline
About Me
About LinkedIn
Analytics products at LinkedIn
Fun Facts: Sequencing the Startup DNA
3

4.
LinkedIn by the numbers
120,000,000+ users (August 4, 2011)
2+ new user registrations per second
81+ Million monthly unique users* (Comscore)
2 Billion People Searches in 2010
7.1 Billion page views in Q2 2010
2+ Million companies with LinkedIn Company Pages
2+ M LinkedIn Groups
4

5.
Broad Range of Products
5

6.
User Profile
6

7.
Connections
7

8.
Communications
Communications
10
8

9.
Hiring Solutions
9

10.
Job Search
10

11.
Company Pages
11

12.
Outline
About Me
About LinkedIn
Analytics products at LinkedIn
Fun Facts: Sequencing the Startup DNA
12

13.
What do I mean by Analytics Products?
13

14.
People You May Know
14

15.
Profile Stats: WVMP
15

16.
Viewers of this profile also ...
16

17.
Skills
17

18.
InMaps
18

19.
Related Searches
Millions of Searches everyday
Help users to explore and refine their queries
19

20.
Analytics Products: Key Ideas
Recommendations
People You May Know, Related Searches, Viewers of
this profile ...
Insight
Profile Stats: Who Viewed My Profile, Skills
Visualization
InMaps
20

21.
Analytics Products: Challenges
LinkedIn: largest professional network
120+ million members on LinkedIn
Billions of pageviews
Terabytes of data to process
21

22.
Outline
Analytics products at LinkedIn
Deep Dive - Connection Strength
Deep Dive - Related Searches
Fun Facts: Sequencing the Startup DNA
22

23.
Connection Strength
Let’s build “Connection Strength”
Systems and Tools we use
Hadoop MapReduce
Managing workflow
Serving data in production
23

24.
Connection Strength
How well do people know each other?
Connection Strength
Application: reorder updates to show updates
from close connections
24

25.
Connection Strength
How well do
Alice
people know each
other?
Bob Carol
25

26.
Connection Strength
How well do
Alice
people know each
other?
Bob Carol
26

27.
Connection Strength
How well do
Alice
people know each
other?
Bob Carol
Triangle closing
27

28.
Connection Strength
How well do
Alice
people know each
other?
Bob Carol
Triangle closing
Prob(Bob knows Carol) ~ the # of common connections
28

29.
Triangle Closing in Pig
-- connections in (source_id, dest_id) format in both directions
connections = LOAD `connections` USING PigStorage();
group_conn = GROUP connections BY source_id;
pairs = FOREACH group_conn GENERATE
generatePair(connections.dest_id) as (id1, id2);
common_conn = GROUP pairs BY (id1, id2);
common_conn = FOREACH common_conn GENERATE
flatten(group) as (source_id, dest_id),
COUNT(pairs) as common_connections;
STORE common_conn INTO `common_conn` USING PigStorage();
29

30.
Pig Overview
Load: load data, specify format
Store: store data, specify format
Foreach, Generate: Projections, similar to select
Group by: group by column(s)
Join, Filter, Limit, Order, ...
User Defined Functions (UDFs)
30

31.
Triangle Closing in Pig
-- connections in (source_id, dest_id) format in both directions
connections = LOAD `connections` USING PigStorage();
group_conn = GROUP connections BY source_id;
pairs = FOREACH group_conn GENERATE
generatePair(connections.dest_id) as (id1, id2);
common_conn = GROUP pairs BY (id1, id2);
common_conn = FOREACH common_conn GENERATE
flatten(group) as (source_id, dest_id),
COUNT(pairs) as common_connections;
STORE common_conn INTO `common_conn` USING PigStorage();
31

32.
Triangle Closing in Pig
-- connections in (source_id, dest_id) format in both directions
connections = LOAD `connections` USING PigStorage();
group_conn = GROUP connections BY source_id;
pairs = FOREACH group_conn GENERATE
generatePair(connections.dest_id) as (id1, id2);
common_conn = GROUP pairs BY (id1, id2);
common_conn = FOREACH common_conn GENERATE
flatten(group) as (source_id, dest_id),
COUNT(pairs) as common_connections;
STORE common_conn INTO `common_conn` USING PigStorage();
32

33.
Triangle Closing in Pig
-- connections in (source_id, dest_id) format in both directions
connections = LOAD `connections` USING PigStorage();
group_conn = GROUP connections BY source_id;
pairs = FOREACH group_conn GENERATE
generatePair(connections.dest_id) as (id1, id2);
common_conn = GROUP pairs BY (id1, id2);
common_conn = FOREACH common_conn GENERATE
flatten(group) as (source_id, dest_id),
COUNT(pairs) as common_connections;
STORE common_conn INTO `common_conn` USING PigStorage();
33

34.
Triangle Closing in Pig
-- connections in (source_id, dest_id) format in both directions
connections = LOAD `connections` USING PigStorage();
group_conn = GROUP connections BY source_id;
pairs = FOREACH group_conn GENERATE
generatePair(connections.dest_id) as (id1, id2);
common_conn = GROUP pairs BY (id1, id2);
common_conn = FOREACH common_conn GENERATE
flatten(group) as (source_id, dest_id),
COUNT(pairs) as common_connections;
STORE common_conn INTO `common_conn` USING PigStorage();
34

35.
Triangle Closing in Pig
-- connections in (source_id, dest_id) format in both directions
connections = LOAD `connections` USING PigStorage();
group_conn = GROUP connections BY source_id;
pairs = FOREACH group_conn GENERATE
generatePair(connections.dest_id) as (id1, id2);
common_conn = GROUP pairs BY (id1, id2);
common_conn = FOREACH common_conn GENERATE
flatten(group) as (source_id, dest_id),
COUNT(pairs) as common_connections;
STORE common_conn INTO `common_conn` USING PigStorage();
35

