Bajaj Allianz Life Insurance Company - Insurer Innovation Award 2024
How to read computer vision-based networks?
1. HOW TO READ COMPUTER VISION-BASED NETWORKS?
Universidade Nova de Lisboa I iNOVA Media Lab
@jannajoceli ˚ thesocialplatforms.wordpress.com
Janna Joceli Omena
The University of Sheffield, 24 June 2020 I Online research seminar hosted by STeMiS and the Digital Society Network.
7. ❏
❏ The definitions & potentialities of computer vision API-based networks.
❏ What precedes the reading of these networks?
❏ What takes place with and in image-label & image-domain networks?
READ
12. Visual representation of “Portuguese” in stock image sites.
ShutterStock and Adobe Stock, January 2019.
13.
14. Pro-impeachment protests in Brazil.
Hashtag engagement I 18 March 2016, Instagram.
Image-label network of the pro-impeachment protests.
15. The imagery of #femboy
Hashtag engagement I June 2015 - August 2017, Instagram.
Image-Label Network of #femboy
16. Image-Label Network of Microcephaly on Instagram.
mother
daughter
cool
The visuality of #microcefalia in Brazil
Hashtag engagement I June 2012 - October 2017, Instagram.
18. the rookie the amateur the expert
* Shutterstock query for austrian
this is outdoor
Microsoft
the rookie
they’re trees
IBM watson
the amateur
they’re black pines
Google Vision
the expert
19.
20.
21.
22. Detecting the mood of Portuguese
Universities through Facebook
Timeline images.
38. query design & grammars
data capture affordances
output file and metadata
building the network
[query]
hashtag engagement
Facebook timeline images
Video/apps/profile img thumbnails
usernames (e.g. botted accounts)
keywords (e.g. climate emergency)
[research software
or python scripts]
API calling
Scraping
output file and metadata
GEPHITABLE2NET
EXCEL & PYTHON
SCRIPTS
GOOGLE VISION API
Full matching images
Images URLS Related data
43. Isolated elements
Periphery
Mid-term
Centre
Node size I Degree Node size I in-degreeComputer Vision API-based Networks
Total number of co-occurrences of
labels used to describe images
XImage-label
(undirected graph)
Total number of co-occurrences of link
domains and images
Image-domain
based web detection - full matching
images (mixed graph)
Node size I out-degree
X
When a link domain hosts one
or more images
When an image appears in
one or more link domains
Node colour
Created attributes, e.g.
clusters, year, image host
Created attributes, e.g.
username, year or TLD, ccTLD
44.
45. query design & grammars
data capture affordances
output file and metadata
building the network
281 visible
[purchase & list of hashtags]
460 invisible
[data analysis & co-tag nets]
[query]
botted accounts
[query]
botted accounts
241 visible
[purchase]
442 invisible
[note section & co-tag nets]
Scraper built by Jason Chao
[last 30 posts]
visible invisible total
4056 7579 11.635
visible invisible total
7082 12.093 19.175
GEPHITABLE2NETEXCEL
GOOGLE VISION API
Full matching images
Images URLS
analytical process
BIG SCREENPRINTED NETGEPHI WEB
analytical process
BIG SCREENPRINTED NETGEPHI WEB
48. Sistema de leitura de redes digitais multiplataforma
Digital Methods for Hashtag Engagement Research.
Call into the platform!
APIs de Visão Computacional: investigando
mediações algorítmicas a partir de estudo de bancos de imagens.
Cross-Platform Digital Networks:
Bots and the Black Market of Social Media Engagement
Interrogating Vision APIs
Reading Digital Networks: Climate Emergency, Bolsonaro & Bot Image Circulation by Vision API.