Menu
Home
Log in / Register
 
Home arrow Computer Science arrow Social Informatics
< Prev   CONTENTS   Next >

3 Engagement Analysis

In this section we present our engagement analysis study. For the purpose of this study we have collected the latest 3,200 posts from the @dorsetpolice Twitter account, published between 2011-12-23 and 2014-06-12. This account has around 14K followers, and over 3.3K tweets in the form of announcements, appeals, crime reports, etc. From the collected 3,200 posts (note that this limit is established by the Twitter API), 733 are not originally written by @dorsetpolice, but are messages retweeted from other sources. Also 74 of the collected posts are not initialisations but replies to other tweets.

To analyse these data we use a two-phase approach. In the first part we apply a machine learning analysis method [1] to identify the key linguistic and time characteristics of those posts attracting attention. In the second part we conduct a semantic analysis to extract the key topics (concepts and entities) of the policing messages. We combine machine learning with semantic technologies to better understand, not only how and when messages should be written to attract attention, but also which topics users are more likely to engage with.

3.1 Expressing Engagement in Twitter

In the Twitter platform, retweeting, favouring and replying are actions that require an explicit interaction from a user towards another one. These actions have been repeatedly considered in the literature of social media as engagement indicators [8, 9, 10, 11, 12]. In total, the posts generated by the @dorsetpolice Twitter account received 30,726 retweets. To provide an overview, the following table shows the top 10 retweeted posts in our collected dataset.

Table 1. Top 10 retweeted posts. Note that mentions and links have been anonymized.

Post

Date

Ret

Regarding tweets to @user1 We are aware of the issue and we are actively looking into it.

2012-07-30

22:47:19

6672

Regarding tweets to @user1 17-year-old man arrested this morning

2012-07-31

5069

at a guest house in the Weymouth area. Enquiries continue.

07:51:26

RT @user2: URGENT ALERT (please RT) Mass ransomware

2013-11-18

1434

spamming event targeting UK computer users. More... URL1

09:39:44

RT @user3: Today is #WorldMentalHealthDay RT if you agree: We

2013-10-10

853

need support and respect. We won't give up. URL2

09:54:56

RT @user4: Please RT: Stay away from the shoreline this evening/tomorrow. Coastal paths could be dangerous. Risk of being swept out to …

2014-01-02

13:48:04

392

RT @user5: Have you seen missing person Richard Brockbank from

2014-05-21

235

Newbury? URL3 @user6 #findbrocky URL4

16:46:58

RT @user7: Severe weather warnings have been issued for the next

2014-02-04

217

five days. More info at URL6, URL7

16:32:23

Wanted Poole man Dean Goodwin has been arrested by armed police

2012-11-27

177

in Poole and is in police custody

18:05:15

Someone must recognise suspect from #Bournemouth robbery. Call

2014-05-07

159

101 if you do. Please RT. #CCTV URL8

22:51:59

RT @missingpeople: Zara went missing from Wimbourne, Dorset last

2013-06-05

136

month. Please #jointhesearch RT and help us find her URL9

16:21:09

As we can see in Table 1, the top two posts talk about the detention of a criminal. The remaining posts focus on a variety of issues, such as sea and weather warnings as well as the tasks of searching for lost people or suspects.

Note that when users retweet they spread the message to their followers (as opposed to favouring or replying) leading to a potential stronger involvement and engagement. In this work we consider retweets as indicator of engagement for the rest of our analysis. Tweets that have been retweeted at least once by the citizens are considered seed-posts. Those tweets that have not been retweeted (i.e., have not obtained any direct engagement from the citizens) are considered non-seed posts. Table 2 summarises the dataset, and shows the number of seeds vs. non seed posts. As we can see from the table, over the course of nearly 3 years, from 2011-12-23 till 2014-06-12, 86% of the tweets received at least one retweet (seed posts).

Table 2. Dataset description (number of seeds vs. non seed posts)

Dataset

Time Spam

Num posts

Seed posts

Non seed posts

Twitter

2011-12-23

2014-06-12

3,200

2,770

430

The next two sections present the analysis of engagement dynamics performed over this dataset. The first part consists on a ML analysis, which aims to detect the linguistic and time patterns of seed vs. non-seed posts. The second part performs a semantic analysis to identify the key topics of seeds vs. non-seed posts.

 
Found a mistake? Please highlight the word and press Shift + Enter  
< Prev   CONTENTS   Next >
 
Subjects
Accounting
Business & Finance
Communication
Computer Science
Economics
Education
Engineering
Environment
Geography
Health
History
Language & Literature
Law
Management
Marketing
Philosophy
Political science
Psychology
Religion
Sociology
Travel