36.
Triangle Closing Example
Alice
Bob Carol
connections = LOAD `connections` USING
1.(A,B),(B,A),(A,C),(C,A) PigStorage();
2.(A,{B,C}),(B,{A}),(C,{A})
3.(A,{B,C}),(A,{C,B})
4.(B,C,1), (C,B,1)
36

37.
Triangle Closing Example
Alice
Bob Carol
1.(A,B),(B,A),(A,C),(C,A)
group_conn = GROUP connections BY
2.(A,{B,C}),(B,{A}),(C,{A}) source_id;
3.(A,{B,C}),(A,{C,B})
4.(B,C,1), (C,B,1)
37

38.
Triangle Closing Example
Alice
Bob Carol
1.(A,B),(B,A),(A,C),(C,A)
2.(A,{B,C}),(B,{A}),(C,{A})
pairs = FOREACH group_conn GENERATE
3.(A,{B,C}),(A,{C,B}) generatePair(connections.dest_id) as (id1, id2);
4.(B,C,1), (C,B,1)
38

39.
Triangle Closing Example
Alice
Bob Carol
1.(A,B),(B,A),(A,C),(C,A)
2.(A,{B,C}),(B,{A}),(C,{A}) common_conn = GROUP pairs BY (id1, id2);
common_conn = FOREACH common_conn
3.(A,{B,C}),(A,{C,B}) GENERATE flatten(group) as (source_id, dest_id),
4.(B,C,1), (C,B,1) COUNT(pairs) as common_connections;
39

40.
Connection Strength
Age
Company Overlap
School Overlap
40

41.
Related Searches
Let’s build “Related Searches”
Systems and Tools we use
Hadoop MapReduce
Managing workflow
Serving data in production
41

42.
Systems and Tools
Kafka (LinkedIn)
Hadoop (Apache)
Azkaban (LinkedIn)
Voldemort (LinkedIn)
42

43.
Data Cycle
43

44.
Systems and Tools
Kafka
publish-subscribe messaging system
transfer data from production to HDFS
Hadoop
Azkaban
Voldemort
44

45.
Systems and Tools
Kafka
Hadoop
Java MapReduce and Pig
process data
Azkaban
Voldemort
45

46.
Systems and Tools
Kafka
Hadoop
Azkaban
Hadoop workflow management tool
to manage hundreds of Hadoop jobs
Voldemort
46

47.
Systems and Tools
Kafka
Hadoop
Azkaban
Voldemort
Key-value store
store output of Hadoop jobs and serve in production
47

48.
Related Searches
Let’s build “Related Searches”
Systems and Tools we use
Hadoop MapReduce
Managing workflow
Serving data in production
48

49.
Hadoop MapReduce
Large-scale distributed data processing
Map phase: partition the problem in smaller
steps
Reduce phase: aggregate the output of smaller
steps
Allows parallelism and fault-tolerance
49

50.
Related Searches
Let’s build “Related Searches”
Systems and Tools we use
Hadoop MapReduce
Managing workflow
Serving data in production
50

51.
Our Workflow
Triangle
Age School Overlap
Closing
Union
push-to-prod
51

52.
Our Workflow
Triangle
Age School Overlap
Closing
Union
push-to-qa push-to-prod
52

53.
Sample Workflow
53

54.
Azkaban Workflow Management
Dependency management
Regular Scheduling
Monitoring
Diverse jobs: Java, Pig, Clojure
Configuration/Parameters
Resource control/locking
Restart/Stop/Retry
Visualization
History
Logs
54

55.
Our Workflow
Triangle
Age School Overlap
Closing
Union
push-to-prod
55

56.
Our Workflow
Triangle
Age School Overlap
Closing
Union
push-to-prod
56

57.
Related Searches
Let’s build “Related Searches”
Systems and Tools we use
Hadoop MapReduce
Managing workflow
Serving data in production
57

58.
Voldemort
Large amount of data/Scalable
Quick lookup/low latency
Versioning and Rollback
Fault tolerance through replication
Read only
Offline index building
58

59.
Voldemort RO Store
59

60.
Outline
Analytics products at LinkedIn
Deep Dive - Connection Strength
Deep Dive - Related Searches
Fun Facts: Sequencing the Startup DNA
60

61.
Related Searches
Millions of Searches everyday
Help users to explore and refine their queries
61

62.
How to build Related Searches?
Collaborative Filtering: searches done in the
same session
Q1 Q2 Q3 Q4
Time
62

63.
How to build Related Searches?
Searches correlated by results clicks
Q1 R1
Qn Rm
63

64.
How to build Related Searches?
Searches with overlapping terms
Q1 Jeff LinkedIn
Q2 LinkedIn CEO
64

65.
Outline
Analytics products at LinkedIn
Deep Dive - Connection Strength
Deep Dive - Related Searches
Fun Facts: Sequencing the Startup DNA
65

66.
Sequencing the Startup DNA
Done by Monica Rogati at
LinkedIn
66

67.
Sequencing the Startup DNA
•Top majors: Entrepreneurship, Computer Engineering
•Bottom: Nursing, Administration
67

68.
Sequencing the Startup DNA
•Most founders age at first startup between 20 and 40
68

69.
Sequencing the Startup DNA
•Top regions: San Francisco, NYC
69

70.
Sequencing the Startup DNA
•Top Business Schools:
Stanford, Harvard, MIT Sloan
70

71.
Things Covered
Analytics products at LinkedIn
Deep Dive - Connection Strength
Deep Dive - Related Searches
Fun Facts: Sequencing the Startup DNA
71

72.
SNA Team
Thanks to SNA Team at LinkedIn
http://sna-projects.com
We are hiring!
72

73.
Questions?
73