adamicCNgraphworkshop

Views:
 
Category: Entertainment
     
 

Presentation Description

No description available.

Comments

Presentation Transcript

Slide1: 

A social network caught in the Web Lada Adamic and Eytan Adar (HP Labs, Palo Alto, CA) Orkut Buyukkokten (Google)

Slide2: 

Outline Intro to Club Nexus Profiles Nexus Net Similarity and distance Association by similarity Nexus Karma Conclusions

Slide7: 

Profiles: status (UG or G) year major or department residence gender Personality (choose 3 exactly): you funny, kind, weird, … friendship honesty/trust, common interests, commitment, … romance - “ - freetime socializing, getting outside, reading, … support unconditional accepters, comic-relief givers, eternal optimists Interests (choose as many as apply) books mystery & thriller, science fiction, romance, … movies western, biography, horror, … music folk, jazz, techno, … social activities ballroom dancing, barbecuing, bar-hopping, … land sports soccer, tennis, golf, … water sports sailing, kayaking, swimming, … other sports ski diving, weightlifting, billiards, …

Slide8: 

Finding correlations between user attributes Are people who consider themselves funny also more likely to enjoy comedies? 518 funny users 74 % of users overall like comedies 416 (80% of) funny users like comedies, this is 3.4 standard deviations (=10) above expected (383) Z score = 3.4 Z scores with absolute value > 2 are significant at the p = 0.05 level. 3.4 is significant at the 0.0003 level small differences (10%) can be significant.

Slide9: 

Personality and tastes (just a few examples)

Slide10: 

Major and personality

Slide11: 

Gender Differences

Slide12: 

Degree Distribution for Nexus Net 2469 users, average degree 8.2

Slide13: 

Shortest paths between users

Slide14: 

Clustering and betweenness Clustering or transitivity: how many of the user’s friends are friends themselves C = # links between friends (# friends)* (# friends - 1)/2 c = 0.17 for Club Nexus Other findings: people who list more buddies list more preferences/activities edges with high betweenness lie between dissimilar people (r = -0.2) people with high betweenness have more links (r = 0.7) - “ - have lower clustering coefficients (r = -0.12)

Slide15: 

Similarity and distance year is more important for undergrads department is more important for grads 1 2 3 4 5 6 7 8 0 0.2 0.4 0.6 0.8 1 distance between users in hops fraction of similar users G residence UG residence G department UG major G year UG year G status UG status

Slide16: 

users who like A all users Association ratios p = (# users who like A)/(total #users) L = # connections A users have m = expected number of links to other A users = L*p r = (# links between A users)/m

Slide17: 

Personality and association ratio

Slide18: 

Interests and association ratios

Slide19: 

Nexus Karma Rank how ‘trusty’, ‘nice’, ‘cool’, and ‘sexy’ your buddies are on a scale of 1 to 4 446 users ranked 1735 different friends correlations between scores given (users were ranked as ‘3,3,3,3’ more often than ‘1,4,2,3’ average scores: nice (3.37), trusty (3.22), cool (3.13), sexy(2.83) trusty--nice and cool--sexy more highly correlated (r = 0.7) vs. trusty--sexy and nice--sexy (r = 0.4) no relationship between average score received and # of friends negative correlation between average score given and # of friends

Slide20: 

How users view themselves vs. how others view them

Slide21: 

Additional insights from Nexus Karma Users receiving higher ‘nice’ scores give higher ‘trusty’, ‘nice’, and ‘cool’ scores (r = 0.14-0.17) If one user gives another user a higher ‘trusty’ or ‘nice’ score than their other friends, that same friend is more likely to reciprocate. Users who share friends are more likely to give each other high scores (r = 0.10-0.13)

Slide22: 

Conclusions Learn about real world social networks from online community Less effort than traditional social network survey methods, almost a side-effect of digital nature of interactions Although most results not surprising, data is very rich - opportunity to simulate search and information spread Karma data can be used to study online reputation mechanisms Longitudinal data can be used to study network evolution

Slide23: 

To find out more: Information dynamics group (IDL) at HP Labs: http://www.hpl.hp.com/shl/ Paper at: http://www.hpl.hp.com/shl/social/

Slide24: 

Free time activity and association ratios

authorStream Live Help