3. Types of Tweet Contents (Naaman et al, 2010)
suggest a new term – “Meformers” (80%). Figure 3 shows
the mean of the average proportion of messages in the top
four categories for each user. For instance, on average
Informers had 53% of their messages in the IS category,
while a significant portion (M=48%) of the messages
posted by Meformers were “Me Now” messages. Indeed,
the figure suggests that while Meformers typically post
messages relating to themselves or their thoughts, Informers
post messages that are informational in nature.
Figure 3. Mean user message proportions for the four main
Dr Max L. Wilson categories, breakdown by http://cluster.
cs.nott.ac.uk/~mlw/
(Table 1). As mentioned, the coders were allowed to assign
multiple categories to each message. Each message was
assigned to two coders; to resolve discrepancies between
coders we simply assigned to each message a union of
categories assigned by the coders. The short length of
Twitter messages meant a lack of context that did not
permit a simple resolution to coder differences. Instead, we
opted to consider all interpretations of the messages by
coders. Over-coding was not a problem as messages had 1.3
sharing (IS; 22% of messages were coded in that category),
opinions/complaints (OC), statements (RT) and “me now”
(ME), with the latter dominating the dataset (showing that,
indeed, “it’s all about me” for much of the time).
Figure 1. Message Category Frequency.
Figure 2 considers the proportion of users’ activity
dedicated to each type of content out of 10 messages coded
for each user. The figure focuses on the four most popular
categories shown above, and the blue area in each section
represents all users. For example, the ME histogram shows
that 14% of all users had 0-10% (left-most column) of their
messages in the “Me Now” category; on average, users had
41% of their messages in “Me Now”. The figure contrasts
the span of activities of the network: most people engage in
some scale of ME activity, while relatively few undertake
information sharing as a major activity..
Code Example(s)
Information Sharing (IS) “15 Uses of WordPress
<URL REMOVED>”
Self Promotion (SP) “Check out my blog I updated 2day 2
learn abt tuna! <URL REMOVED>”
Opinions/Complaints
(OC)
“Go Aussie $ go!”
“Illmatic = greatest rap album ever”
Statements and
Random Thoughts (RT)
“The sky is blue in the winter here”
”I miss New York but I love LA...”
Me now (ME) “tired and upset”
“just enjoyed speeding around my
lawn on my John Deere. Hehe :)”
Question to followers
(QF)
“what should my video be about?”
Presence Maintenance
(PM)
“i'm backkkk!”
“gudmorning twits”
Anecdote (me) (AM) “oh yes, I won an electric steamboat
machine and a steam iron at the
block party lucky draw this morning!”
Anecdote (others) (AO) “Most surprised <user> dragging
himself up pre 7am to ride his bike!”
Table 1. Message Categories.
(Table 1). As mentioned, the coders were allowed to assign
multiple categories to each message. Each message was
assigned to two coders; to resolve discrepancies between
sharing (IS; 22% of messages were coded in that category),
opinions/complaints (OC), statements (RT) and “me now”
(ME), with the latter dominating the dataset (showing indeed, “it’s all about me” for much of the time).
Figure 1. Message Category Frequency.
Figure 2 considers the proportion of users’ activity
dedicated to each type of content out of 10 messages coded
for each user. The figure focuses on the four most popular
categories shown above, and the blue area in each section
represents all users. For example, the ME histogram shows
that 14% of all users had 0-10% (left-most column) of messages in the “Me Now” category; on average, users 41% of their messages in “Me Now”. The figure contrasts
Code Example(s)
Information Sharing (IS) “15 Uses of WordPress
<URL REMOVED>”
Self Promotion (SP) “Check out my blog I updated 2day 2
learn abt tuna! <URL REMOVED>”
Opinions/Complaints
(OC)
“Go Aussie $ go!”
“Illmatic = greatest rap album ever”
Statements and
Random Thoughts (RT)
“The sky is blue in the winter here”
”I miss New York but I love LA...”
Me now (ME) “tired and upset”
“just enjoyed speeding around my
lawn on my John Deere. Hehe :)”
Question to followers
(QF)
“what should my video be about?”
Presence Maintenance
(PM)
“i'm backkkk!”
“gudmorning twits”
Anecdote (me) (AM) “oh yes, I won an electric steamboat
machine and a steam iron at the
block party lucky draw this morning!”
Anecdote (others) (AO) “Most surprised <user> dragging
himself up pre 7am to ride his bike!”
Table 1. Message Categories.
