Académique Documents
Professionnel Documents
Culture Documents
Los
Alamitos, CA: IEEE Press.
Figure 1. Tweets posted by language and time (all times Eastern Standard Time)
The researchers found that for any given time period, that 12.5% of all tweets in their collection contained an
the communication appeared highly incoherent; almost @ sign. This suggests that interactive uses of Twitter
no conversational exchanges seemed to be taking are increasing rapidly and that this is a global trend.
place. When messages unrelated to exchanges between Although English tweets were the most common in
dyads were cleaned from the data, however, extended all four time periods, they were most frequent (68%) in
dyadic conversations could be identified that were two time periods: 2 p.m. and 6 p.m. For the remaining
moderately coherent. However, they were far less co- research questions, we focused on tweets posted in the
herent than IM or IRC exchanges visualized using the 6 p.m. sample, as it had fewer warnings of missed
same method. The present study partially replicates tweets from Twitter Scraper than the 2 p.m. sample.
Zelenkauskaite and Herring’s [18] methods, so as to be
able to compare the Twitter coherence results with 5.2. Functions of @
those for other CMC conversation.
The second research question sought to categorize
5. Findings and quantify the varying uses of the @ sign in English
tweets. In the 6 p.m. sample, 1472 English tweets em-
5.1. Language of tweets ploying an @ sign were found. The breakdown of their
functions is summarized in table 2.
As might be expected, Twitter posting activity var- Nearly 91% of the @ signs appearing in the 6 p.m.
ies according to time zone. English tweets are most sample were used to direct a tweet to a specific ad-
common overall, followed by Japanese and Spanish. dressee. The next most common use of the @ sign was
These results are displayed in Figure 1. to refer to another user, followed by indicating one’s
The different language groups appear to make use current location. In this sample, 28 users included mul-
of the @ sign with equal frequency. Overall, 30% of tiple @ signs in their tweets for different purposes. An
tweets in the 6 a.m. and 6 p.m. samples, 31% of tweets analysis of a random sample of 50 English tweets em-
in the 10 a.m. sample, and 32% of the tweets in the 2 ploying the @ sign from the other time periods sug-
p.m. sample use the @ sign. This is more than twice as gests that the percentages in Table 2 are representative
high as the rate reported in Java et al. [8], who found of the larger corpus.
Table 2. Functions of the @ sign in English tweets The most popular content of tweets is reporting one’s
own experience, consistent with the stated purpose of
Number % Twitter to answer the question “What are you doing?”
Addressivity 1339 90.96% Fifty-one percent of tweets without @ signs, and 17%
Reference 80 5.43% of tweets with @ signs, appear to address this question.
Locational ‘at’ 11 0.75% Moreover, interesting differences can be observed
Non-locational ‘at’ 5 0.34% between the content of tweets with and without the @
sign. Tweets with @ signs are more focused on an ad-
Email 1 0.07%
dressee, more likely to provide information for others,
Emoticon 1 0.07% and more likely to exhort others to do something—in
Other 7 0.48% short, their content is more interactive. In contrast,
Address / Refer 20 1.36% tweets without @ signs are more self-focused, although
Address / Location 3 0.20% they also report other’s experiences, and they make
Address / Email 2 0.14% more general announcements. The low incidence of
addressee orientation (2%) and exhortations (1%) in
Address / Other 1 0.07%
tweets without @ suggests that, while users may
Refer / Other 2 0.14% choose in theory to respond to other tweets without
Total 1472 100.00% using the @ sign, in practice such unmarked responses
are infrequent.
5.3. Content of tweets with and without @
5.4. Responses to @
We next asked what people twitter about, and if the
content of tweets differs when the @ sign is used to Because an initiation must receive a response in or-
direct a message to a specific individual, as opposed to der for conversation to occur, the fourth research ques-
posting it for oneself or the general Twitter readership. tion asked how many messages directed to others re-
The main themes of a random selection of 207 English ceived one or more responses. To answer this question,
tweets in the 6 p.m. sample are summarized in Table 3. we examined all English messages with and without @
signs posted to Twitter within the first 30 minutes of
Table 3. Content of a random sample of English the 6 p.m. sample and identified public responses (by
tweets searching for the @ sign) appearing in the remainder of
w/@ w/o @ Total the hour. Messages without @ signs were included in
about 21 2 23 this analysis, because we observed that some undi-
addressee (33%) (1%) (11%) rected messages (without @ signs) were responded to,
announce/ 0 14 14 and thus became post facto initiations. However, it was
advertise (0%) (10%) (7%) less common for responders not to use an @ sign to
7 1 8 indicate the addressee, as noted above.
