What People Say When They Tweet
Everyone is talking about Twitter (to the consternation of many of our readers, I'm sure), but what are people talking about on Twitter? It is really just a flood of inane status updates and fragmented chat, or are people actually talking about things that matter to them via the service? We've talked about Twitter as a platform for serious discussion, but is anyone really using it that way? We teamed up with Summize to take a closer look about what people are talking about in the Twitosphere.
Summize looked at about 4 million Twitter status update messages (tweets) collected from the public time line over a seven day period running from April 27 - May 3. We saw approximately 200,000 active users (users that sent at least one message) during that period, of which 60% tweeted in English. Japanese was the second most popular language on Twitter with about 9% of the tweets that we gathered.
Most strikingly, we found that a whopping 5% of all Twitter (in terms of tweets) is powered by the top 100 active accounts. Those 100 users post updates 200-3000 times a day, which might even impress uber-Twitter users like Robert Scoble or Jason Calacanis.
What we saw was that the top Twitter users are not always people, but rather, they're sources using Twitter as a feed publishing platform. For example, the most active user we saw was an account called "lejddfr," which does "push journalism" (sending out frequent links to new stories) for French news service Le Journal du Dimanche. lejddfr has just 315 followers, is following no one, and has made over 101,000 tweets. Another example we saw is "GSSP," which tweets links to stories on the news web site NewzOf.com. Our favorite high volume Twitter user, though is "chandraxray," a space station that tweets its location up to 350 times per day.
To get a feel for the distribution of tweeting activity across the Twitter universe, we sorted users by the number of tweets per day and graphed that vs. the total traffic. 66% of the users only tweet once per day.
But that still doesn't answer the question of what people are talking about. So we started analyzing tweets and came up with a list of the top five words people send. Apparently, the first thing anyone sends out on Twitter is a "test" message.
- test
- lol
- working
- :)
- sleeping
Unfortunately, while we can guess from that list that Twitter users tend to be sleepy workaholics who are generally happy, we still can't tell much about what they're really talking about on the service. So we next applied some fancy topic extraction and started calculating trends over our week's worth of data.
What we found is that there are three main types of conversations going on. First, there are status updates of every day occurrences such as, "getting coffee," "check out this post on X," "going to sleep," or other mundane life things. Second, there are short term memes where many people talk about some event before, during, or after it. These conversations are usually short lived -- ranging from a few minutes to a few hours. For example a TV show like "Lost" will have some buzz, before, during, and for a short time after the show airs, but will drop out of the stream very quickly. We saw that happen with "LSD" when the drug's creator Albert Hoffman died last week. The final type of discussion we see on Twitter, are long term memes. These are topics of interest that people talk about for days, weeks, or even months. Politics or new video games are great examples of these longer term discussions happening on the platform.
Below we selected a subset of the automatic trends found for the week to illustrate this phenomenon. Each column from left to right shows the days of the week. Topics on the top of each stack represent shorter lived memes, while topics lower on the chart represent items of longer term interest.
You can click on the links below to explore the memes on those days:
Sunday, April 27:
Obama,
GTA IV,
Mario Kart Wii,
Coachella,
Facebook,
NFL,
BBQ,
Ubuntu,
Spurs
Monday, April 28:
Obama,
GTA IV,
Mario Kart Wii,
Coachella,
Rev Wright,
iMacs,
Facebook,
Social Brew
Tuesday, April 29:
Obama,
GTA IV,
Iron Man,
Rev Wright,
Ben Jerry,
Celtics,
Lakers,
Deal or No Deal
Wednesday, April 30:
Obama,
GTA IV,
Iron Man,
American Idol,
LSD,
iGoogle,
Violet Hill,
Neil Diamond,
Spurs
Thursday, May 1:
Obama,
GTA IV,
Iron Man,
RSS Awareness Day,
Diggnation,
Baskin Robbins
Friday, May 2:
Obama,
GTA IV,
Iron Man,
DC Madam,
Hawks,
BSG,
Lost
Saturday, May 3:
Obama,
GTA IV,
Iron Man,
Kentucky Derby,
Free Comic Book Day,
Maker Faire,
YouTube,
Boris
Conclusion
While technology, politics, and geekery (sci-fi movies and video games) tend to dominate the long term memes, people are discussing all sorts of things on Twitter -- from sports to pop culture to cooking. Could that indicate that the site it starting to have some mainstream appeal? Or maybe just that even us tech geeks occasionally find time to talk about things other than technology? Either way, the way memes flow on Twitter is an interesting topic and one that we had fun looking at.
