advertisement

Coding Social Imagery: Learning from a #selfie #humor Image Set from Instagram

20 %
80 %
advertisement
Information about Coding Social Imagery: Learning from a #selfie #humor Image Set from...

Published on April 6, 2016

Author: ShalinHaiJew

Source: slideshare.net

advertisement

1. Coding Social Imagery: Learning from a #selfie #humor Image Set from Instagram DIGITAL POSTER SESSION BIG XII TEACHING AND LEARNING CONFERENCE (3RD ANNUAL) THEME: TRANSFORMATIONAL TEACHING AND LEARNING KANSAS STATE UNIVERSITY JUNE 2 – 3, 2016

2. 2

3. Definition: Social imagery Digital images shared through social media platforms 3 Social networking sites (Facebook) Work-based social networking sites (LinkedIn) Digital content sharing sites Image-sharing sites (Flickr) Video sharing sites (YouTube, Vimeo) Microblogging sites (Twitter) Web logs / blogs Wikis Crowd-sourced online encyclopedias (Wikipedia) Email systems SMS systems (short message systems) Work-based collaboration systems The Web and Internet (broadly speaking), and others

4. Social imagery SOME CHARACTERISTICS In the wild (shared on social media platforms) Social Plentiful (“big data”) Multilingual Opinion-ful Repurposed, mash-ups Low-res, medium-res, high-res Uncensored, low-censored, censored USE FOR RESEARCH Scrape-able / mass collectible High-dimension data Codable (by type, by concept, by content, and other aspects) May be harnessed for research…but how? 4

5. #selfie and #humor(Venn diagram) 5

6. 6

7. Overview Social media messaging has long been harnessed to inform faculty about their respective learners. The textual channel is often used because of the ease of interpretation and analysis. Social imagery—tagged images, #selfies, grouped imagery, and others—has been less used, in part because images are more complex and multi-meaninged to analyze. Also, there are not many generalist models that inform how to code or even understand social imagery in an emergent way. (There are large-scale computational means to interpret online images, such as the AlchemyAPI of IBM Watson, for various types of feature extractions. There are ways to code imagery based on specific research questions in particular fields-of-practice.) 7

8. Overview(cont.) The presenter recently analyzed a 941-image #selfie + #humor image set from Instagram, with three main research questions: (1) What does identity-based humor look like in terms of a #selfie #humor- tagged image set from the Instagram photo-sharing mobile app? (2) Do more modern forms of mediated social humor link to more traditional forms theoretically? Is it possible to apply the Humor Styles Model to the images from the #selfie #humor Instagram image set to better understand #selfie #humor? (3) What are some constructive and systematized ways to analyze social image sets manually (with some computational support)? This digital poster session will highlight some of the initial research findings (forthcoming in a near-future publication) and share insights about effectively coding social imagery in a bottom-up and emergent way. 8

9. Study of #selfie #humor Image Set from Instagram 9

10. Instagram (2016) Initially designed as mobile app for mobile-image sharing (and now also short video sharing) Conceptualized as an “instant” “telegram” (= Instagram) By the numbers: ◦ 400 million monthly active users ◦ More than 75 million daily active users ◦ 20% of Internet users use Instagram ◦ 77.6 million Instagram users in the U.S. ◦ 51% male; 49% female Instagram users (Craig Smith, April 5, 2016, DMR) A young demographic user base (55% in 18 – 29 age group, 28% in 30 – 49 age group) (Pew Research Center, Mobile Messaging and Social Media, 2015) Purchased by Facebook in 2012 for $1 billion 10

11. Research method Study of identity-based humor based on on-the-fly digital self-portraits (“selfies”) from imagery shared on Instagram tagged with #selfie #humor ◦ Review of the Literature ◦ Extraction of Targeted Imagery (based on folk tagging) ◦ Ingestion of Imagery into NVivo 11 Plus ◦ Light data cleaning ◦ Analysis and Interpretation ◦ Theme extraction (qualitative analysis) ◦ Image categorization (quantitative analysis) ◦ Comparison with humor theories, humor styles, and other prior research) ◦ Consideration for Future Research and Methodologies 11

12. 12

13. Some features of the #selfie #humor image set Imagery tagged with #selfie and #humor on Instagram ◦ “Selfie” defined generally as an on-the-fly digital self-portraits ◦ “Humor” defined as something created to induce amusement “Found” imagery vs. created imagery in user-generated contents Digital image editing with some text overlays, melded images, image filters, and higher end image editing (image masking, layering) Identifiable social messaging and statements about self and about others Some recurring themes and apparent broad emulation of others Some elicitations for topical image sharing, such as through hashtag campaigns and event-based calls 13