exhort
(11%) (1%) (4%) This is a very conservative measure of responsive-
info for 10 1 11 ness, in that it does not include responses that were
others (16%) (1%) (5%)
info for 2 9 11
posted after the one-hour time frame, responses that
self (3%) (6%) (5%) did not employ the @ sign, or private responses. Even
meta-com- 0 4 4 without including these, of the 785 public English
mentary (0) (3%) (2%) tweets that used the @ sign to direct the message to a
media 4 14 18 particular individual that were posted in that half hour,
use (6%) (10%) (9%) 245 messages (31.2%) received a public response.
express 5 8 13 This rate is considerably higher than the rate of 7%
opinion (8%) (6%) (6%) reported by Zelenkauskaite and Herring [18] for public
other’s 1 10 11 responses received to iTV SMS within a two-day pe-
experience (2%) (7%) (5%)
self 11 73 84
riod. Moreover, the actual response rate to tweets with
experience (17%) (51%) (41%) @ signs, including data not accessible to us, is almost
solicit 0 3 3 certainly higher. This measure thus suggests that the
info (0) (2%) (1%) responsiveness of Twitter as a conversational environ-
other 2 5 7 ment is at least at the low end of moderate and proba-
misc (3%) (3%) (3%) bly higher.
Total 63 144 207
5.5. Coherence of Twitter conversations discussion, the conversation drifts towards comment-
ing on the tool used to create the screen capture.
The fifth research question asked: How long, and Figure 2 displays the evolution of topics in this ex-
how coherent, are Twitter conversations? The 6 p.m. change over time as diagramed using VisualDTA. In
sample contained 123 exchanges in English, where an DTA diagrams, T indicates a proposition that is nar-
exchange was defined as at least one initiation and one rowly on-topic with the proposition to which it re-
response involving at least two people, with or without sponds, while P indicates a parallel shift that introduces
use of the @ sign. Table 4 shows descriptive statistics a new idea into the discourse. The y-axis represents the
of the exchanges for five categories: the number of messages in chronological order, and the x-axis repre-
participants in each exchange; the length of the ex- sents cumulative semantic distance between proposi-
change (within the one-hour time frame); the number tions. Figure 2 shows that this conversation is continu-
of exchanges; the percentage of tweets employing the ous and proceeds in a mostly gradual, step-wise fash-
@ sign to address the tweet to a participant; and the ion (via parallel shifts with a semantic distance of 1)
average time between tweets within the exchange. from the upper left to the lower right. This pattern,
including its tendency toward topic drift, is characteris-
Table 4. Characteristics of Twitter conversations tic of a coherent dyadic conversation, such as takes
place, for example, via instant messaging [18].
Range Mean Median Mode
# Participants 2 - 10 2.5 2 2
Length 00:25 -
(min:sec) 54:22 17:41 26:33 -
# Exchanges 2 – 30 4.62 3 3
% Tweets
using @ per 20% -
exchange 100% 86% 100% 100%
Avg time bt
tweets in ex- 00:25 -
changes 34:05 06:43 04:24 -
[6] Herring, S. C. (2003). Dynamic topic analysis of syn- [16] Werry, C. C. (1996). Linguistic and interactional fea-
chronous chat. In: New Research for New Media: Innovative tures of Internet Relay Chat. In S. C. Herring (Ed.), Com-
Research Methodologies Symposium Working Papers and puter-mediated communication: Linguistic, social and cross-
Readings. Minneapolis, MN: University of Minnesota School cultural perspectives (pp. 47-63). Amsterdam: John Benja-
of Journalism and Mass Communication. Retrieved June 15, mins.
2008 from http://ella.slis.indiana.edu/~herring/dta.2003.pdf
[17] Wikipedia. (2008, September 7). Twitter. Retrieved
[7] Herring, S. C., & Kurtz, A. J. (2006). Visualizing Dy- September 7, 2008 from http://en.wikipedia.org/wiki/Twitter.
namic Topic Analysis. Proceedings of CHI’06. ACM Press:
New York. Retrieved September 8, 2008 from [18] Zelenkauskaite, A., & Herring, S. C. (2008). Television-
http://ella.slis.indiana.edu/~herring/chi06.pdf mediated conversation: Coherence in Italian iTV SMS chat.
Proceedings of the Forty-First Hawai’i International Con-
[8] Java, A., Song, X., Finn, T., & Tseng, B. (2006, August). ference on System Sciences (HICSS-41). Los Alamitos, CA:
Why we Twitter: Understanding microblogging usage and IEEE Press.