How Useful Is A Tweet? iHub Research’s 3Vs of Crowdsourcing

Framework

Angela Crandall Nanjira Sambuli Chris Orwa

This research was funded by Canada’s International Development Research Centre.

Twitter: Some Facts and Figures

•  Launched in 2006 •  Approximately 550 million active users

worldwide •  About 200 million monthly active users •  An average of 400 million tweets are sent

everyday globally •  60% of the monthly active users log on using

a mobile device at least once every month

Twitter and the Tweeps

•  Twitter…can be more of a news media than even a social network (Kwak et al, 2010)

•  Breaking news and coverage of real-time events are all shared under the 140-character limit

•  Twitter users search for up-to-the-second information and updates on unfolding events

Twitter for Crowdsourcing. That Is…

Collecting information from the “crowd” •  Allows for a wide reach of people in inexpensive ways •  Large amounts of data can be obtained quickly, and often in real time •  Not necessarily through tech, but nowadays most use tech such as online or via mobile phone • Crowdsourcing fosters citizen engagement with the information—to dispute, confirm, or acknowledge its existence.

Mapping Kenyan Election Events, Thanks to crowdsourcing!

What is there to (Twitter) crowdsourcing?

Viability: In what situation/events is crowdsourcing a viable venture likely to offer worthwhile results/outcomes? Validity: Does crowd-sourced information offer a true reflection of the reality on the ground? Verification: Is there a way in which we can verify that the information provided through crowdsourcing is indeed valid? If so, can the verification process be automated?

Crowdsourcing during an Election •  What, if any, particular conditions should be in place

for crowdsourcing of information to be viable during an election period?

•  Can crowd-sourced information be validated during

an election period? If so, what is the practical implementation of doing so?

•  How do different crowdsourcing methods contribute

to the quality of information collected?

Why Elections? o  Elections in Kenya have been noted to spark many

online conversations, especially with the continued uptake of social media;

o  Citizens have an important role to play to contribute

information from the ground; o  Existing election crowdsourcing initiatives (such as

Uchaguzi), but none use passive crowdsourcing; o  Research exists around crowdsourcing during

disasters, but does not yet exist around elections.

Why Crowdsourcing, Kenyan Elections and #KoT

•  #KoT have participated in crowdsourcing activities severally, under hashtags such as #CarPoolKE, #findfuel, #SomeoneTellCNN etc.

•  Approximately 90,000 tweets generated during the first Kenyan Presidential Debates (as monitored using popular hashtags)

•  Election-campaigning was also digital

(Online) Passive Crowdsourcing vs. Active Crowdsourcing

• Active – Open call made for participation (e.g. Ushahidi’s Crowdmap).

• Passive – Sifting through content already being generated (e.g. on Twitter/Facebook) to capture relevant information.

What we did

Cross-comparison of different media sources: o  Traditional Media o  Data mining from Twitter o  Uchaguzi Crowdsourcing o  Fieldwork

Research Findings

Passive Crowdsourcing

is Viable During the

Elections in Kenya

Twitter Breaks

News

An Example from the Westgate Incident

First tweet about the attack at 12:38PM

First tweet by media about the attack

First tweet by a government institution about the attack

Mining Of Twitter Data without Machine Learning is Not Feasible

Search method

Time taken

Number of Newsworthy Tweets

Search time for whole data set

Viable for real time analysis

Viable for post-data analysis

Linear search

90 hrs 100 270 days No No

Keyword search

4.5 hrs 400 27 days No In a very limited way

ML, supervised learning

Less than 6 mins, 1.5 hrs labeling

12,208 Less than 1 sec

Yes Yes

From the Westgate Incident… Mining tweets from the Westgate attack manually has been labour-intensive, limiting us to sufficiently analysing the first half hour (12:38 PM – 1:18 PM GMT+ 3) Further analysis into Twitter data from the incident will require machine learning techniques.

In Summary: o  Kenyan social media content is rich with real-time

updates of happenings that might not be present in mainstream media reports.

o  Mining of crowd-sourced data appears to be high value

when one is looking for timely, local information. o  There are indeed considerations that are useful for

assessing and running an election-based crowdsourcing activity.

The 3Vs Crowdsourcing Framework

AVAILABLE FOR FREE DOWNLOAD HERE: http://www.ihub.co.ke/blog/2013/08/3vs-crowdsourcing-framework-for-elections-

launched/

Next Steps

•  Testing the 3V’s Framework on other election-related crowdsourcing opportunities

•  Move to real-time analysis of tweets •  Provide tools for verifying crowdsourced

information. •  Integrate research to media practices •  Working with local media organizations to build a

useable tool for collecting real-time newsworthy incidents from the crowd

research@ihub.co.ke

?

?

?