computational social media lecture 04: shooting

Transcription

1 computational social media lecture 04: shooting daniel gatica-perez

2 this lecture 1. a snapshot of the present flickr, instagram, snapchat 2. a look into the past 20th century image sharing practices 3. understanding research on social image systems flickr: tags, communities, geo-tagged media instagram: selfies & engagement snapchat: ephemeral social media

3 1. a snapshot of the present

4

5 reminder: participation in social media 68% 29% 73% 22% 35% 25% 24% 27%

6 flickr 1. help people make their photos available to the people who matter to them 2. enable new ways of organizing photos and video launched bought by Yahoo hosting over 6 billion images 87 million users 3.5 million images per day bought by SmugMug

7 instagram capture and share the world's moment launched bought by facebook monthly active users 300M 400M users outside US 70% 75% shared photos 30B 40B uploaded photos per day 70M 80M daily stories n/a n/a likes per day 2.5B 3.5B M n/a n/a n/a 300 n/a accessed: accessed

8 2013 M. Duggan and A. Smith, Social Media Update 2013, Pew Research Internet Project,

9

10 snapchat We are not the sum of everything we have said or done or experienced or published we are the result. We are who we are today, right now launched 80% of users are in US 70% users are women 400 million snaps received per day declined offers from facebook & google 100 million monthly active users en.wikipedia.org/wiki/snapchat

11 2013: first Pew survey on Snapchat use: 9% of phone users; more popular among youth 2018: Pew survey: up to 27%

12 2. a look into the past

13 Everything exists to end up in a book Stéphane Mallarmé, credit: wikipedia Everything exists to end up in a photograph Susan Sontag, On Photography, 1977 credit: Picador

14 why photos? Van House (2007): memory & identity maintaining relationships self-representation self-expression credit (cc): Prince flickr

15 kodak moments (before digital): few people take few pictures share photos with family & close friends as single pics or albums kodak moments (after digital): more people take more pictures but share photos in the same way as before credit (cc): Prince flickr (cc):

16 Ofoto (1999), bought by Kodak 2001 Kodak EasyShare Gallery (2005) 2008: 60 million users, billions of images 2012: Kodak goes bankrupt Shutterfly buys it and shuts it down

17 participatory culture and photo communities (Henry Jenkins) a culture in which fans and other consumers are invited to actively participate in the creation and circulation of new content photo communities have existed for as long as photos and cameras have been available fan clubs, sport fans, birdwatchers, amateur photographers credit (cc): Bradley flickr web 2.0 empowered communities H. Jenkins, Convergence Culture: Wher Old and New Media Collide, New York University Press, 2006.

18 stock photography Corel Gallery (1st ed. 1994) PhotoDisc (2000), now gettyimages

19 3. understanding research on social image systems

20 tags, communities, geo-tagged media selfies & followers ephemeral social media

21 flickr

22 tags, communities, geo-tags

23 tagging on flickr

24 tags in flickr

25 a basic tagging model C. Marlow, M. Naaman, D. Boyd, & M. Davis, HT06, tagging paper, taxonomy, Flickr, academic article, to read, in Proc. ACM HT 2006

26 why do people tag? M. Ames and M. Naaman, Why we tag: motivations for annotation in mobile and online media, in Proc. ACM CHI, 2007

27 flickr communities

28 flickr groups

29

30 flickr groups: aggregates of images, tags, users self-organized users, common interests: photographic techniques specific content geographical locations social events just social interaction by sharing photos with groups, they create pools of photos and tags that characterize a group R. Negoescu and D. Gatica-Perez, Analyzing Flickr Groups, in Proc. ACM CIVR 2008

31 analyzing flickr groups what are photo-sharing practices of groups? participation: how much content do people share with groups? group loyalty: do people belong to many groups? photo recycling : do people shop around with the same photos? how to discover groups in flickr? topic-model group representation probabilistic, unsupervised learning 50k groups, 6.9M photos, 1.9M unique tags

32 sharing photos with groups users who share in groups do it often: % of photos shared in groups: mean 29% median 17%