14. Some features of the #selfie #humor image set (cont.) Word play and visual gags, name calling, and public call-outs and shaming Visual synecdoche and visual metonymy; visual symbolism Cause-and-effect narratives (if this, then this), side-by-side image comparisons (before-after, analogies in imagery), and image sequences (implied chronologies) Images separated from original contexts (in some cases), such as from blog posts and Tweetstreams and websites ◦ Varying degrees of messaging coherence / effectiveness as stand-alone images 14

15. What do people find funny (based on research)? ABOUT HUMOR People are hard-wired to respond instinctually to the ludicrous and the incongruous; there has to be surprise (the unexpected) but not contravening of internalized social ethics (humor can be edgy, but if it is seen as truly wrong, people will respond with anger, not amusement) Appreciation for humor is linked to cognition and emotion…and personality ABOUT LAUGHTER A lot of laughter is not related to humor per se but may be used to relieve tension in a social context ◦ Laughter may be used to emphasize certain points made in speech ◦ Shared laughter raises people’s moods Laughter may be voiced or unvoiced ◦ There are a wide array of laugh sounds 15

16. What do people find funny (based on research)? (cont.) ABOUT HUMOR Areas of the brain have been linked to response to jokes, such as one area for word play and puns Certain types of humor may be preferred by certain individuals in certain demographics, such as a preference for pratfalls and fart jokes among younger males tragedy + time = comedy ABOUT LAUGHTER Laughter tends to be social (much more likely in a context where people are in the company of others) ◦ People who have lived for long periods in isolation still have a laugh response but it tends to be silent or unvoiced People who tell jokes tend to laugh more at their own jokes 16

17. What do people find funny (based on research)? (cont.) ABOUT HUMOR There are social power dynamics in the expression of humor in the workplace ◦ Supervisors are more likely to tell jokes than subordinates ◦ Humor can have salutary effects in work places (those that are not NSFW) Males tend to tell more jokes than females ABOUT LAUGHTER Subordinates are more likely to laugh at their supervisor’s jokes than vice versa (and more likely to laugh at their supervisor’s jokes than their peers’ jokes) 17

18. High-level topic-based image categories in the #selfie #humor image set 1. Truth-telling about the self 2. Un-selfie (counter-messaging) 3. Animals (as self and other) 4. Inspirational 5. Human sociality and social media (meta-perspective) 6. Human tensions (and issues) 7. Funny faces 8. Spectacle 18

19. High-level image categories in the #selfie #humor image set (cont.) 19 1. truth- telling about self 2. un-selfie (counter- messaging) 3. animals (as self and other) 4. in- spirational 5. human sociality and social media (meta- perspective ) 6. human tensions and issues 7. funny faces 8. spectacle 8 40 31 57 78 54 8 12

20. 20

21. Understanding the frequency counts Image set was scraped over two days using two different web add-ons to extract images that came up from <#selfie #humor Instagram> Google Images and then Microsoft’s Bing Images ◦ Range of dependencies (time, technological) Not an N-of-all (by any means) Not a random sampling of the target image set (so not generalizable or directly representational) ◦ Unclear what sampling bias was inherent in the data extraction ◦ = Convenience sample descriptions Unweighted count of the respective images 21

22. Three main types of organizing themes from the image set (a) the purposes of the selfies ◦ 1. truth-telling about the self, 4. inspirational, and 8. spectacle; (b) the messaging ◦ 2. un-selfie (counter-messaging), 5. human sociality and social media (meta), 6. human tensions (and issues), and (c) the types ◦ 3. animals (as self and others), 7. funny faces 22

23. Critique of this bottom-up coding of complex social image sets WEAKNESSES Non-alignment in using three different approaches and conceptualizations to organize images (purposes / intentional communication, messaging, and types) ◦ Why not purposes alone? Or messaging alone? Or types alone? Inelegance in terms of lack of mutual exclusivity in terms of image coding (social images that can fit multiple categories simultaneously) ◦ The practice of analyzing for salience but lacking in sufficient explanatory power STRENGTHS Mixed approaches necessary given complexity of social image sets ◦ is some degree of coherence Fairly parsimonious in its approach Somewhat transferable to other types of data sets: purposes, messaging, and types Tapping into human perceptual analytics (without necessarily a need for computational supports) 23

