Predicting Tie Strength with the Facebook API - PowerPoint Presentation

Presentation on theme: "Predicting Tie Strength with the Facebook API"— Presentation transcript

Slide1

Predicting Tie Strength with the Facebook API

Tasos

SpiliotopoulosMadeira-ITI, University of Madeira, Portugal / Harokopio University, GreeceDiogo PereiraUniversity of Madeira, Portugal Ian OakleyUlsan National Institute of Science and Technology, Republic of Korea

18th

Panhellenic

Conference on Informatics (PCI 2014

), 2-4 October 2014, Athens, Greece

1Slide2

“a (probably) linear combination of the amount of time, the emotional intensity, the intimacy (mutual confiding), and the reciprocal services which characterize a tie

” Mark Granovetter (1973) in The Strength of Weak TiesStrong ties.Weak ties.Tie strength2Slide3

Gilbert & Karahalios: a browser script that crawled Facebook web pages Panovich

et al: Facebook’s “Download Your Data” featureBurke & Kraut: Facebook server logsOthers: Publicly available datasetsAsynchronous calculationNon-standard tools and technologies Tie strength calculation and Facebook3Slide4

90 participants1728 friendships rated18 variables collected via the Facebook API

Study description4Slide5

Study description

5Slide6

Study description6Slide7

Study description7Slide8

Study description8Slide9

Study description9Slide10

Study description10Slide11

Study description11Slide12

Results12

90 participants (59% male)1728 Facebook friendshipsMean age: 26.9 years (SD = 8.7) From 11 countries (85.6% from Portugal)Mean number of Facebook friends: 355 (SD = 218.9, range = 28 – 872)Using Facebook for an average of 13.4 (SD = 15.1) hours per weekSlide13

Results – data collected13

18 predictive variables based on:

privacy preservationprevious literatureSlide14

Results – regression model of tie strength

14Slide15

Results – tie strength distributions

15

The model underestimates tie strength (mean: 0.29 vs 0.13, median: 0.21 vs 0.1), but that’s common.19.7% of friendships rated by the participants were set to zero.Slide16

Results – classification16

65.9% accuracy in differentiating between strong and weak ties, χ2 (1, N = 3456) = 135.08, p < 0.001 86.3% accuracy in differentiating between very strong and weaker ties, χ2 (1, N = 3456) = 107.83, p < 0.001 Slide17

Assessing tie strength calculation in real timeEnables automated friend characterization -> friend grouping, customized feeds, adaptive privacy controls, friend recommendations, content recommendations, more efficient information seeking

Enables more sophisticated social network analysisContributions17Slide18

Contributions18Slide19

Contributions19Slide20

Assessing tie strength calculation in real timeEnables automated friend characterization -> friend grouping, customized feeds, adaptive privacy controls, friend recommendations, content recommendations, more efficient information seeking

Enables more sophisticated social network analysisBetter understanding of tie strengthA model of tie strengthWeights of the predictor variablesInsights for computational social science studiesContributions20Slide21

Assessing tie strength calculation in real timeEnables automated friend characterization -> friend grouping, customized feeds, adaptive privacy controls, friend recommendations, content recommendations, more efficient information seeking

Enables more sophisticated social network analysisBetter understanding of tie strengthA model of tie strengthWeights of the predictor variablesInsights for computational social science studiesContributions21Thank you!