33 probabilistic topic models: generative models of text + documents: bags of words + topics: soft clusters of words + documents are generated by sampling words from the topics + unsupervised learning in graphical models M. Steyvers and T. Griffiths, Probabilistic topic models, in Handbook of Latent Semantic Analysis, Erlbaum, 2007

34 topic models: Latent Dirichlet Allocation (LDA) Assume D documents, N words per document, and K latent topics Topics are 1:K,where each k is a distribution over the vocabulary For each document D Sample topic proportion d ~ Dirichlet( ) For each of the N words wd,n Sample topic assignment zd,n ~ Multinomial( d) Sample word wd,n ~ p(wd,n 1:K, zd,n), a multinomial conditioned on the topic specified by zd,n (out of all possible topics) D. M. Blei, A. Ng, M. Jordan, Latent Dirichlet allocation, Journal of Machine Learning Research, 3: , Jan D. M. Blei. Probabilistic topic models. Communications of the ACM, 55, 4, 77-84, Apr. 2012

35 topic modeling for Flickr groups p(t z) = p(w z) + groups are documents (bags of tags) + groups described by the topics they are about + inference using Markov Chain Monte Carlo (MCMC) + big-data versions available (PLDA, Chang et al. 2009) p(z G)

36 learned topics: top words and top groups (using PLSA, precursor to LDA)

37 visualizing the top groups for topic 1

38 learned topics (2)

39 learned topics (2)

40 learned topics (3)

41 learned topics (4)

42 learned topics (5)

43 how many topics are flickr groups about? topic-expert groups have spiky topic distributions, with one dominant topic ~35% of groups with 3-5 topics: focused interests (content-oriented groups) less semantic coherence in large social-oriented groups (20+ topics)

44 topic decomposition for group Portrait

45 topic decomposition for group Flickr Central

46 from communities to the commons

47

48 B. Thomee, D. A. Shamma, G. Friedland, B. Elizalde, K. Ni, D. Poland, D. Borth, and L.-J. Li. YFCC100M: the new data in multimedia research. Communications of the ACM 59, 2, 64-73, January 2016

49 flickr and stock photography

50 flickr monetizing CC images?

51

52 comparing image sharing practices

53 two image sharing practices kodak culture:: take photos to share with a small existing social group & to archive snaprs: : take photos with the primary objective of sharing them with the world credit (cc): john flickr: A. Miller & W. Edwards, Give and Take: a Study of Consumer Photo-sharing Culture and Practice, in Proc. ACM CHI, 2007.

54 two image sharing practices two datasets flickr: images with tags kodak gallery: images with free text 10,000 most popular words (in terms of number of users) for each source credit (cc): Klearchos R. Negoescu, A. Loui, & D. Gatica-Perez, Kodak Moments and Flickr Diamonds: How Users Shape Large-Scale Media, in Proc. ACM Multimedia, 2010

55 comparing kodak & flickr language Distribution of 10-category tag taxonomy (top 200 tags) + flickr but not kodak macro, selfportrait, blackandwhite, photoshop, flickr, abigfave, geotagged + kodak but not flickr enjoying, showing, giving, checking, loved, visiting, dressed, wearing

56 using topic models to study topic specificity for kodak & flickr kodak-specific users for each source + learn LDA on joint vocabulary + determine relevant topics for each user based on topic distribution + topic specificity flickr topic: abigfave, flickrdiamond,anawesomeshot flickr-specific flickr topic: nature, landscape, flora, ilovenature, plant, animal kodak topic: i, time, daddy, ready, mommy, love, big, happy, playing kodak topic: picture, pictures, edited, png

57 geo-tagged images (reading #5)

58 reading #5: need presenter & discussant by tonight To be presented next week (11.05) D. Crandall, L. Backstrom, D. Huttenlocher, and J. Kleinberg Mapping the World's Photos Proc. WWW Reminder: Everybody needs to PRESENT or DISCUSS one paper to get the grade for this part of the course. 5 papers left for the course (including this one)

59 questions? @dgaticaperez