24. Category 1: Truth-telling about the self Owning one’s own reality Owning one’s own laziness Avoiding gullibility 24

25. Category 2: Un-selfie (counter- messaging) Calling out humble-brags Mistaking the virtual for the real Using others to see the self 25

26. Category 3: Animals (as self and other) A literal animal selfie (“Monkey Selfie”) Animal selfie humor Straight animal images 26

27. Category 4: Inspirational Going forth and conquering Human predicaments Not Monday! Text manifestos about life Time Expressions of gratitude 27

28. Category 5: Human sociality and social media (meta-perspective) Over-focusing on looks Relating around money and purchases Social exclusion and inclusion Lots of talk and word play Social media socializing ◦ But first, let me take a selfie! ◦ Sharing food…images ◦ Dismissing others Truth behind the screen 28

29. Category 6: Human tensions (and issues) Racial tensions Gender group tensions Critiques of and comments about celebrities No drunkenness please Contemporary social and political issues 29

30. Category 7: Funny faces Funny faces 30

31. Category 8: Spectacle Spectacle (acts of derring-do) 31

32. 32

33. Gender counts and number presentation Male-female gender parity in single selfie image counts (selfies showing one individual in the image) but slight higher count for male subjects A larger male-to-female gap for duo and group selfie images Majority of selfie images in set were singletons, followed by duo and group selfies (the latter two at about the same rates) Minority of selfie images had unclear gender (in all three categories) 33

34. 34

35. 35

36. Selfie types and counts Photos predominated over drawings / illustrations and mixed text and visualizations Text -only selfies tend to be drawings or illustrations mostly, not photos “Extended self” analyses of selfies showed uses of the following to represent the “self”: people, animals, figures (objects), and texts Animal selfies tended to combine images and text 36

37. 37

38. Humans by general age categories Adults: 526 Children: 32 Toddlers: 4 Babies: 6 Numbers may be indicative of ease-of-access to imagery of adults by adults Child images are usually from a mix of child stars (through screen grabs) and family images Likewise, toddler and baby images seem to be either in the public arena (likely copyrighted) or personal 38

39. 39

40. Three common imagery format types Photographs: 429 Drawings / illustrations: 91 Mixed images and text visuals: 63 40

41. Observed humor styles Predominant humor style may have an effect on one’s social-psychological health. R. Martin, P. Puhlik-Doris, G. Larsen, J. Gray, and K. Weir (Feb. 2003) shared a piece “Individual differences in uses of humor and their relation to psychological well-being: Development of the Humor Styles Questionnaire” in the Journal of Research in Personality about four types of humor styles: 1. Affiliative humor (social): used to charm and amuse others so as to benefit relationships 2. Aggressive humor (social): used to critique and ridicule others so as to put others down 3. Self-enhancing humor (self): used to relieve tensions and stress so as to aid in coping 4. Self-defeating humor (self): used to put oneself down to make others laugh (at a cost to one’s dignity) 41

42. Observed humor styles (cont.) Humor styles have implications for healthy / unhealthy self-concept and for constructive / non- constructive social interactions with others. Humor styles may be seen in what types of humor people engage in and prefer. Humor styles may be inferred from shared messaging in the #selfie #humor image set. 42

43. An expanded 2 x 2 table of the four dimensions of humor styles with linked #selfie #humor image themes Enhance self Enhance social relations Benign Self-enhancing (adaptive) Affiliational (adaptive) 1. Truth-telling about the self 2. Un-selfie (counter-messaging) 3. Animals (as self and other) 4. Inspirational 5. Human sociality and social media (meta-perspective) 7. Funny faces 8. Spectacle Injurious Aggressive (mal-adaptive) Self-defeating (mal-adaptive) 6. Human tensions (and issues) 6. Human tensions and issues 43

44. Some Early Insights re: Coding Social Imagery 44

45. Image data cleaning Keep a pristine master collection of the raw images before any data cleaning is done. ◦ Avoid losing data from data cleaning. Early data cleaning involves deciding what belongs in the research set and what doesn’t. ◦ Spell out standards for inclusion / exclusion. Be consistent. ◦ What is a #selfie? What is #humor? Remove duplicate images. ◦ Different messaging using the same underlying images were considered to be different #selfie #humor images. 45