Meformers vs Informers
4. Likelihood of ReTweeting a Tweet
(Naveed et al, 2011)
Increasing Likelihood Decreasing Likelihood
• URLs
• Especially with @username or
#hashtags
• More intense than plain
- positive or negative
• Using negative emoticons
• Using a question mark
• Directed at a Person
• Using positive emoticons
• Using an exclamation mark
Dr Max L. Wilson http://cs.nott.ac.uk/~mlw/
5. Tweets about Depression
Information Dissemination
Self Disclosure
Social Engagement
Self Disclosure is more angry
Dr Max L. Wilson http://cs.nott.ac.uk/~mlw/
7. Tweet Usefulness: Some factors make tweets more
useful to consumers
Favouriting Tweets: lots of uses,
but only 1 button
Tweet Style: should businesses be fun, serious, polite,
cheeky? You wont believe the answer. #4 is my favourite.
Dr Max L. Wilson http://cs.nott.ac.uk/~mlw/
8. Tweet Usefulness
Hurlock, J. and Wilson, M. L. (2011) Searching Twitter: Separating the Tweet from the
Chaff. In: 5th International AAAI Conference on Weblogs and Social Media (in press).
• 1) a temporal monitoring task
- Our Task: whats happening at a current festival
• 2) a subjective product task
- Our Task: information about the forthcoming iPhone
• 3) a location-sensitive planning task
- Our Task: where to eat in a part of London
Dr Max L. Wilson http://cs.nott.ac.uk/~mlw/
10. Useful Tweets: 6 Factor-Groups
• 4 Content Factors
Personal Experience Direct Recommendation
Social Knowledge Specific Information
• 2 Subjective Factors
Entertaining Shared Sentiment
• 2 Relevance Factors
Recency (Time) Correct Location
Dr Max L. Wilson http://cs.nott.ac.uk/~mlw/
11. Useful Tweets: 6 Factor-Groups
• 3 Trust Factors
Trusted Author Trusted Avatar Trusted Link
• 3 Link Factors
Actionable Link Media Link Info. Link
• 2 Response Factors
Retweeted Lots Real Conversation
Dr Max L. Wilson http://cs.nott.ac.uk/~mlw/
12. Not-Useful Tweets: 5 Factor-Groups
• 2 Anti-Trust Factors
Un-trusted Author Un-trusted Link
• 2 Irrelevance Factors
Out of Date Incorrect Location
• 2 Response Factors
Question without Answer Repeated Content
Dr Max L. Wilson http://cs.nott.ac.uk/~mlw/
13. Not-Useful Tweets: 5 Factor-Groups
• 8 Content Factors
No Information Introspective Off Topic
Too Technical SPAM Content Dead Link
Poorly Constructed Wrong Language
• 3 Subjective Factors
Too Subjective Disagreeable Not Funny
Dr Max L. Wilson http://cs.nott.ac.uk/~mlw/
14. Useful Tweets have Multiple Factors
Specific Fact
Useful Information
Useful Link
Dr Max L. Wilson http://cs.nott.ac.uk/~mlw/
15. Not-Useful Tweets have a Clear Flaw
Its in dollars!
Dr Max L. Wilson http://cs.nott.ac.uk/~mlw/
16. Tweet Usefulness: Some factors make tweets more
useful to consumers
Favouriting Tweets: lots of uses,
but only 1 button
Tweet Style: should businesses be fun, serious, polite,
cheeky? You wont believe the answer. #4 is my favourite.
Dr Max L. Wilson http://cs.nott.ac.uk/~mlw/
18. What motivates people to
favourite tweets?
Work with Meier & Elsweiler
• Large-scale survey (n=606)
- Generic subjective questions
- Actual Favourited Tweets
- Critical Incident questions
• Analysis - ‘Almost Perfect Agreement’
- Iterative Content Analysis
- Affinity Diagramming
Dr Max L. Wilson http://cs.nott.ac.uk/~mlw/
19. “25” Reasons Found
• Actually more of a hierarchy of reasons
• 1 Category was ‘no reason’
• 2 Main Categories:
- A response to the tweet, content, user, situation
- A functional purpose
Dr Max L. Wilson http://cs.nott.ac.uk/~mlw/
31. Implications of Favouriting
• The use of the fav button is really overloaded
- agreeing, liking, re-finding, to-do
• Favouriting vs RTing
- they imply different things
• Several platforms have single entities, and a similar button
- do these situations apply in e.g. tumblr?
Dr Max L. Wilson http://cs.nott.ac.uk/~mlw/
32. Implications for Content Marketing?
• To increase favourites?
- post things that need follow-ups
- post things that people want to keep
- post things that are objectively likeable
- post things that invoke emotion/memory
- post things that are subjectively likeable
(understand your audience)
- make your campaign ‘human’ so people engage non-verbally
- post about people - not just too people?
