IEEE Consumer Electronics Magazine - July 2016 - 92

These data indicate that people
were making decisions based on
tweet content.
system that is portable across multiple social networks. In
our experimental setup, we use Twitter as the social media
platform. Twitter was selected due to its follower and following structure, as this can be informative of the Twitter
user's personality traits.
We foresee use cases for citizen actuation in environments
from small scale, like a neighborhood/community, or a small to
middle enterprise but also to medium- and large-scale entities
such as a city. We envisage citizen actuation as forming part of
the design process of future smart devices and environments as
a method to keep users engaged with their surroundings. Our
research survey discussed in the next section gathered information regarding how people would use their own experiences
and background knowledge on Twitter to pick suitable people
that they would feel would complete a task.

Survey
Our survey ran July-October 2014 and was shared through
online social networks. In total, 136 people entered the survey,
and 92 people completed the survey. Only fully completed
responses were taken into consideration for analysis. The
respondents were 69.6% male, and 30.4% female. 82.6% of
the respondents replied that they had a Twitter account, and
59.4% of these stated that they posted to Twitter at least once
or twice a week; similarly to Java et al. [15], we define these
respondents as "active users." Our survey aims to measure the
participant's opinion on whether the owner of a user profile
would complete a small task. We defined small task as being a

50
45
40
35
30
25
20
15
10
5
0

Very Unlikely

Unlikely

Neither Likely or Unlikely
FIGure 1. Question 1 results.
92 IEEE ConsumEr ElECtronICs magazInE

JULY 2016

Likely
Very Likely

short (timewise) action taken to effect or report on the person's
environment (for example, opening or closing a window, turning off electrical appliances, or taking a picture on their smartphone and posting it to Twitter). In particular, the questions
looked at requests (to complete tasks) sent to users through a
microblogging platform and their likelihood to complete these
tasks based on profiles and their features.
Twitter's API allows for the programmatic exploration of
user profiles and users' posts. After initial experimentation
with Twitter's API, it was discovered that we could not rely
on just using Twitter's API to get users, as users were
required that varied in a wide range of activity levels, and this
method would generally return users that are more active.
Followerwonk, a Twitter analytics tool, was used to select
users from the chosen location, as this allows selecting/sorting Twitter users through multiple measures. Screenshots
were taken of the chosen profiles and then edited to show the
desired content. The use of screen shots allowed each participant to view the profile in the same state, with the same
tweets and features. The participants only saw the eight
tweets from each user profile. The main questions asked in
relation to the chosen profiles in the survey are shown below.
The users viewed the profiles shown in a random order to
minimize question order bias.
▼ Q1 In your opinion how likely, would this Twitter account
holder be to complete a task?
▼ Q2 Rate the importance of different features of the profile
in helping you form that opinion of the profile:
a) number of tweets
b) number of followers
c) number of people following
d) description text
e) other.
Figure 1 shows the survey results from Question 1, which
relate to the participants' opinions on how likely each profile
is to complete a task. Question 2e was an open question asking whether there were any other elements of the profile that
influenced the participants' answers to Question 1. These
answers mainly related to tweet content and opinions people
formed around the tweet content and will be discussed further
in the "Computed Results" section. As mentioned previously,
59.4% of our respondents stated that they posted to Twitter at
least once or twice a week, and we defined these as active
users. In our analysis of our survey data, we also compared
our active user data separately to all our data and to our less
active users and found no significant difference in their
responses overall. This could be attributed to the survey
design and questions that highlighted for participants where
all the major features of a profile were.

Computed reSultS
In parallel with our survey, Twitter's API was used to collect
information from the ten profiles to examine any links
between the survey results and profile features. The main features extracted were the number of followers, the number of
people the profile follows (following count), a calculated

Table of Contents for the Digital Edition of IEEE Consumer Electronics Magazine - July 2016