46. Image sufficiency Knowing how many images to collect was not clear initially, or even later in the work. ◦ Saturation would suggest that images should be collected until there are not relevant new themes identifiable for a fair emergent representation. Amount of effort required to iteratively code the imagery manually was a deterrent against searching for more images. Check-backs at multiple periods thereafter (over months) using the same seeding terms (#selfie #humor Instagram) resulted in many images that were ◦ visually and thematically similar to those identified in this exploratory research and ◦ some recognizable images (from the initial image sets); ◦ novelty of concept, along with production quality, tends to be rare. 46

47. Image sufficiency (cont.) Originality is a rarity in this context (combined folk-labeled #hashtagged topics, the selected images from Instagram). ◦ Contents on Instagram are generally conceptualized to be “instant” “telegrams”—so the speed of creation is a factor. ◦ Social media account follower-ship may encourage emulation of others. ◦ Trending memes, which encourage copycats, apparently lead to repeated types of messaging. ◦ Practice of “photoshop battles” (such as on Reddit) appear to lead to particular types of visual expressions based on specific digital image editing / visual expressions. ◦ Instagram “filters” (puppy filter, flower crown filter, and others) result in some common image overlays. 47

48. Some benefits to emergent coding Emergent coding begins with the imagery set…and not anya priori coding theory, framework, model, research question, or other approach. ◦ This bottom-up approach starts with the minutiae of specific dimensions of the images in the image set. This type of coding is closer to the data (rather than starting from a top-down theory and seeing how the data fits). ◦ This does not assume that the data is somehow speaking for itself. Rather, this still acknowledges the subjectivity of coding achieved by a person or people. Setting a baseline description for an image set provides some useful insights…which may be built upon with more targeted types of coding (see reference above). ◦ “Baseline” is used as a limited descriptor of the particular limited image set. This is not understood in any way as generalizing out to the N=all image set. 48

49. Some benefits to emergent coding (cont.) Baseline-setting of social image sets include qualitative (descriptive) and quantitative (count) approaches: categorization of images based on … ◦ concepts (themes, messaging) ◦ contents (gender) ◦ types (single; duo; group) (photos, drawings, combined text and visualization) Identification of anomalies in image sets ◦ It is helpful to have a sense of general tendencies of the image set. ◦ It is helpful to identify outliers and to be able to describe why and how the selected images are outliers. 49

50. Social image data extraction Only 62% of extracted images fit a broad definition of #selfie #humor (the seeding hashtags) ◦ Using “extended self” conceptualizations ◦ “Self” is not only the individual but relationships, possessions, employment, and other aspects ◦ Including non-human animal and object-based and word representations of a self ◦ Plenty of images that were “selfies” but with no apparent “humor” (except maybe in the sense of “in good humor” as in a good mood (a smiling face but no funny or attempt at funny) Selfie multiple ways (in terms of counts): ◦ self as individual “I,” ◦ self as collective “we,” ◦ self as individual “other,” ◦ self as collective “other” Messaging on continuum of empathy and sympathy to non-empathetic and antagonistic A fair amount of noise (vs. signal) to folksonomic tagging of socially-shared images 50

51. Social image data extraction (cont.) Methods for image data capture limit amount of dataand metadata captured ◦ Some data lossiness such as non-capture of the original names of images (using Firefox browser add-on DownThemAll and Google Chrome add-on Chronos Download Manager) ◦ Google Images more effective than Bing Images in capturing larger sets with more original images, using the same data find parameters (#selfie #humor Instagram) Image data scraping using Python or R would likely be much more effectivein terms of amounts of images and additional data collection beyond the images Manual image downloads could enable the capture of more information (higher resolution images, more metadata), but that process is time-costly and slow and not particularly scalable (maybe except through crowd-sourcing) 51

52. Social image data extraction (cont.) May look for “semi-automatic” social image-capture approaches in the future (part machine scraping, part human downloads) 52

53. Some observations about social image coding Image pixelation can be a problem given the low resolution of the scraped images. ◦ Some research, such as through reverse image searches online (like TinEye), can be helpful to establish provenance. ◦ However, it is important to set research limits for interpretation of the images in the #selfie #humor set. Some research led to some iffy sites. Others dead-ended in misspellings. ◦ If an image is too muddy on its own, it was generally omitted from the research set. ◦ Obvious ads were also omitted. There is more ambiguity in imagery than one might initially assume. For example, coding for gender could be challenging. Coding for age can be challenging; for example, does the image show a child or a young adult? Coding for race was avoided because it is not verifiably possible to. Honest and thorough coding means handling some images that can be offensive and socially questionable. 53