Dr Max L. Wilson http://cs.nott.ac.uk/~mlw/
33. Tweet Usefulness: Some factors make tweets more
useful to consumers
Favouriting Tweets: lots of uses,
but only 1 button
Tweet Style: should businesses be fun, serious, polite,
cheeky? You wont believe the answer. #4 is my favourite.
Dr Max L. Wilson http://cs.nott.ac.uk/~mlw/
34. blocks of social media that executives could use media [41].
How should businesses behave?
(Work with Nathan Bratby)
According to the paper, each of these blocks to provide a trait of functionality and a resulting feature. These blocks are show in figure 2.2.
• Keitzmann, Hermkens
& McCarthy (2011)
- 7 functional uses
• Lots of horror stories
- nestle arguing with customers
- #mcdstories
- Luton Airport
Dr Max L. Wilson http://cs.nott.ac.uk/~mlw/
35. Mixed Method Investigation
21
• Business Account example:
data analysis
- Tweets per day, RTs, Favs, Popularity, Links, Hashtags
- Use of Emoticons, Signing name
- Formality Analysis
“They destroyed a building.” and “A building was destroyed.”
The latter statement is more formal. This process is called nominalization [30].
Heylighen used all this information to deduce the following formula:
Formula 1 - Heylighen
• Consumer Survey Data
- Ideal posting frequency
- Preference for message/formality types
The frequencies are expressed as percentages of the number of words in that
category, with respect to the total number of words. The more formal the language
excerpt, the higher the value of F is expected to be, given in a percentage [30]
By using this formula on the dictionaries of Italian, Dutch, French, and English,
the researchers found similar results. They also discovered that written language
scored a much higher formality frequency than that of spoken. In order to test this
formula further, the researchers opted to compare their results to the Dutch list
Dr Max L. Wilson http://cs.nott.ac.uk/~mlw/
36. Emoticons & Name Signing
62
The formality analysis provided more insight into the tweeting style some of
the more popular accounts use. Before that though, manual analysis was used to
discover the traits of the accounts, in terms of regular emoticon use and employees
signing their names at the end of tweets.
Figure 4.17: Table showing use of emoticons Dr Max L. Wilson http://cs.nott.ac.uk/~mlw/
37. Formality by Type
News Companies
Retail Companies
Support Accounts
0 15 30 45 60
Formality Score
Dr Max L. Wilson http://cs.nott.ac.uk/~mlw/
38. The final part of the formality analysis was to look at a company in depth,
Different Types of Posts
which provide a range of di↵erent type of tweets to see if their formality di↵ers
in each situation. Due to time restrictions, only one business was selected for
this. The company chosen was NandosUK, who as well as promotional tweets,
also regularly reply to customers with queries, complaints or praise. 100 of each
tweets were analysed for formality and the results are shown in figure 4.24.
Figure 4.24: Table showing the results of each type of tweet from Nando’s
Surprisingly, the responses to complaints were least formal. This can be seen
more easily on the graph in figure 4.25.
Once all the necessary results were gathered it was possible to compare the
formality of each of these companies with the popularity of the tweets to see if
there was any sort of correlation that indicated the best formality practice. For this
graph, the ArgosHelpers outlier has been removed. This is shown in figure 4.26.
Dr Max L. Wilson http://cs.nott.ac.uk/~mlw/
39. Mimicking Audience Language
So far - an informal observation
• “Popular” companies ‘mimic’ audience language
• Mirror Football
- more ‘football banter’ than GuardianSport
- GuardianSport don't use emoticons, MirrorFootball does
• Tesco Mobile
- less formal than ThreeUK
Dr Max L. Wilson http://cs.nott.ac.uk/~mlw/
40. The appropriate language for a company to use on Twitter is not always
straight forward. A recent example of this is an Argos employee’s response to
a disgruntled customer, whom complained to their Twitter account with the use
of heavy slang. The response was Argos, mimicked this linguistic style, to the
extent that some argued they were mocking the customer. This reply went down
well with both the customer, and general population, quickly going viral and accu-mulating
Extreme Mimicking
thousands of retweets [52]. The tweet in question is shown in figure 2.17.
Figure 2.17: Image displaying Argos’ viral tweet
Dr Max L. Wilson http://cs.nott.ac.uk/~mlw/
43. Accommodation Theory
• People adapt their communication style
• To accommodate the receiver
• This makes the receiver more relaxed
- and ready to engage
• Open Question: Is this an effective business strategy?
Dr Max L. Wilson http://cs.nott.ac.uk/~mlw/
44. Tweet Usefulness: Some factors make tweets more
useful to consumers
Favouriting Tweets: lots of uses,
but only 1 button
Tweet Style: should businesses be fun, serious, polite,
cheeky? You wont believe the answer. #4 is my favourite.
Dr Max L. Wilson http://cs.nott.ac.uk/~mlw/