19
“Make New Friends, but Keep the Old” - Recommending People on SN sites Jilin Chen, Werner Geyer, Casey Dugan, Michael Muller, Ido Guy CHI2009 June 1, 2011 Hyewon Lim

“Make New Friends, but Keep the Old” - Recommending People on SN sites Jilin Chen, Werner Geyer, Casey Dugan, Michael Muller, Ido Guy CHI2009 June 1, 2011

Embed Size (px)

Citation preview

“Make New Friends, but Keep the Old”- Recommending People on SN sites

Jilin Chen, Werner Geyer, Casey Dugan, Michael Muller, Ido GuyCHI2009

June 1, 2011Hyewon Lim

2

Outline Introduction Beehive People Recommendation Algorithms Experiment I: Personalized Survey Experiment II: Controlled Field Study Discussion and Conclusion

3

Introduction Users connect to both friends they already know off-

line &new friends they discover on the site– Many users of popular social networking sites primarily

communicate with people they already know offline– Users of enterprise social networking sites find valuable con-

tacts not yet known to them, or connecting to weak ties

Finding known contacts and interesting new friends to connect with on the site can both be a challenge

4

Introduction “People You May Know” on Facebook

– Based on a “friend-of-a-friend” approach

Recommending people on social networking sites– Different from traditional recommendations of books,

movies, etc., due to the social implication of friending

5

Introduction Social implications of “friending”

– Social dynamics can be obstacles in accepting recommenda-tions

Could be more prominent if unknown or barely known people are recommended

– Lack enough motivation

Despite the difficulty…– Connecting with weak ties or unknown but similar people can

be more valuable to users than merely re-finding existing strong ties_Granovetter73

6

Beehive An enterprise social networking site within IBM Launched in Sep. 2007 38,000 users with an average of 8.2 friends per user

(2008. 7) Friends are directional

– Could be a non-reciprocal friendship A user can request to be introduced to another user

through Beehive

7

People Recommendation Algorithms Algorithms

– Utilize social network structure and based on content simi-larity

Content Matching Content-plus-link Friend-of-a-friend SONAR

8

People Recommendation Algorithms

Algorithm 1: Content Matching

Find users associated with similar content on Beehive

Create a bag-of-words representation of each user– Word vector to describe u:

Vu = (vu(w1), …, vu(wm)) vu(wi) describes the strength of u’s interest in word wi,

calculated using a TF-IDF

Similarity of two users – measured by the cosine similarity of Va and Vb

9

People Recommendation Algorithms

Algorithm 2: Content-plus-Link

Motivation– Disclose a network path to a weak tie or unknown person

Introduce ‘valid social link’– Computes similarity in the same way as the content match-

ing algo– Boost the similarity by 50% if a valid social link

Favors people in close social network proximity to the user over people more disconnected from the user in the social network

10

People Recommendation Algorithms

Algorithm 3: Friend-of-a-Friend

Leverages only social network information of friend-ing– Requires existing friends

Recommendation candidate setRC(u) = { user c | ∃user a s.t. F(u, a) and F(a, c) }

Mutual friend setMF(u, c) = { user a | F(u, a) and F(a, c) }

– Size of MF(u, c) = score of each candidate c for u

11

People Recommendation Algorithms

Algorithm 4: SONAR

Based on the SONAR system– Aggregates social relationship information from different pub-

lic data sources within IBM organization chart, public DB, patent DB, friending system, peo-

ple tagging system, project wiki, blogging system

– For each data source SONAR computes a normalized rela-tionship score in the range of [0, 1]

SONAR returns a list of users related to u and their aggregated relationship score

12

Experiment I: Personalized Survey Methodology (with 500 active users)

– Overlap ratios between recommendations

– Survey

13

Experiment I: Personalized Survey Result 1: Understanding users’ need

– 95% of the users considered people recommendations to be useful

– Users are interested in connecting to weak ties and meeting new people

– Useful information for connecting users to an unknown peo-ple

no

maybe

yes

0 10 20 30 40 50 60 70

7.4

31

61.6

other

division within IBM

geographical location

common contents

common friends

0 10 20 30 40 50 60 70 80

14.5

27

39.2

74.4

75.2

14

Experiment I: Personalized Survey Result 2: Known vs. unknown, good vs. not good

– Result by algorithm

15

Experiment I: Personalized Survey Result 3: Immediate actions resulted from recom-

mendations– Good recommendations that resulted in actions

16

Experiment II: Controlled Field Study Methodology (with 3,000 random users)

– Divided the 3,000 users randomly into 5 groups 4 groups get recommendations using algorithms each 1 group was a control group that did not get any recommenda-

tions

– New recommender widget on users Beehive homepage

17

Experiment II: Controlled Field Study Result 1: Effectiveness of recommender algorithms

– Recommendations resulting in connect actions In contrast to the survey, users rarely chose the introduction op-

tion

– Users can click a link in widget to view the profile 8% of content matching recommendations, 2.9% for SONAR

18

Experiment II: Controlled Field Study Result 2: Impact of people recommendations

– Immediate goal of recommending people in social networking sites is to increase a user’s network of friends

– Compare the number of friends before and after the experi-ment

– People recommendations would impact user activity on the site Viewed 13.7% more pages vs. 24.4% less pages Increase in content and comment creation in the exp. Groups (too

low)

19

Discussion and Conclusion In the experiments

– Relationship-based algorithms outperform in terms of user response

Relationship-based algorithms – Better at finding known contacts– Perform particularly well for newer users

But, too weak to be meaningful for more established users

– FoF can expand their contact list from a few existing contacts– SONAR-like aggregation can take advantage of additional

data

Combine the strengths of both types of algorithms– Have an additional benefit of increasing new users’ trust in

the system