54. Some observations about social image coding (cont.) An image may be coded multiple ways because the categories tend not to be mutually exclusive. ◦ An image can fit fully and / or partially in multiple coding categorizations. Non-English languages (or non-base languages) can be a problem, even with the help of Google Translate. (Original language use may be slang, and many textual elements contain misspellings.) It is helpful to iterate over image sets multiple times with different focuses each time in order to capture accurate information. Image coding is not going to be a once-through sort of activity. ◦ It helps to have a clear and focused purpose for each iteration. Guess-ti-mating numbers is not helpful, not accurate. For an accurate manual count, it helps to go through and count attentively…and even recount. 54

55. Coding social imagery in NVivo 11 Plus NVivo 11 Plus (a qualitative data analytics research suite) enables the curation of a large number of images for data analytics. To benefit more from NVivo 11 Plus, it helps to add plenty of insightful descriptive text in the notes fields…in order to have textual contents against which to run data queries and autocoding. ◦ Without text annotations, word frequency counts, word searches, and other types of data queries cannot be run against the image data. ◦ Without detailed text annotations, no sentiment and no theme extractions will be seen through the autocoding. It helps to also do other coding using other software, such as Word (for notetaking) and Excel (for quantitative representations). 55

56. Social image exploration to create a social image codebook An initial image set may be used to create analytical tools to apply to larger sets of images. ◦ For example, it may help to use an image sample set to create an initial codebook. A basic codebook is comprised of a code and then a description of the standards for what would be coded to that particular node (or code category). ◦ It may help to have some digital exemplars for those categories as well (to better explain the coding). 56

57. Copyright challenges It is hard to chase copyright on such user-shared images since there is so much cooptation of images and ideas from others. ◦ The ostensible users of the images were often not likely the owners of the images and so do not have standing to release copyright for an image. ◦ It’s easier to not publish any of the images without established image ownership and a legal release. ◦ Descriptions of #selfie #humor imagery with proper uses of quotation marks was preferable to actual inclusion of the images. The Instagram End User License Agreement (EULA) requires people who post to own copyright to the materials that they upload and to release rights to Instagram, but there is no blanket copyright release to users of the service. ◦ Users of Instagram indemnify Instagram as a service and platform only. 57

58. Other image coding methods in the academic literature In the research literature, there have been some early works on digital image coding: ◦ using algorithms to analyze social images for contents (understanding of one central form) and facial recognition, and sentiment (such as AlchemyAPI, now of IBM Watson) / algorithms trained on Web- scale data ◦ seems to enable broad-scale summary data ◦ seems to be applied to trending issues ◦ using crowd-sourcing (CrowdFlower) to have people label social and other images (with fairly high levels of confidence) ◦ seems to require a targeted question or research aim Both above methods involve commercial entities and additional costs. 58

59. Other image coding methods in the academic literature (cont.) There are domain-specific works as well, with coding for specific research and queries (a priori coding)…but nothing the author could find about emergent manual coding of social image sets. Some generic types of research questions based on image coding include the following: ◦ Are there gender biases in how people are depicted in mass media around particular social issues? ◦ Are there geographical tendencies in terms of visual depictions based around particular topics? ◦ What are differences between image sets tagged with the same term (keyword or hashtag) from different geographical regions? Different cultures? Different people groups? 59

60. Conclusion and contact Dr. Shalin Hai-Jew ◦ iTAC, Kansas State University ◦ 212 Hale / Farrell Library ◦ 785-532-5262 ◦ shalin@k-state.edu Thanks to Dr. Jana R. Fallin and her team at the K-State Teaching & Learning Center for accepting this digital poster session for the Big XII Teaching and Learning Conference (2016)! This digital poster session was created from a chapter that is forthcoming in a text scheduled for publication in 2017. All contents are copyrighted. 60

#selfie presentations

Add a comment

Related pages

Instructional Coaching Menu-New teachers who need a plan ...

FREE * READY FOR THE NEW SCHOOL YEAR * All teachers know how important it is to be in contact with parents on the regular. With this free contact log ...
Read more

wiki.cfcl.com

nilofer: We live in an era that screams for transformation, and the best we seem capable of is tweaks. http://itssaulconnected.com/archives/2015/06/where ...
Read more

Salud, Smartphone and Infographic on Pinterest

from Frikipandi - Blog de Tecnología - Lo más Friki de la red. What's more intriguing than an #infographic about social media and #health!? Aside from ...
Read more