Special thanks to Dr. Abdur Chowdhury, co-founder of Summize, a conversational search engine. Dr. Chowdhury did all of the data mining and analysis for the this post, as well collaborated on the text and created the charts.
Comments
Subscribe to comments for this post OR Subscribe to comments for all ReadWriteWeb posts
thank you, it is very intersting post !
Posted by: Peter | May 9, 2008 3:18 PM
Josh, this is super interesting. I like http://www.tweetcloud.com/ for providing a bottom up view of my own activity. Turns out that I am a super-passionate advocate NOT of my own startup UpTake, but of Twitter itself! I figured that was true for all users (since everyone I follow talk about Twitter) but apparently that is not true.
@elliottng
Posted by: Elliott Ng | May 9, 2008 3:53 PM
Or how about the dude who has 1 follower, is following 1 person, but has >20,000 updates?...
http://logicalextremes.blogspot.com/2008/05/in-twitter-about-twitter-gross.html
Posted by: Logical Extremes | May 9, 2008 5:32 PM
I find it interesting that 66% of Twitter users tweet only once per day!
Would be great if you can graph the pattern of your database based on number of tweets per day - e.g. 1, 5 or less, 6-25 and more than 25 per day - just to see what fraction falls into each group.
Without recourse to hard data that you have, I still did some guesstimating about Twittographics!
http://www.MoneyPowerWisdom.com/twittographics/
All success
Dr.Mani
Posted by: Dr.Mani | May 10, 2008 8:37 AM
And I have re-tweeted this!
Posted by: Deeps | May 10, 2008 9:18 AM
Boris Johnson and Boris Yeltsin on the same day (May 3rd). Now that shows the diversity of Twitter ;-)
In case you don't know, Boris Johnson ran for the UK Mayor of London around that date. (BTW he won).
I think we also need to consider that like most technologies, there are some very distinct "modes"
Loved ones and friends will tend to twitter only the life stuff.
Professionals will tend to meme about their industry and/or their own company.
Services will tend to just ping updates.
What I think is interesting, is that there is little overlap.
I think this is because your "Twitter circle" is either social or business.
Both exist on Twitter, but i think disparately
Posted by: Peter | May 10, 2008 11:12 AM
Interesting but I would argue that stats can only take us so far. Even knowing the memes does not enable us to see how the broader discourse has been put together, what issues and ideas are being brought together or linked (even in just 140 characters). We need to look at the grammatical structures, the tenses, active or passive voice etc as well as the vocabulary if we are to understand the way it works and how we/our issue/our 'brand' is being talked about and our 'business story' being co-told. We need a qualitative Conversation Audit (as I call it) not just a quantitative monitor.
Posted by: Paul Caplan | May 12, 2008 11:20 AM
I've decided to use it to extend my own blog and help form a community of beer homebrewer online. It seems to be working a bit. Its novel and interesting so it attracts.
Our convo is generally craft beer/homebrew related, however, we seem to be beer people who sometimes talk about beer. Many other topics enter into it...like a first hand tornado experience. Rather than always talking (twittering) about beer.
Imagine for a minute, that we could hear a sample of a thousand traditional (verbal) conversations from all over the world, what would we hear? How long would we have to listen until we heard common threads of current events being mentioned? Just thinking out loud here...
The faster info travels the faster disinformation travels.
The faster info travels the faster useful information travels.
I suspect there will be an improved level of fidelity on the edge of this new wave of twitter communication. Once we all get used to it and learn how to exploit it noise will increase and fidelity will decrease.
All the time we will keep asking the question. "What are people communicating about."
Posted by: Adam @ Beer Bits 2 | May 12, 2008 12:16 PM
Paul: I agree with you that a more complete language evaluation can take us much further, but it's still quite hard to automate.
However, one can do a lot (more) with quantitative methods as well, like extract the main themes of every post and see the combination of the brand name you are looking for and other themes (critical, comparative etc.) that come with it. One doesn't get the exact attitude towards the brand maybe, but would still get some meaningful information I would guess.
Trying to explore this more in our Topify project (link in name). Critics welcome :)
Posted by: Martin | May 13, 2008 7:09 AM
I really think summize is fine, still deliberating between it and tweetscan to be quite honest. I am running some research and from that data, i'll provide detailed comments about both (may take a month or so).
I did want to bring to your attention (being an academic myself) that a Master of Science dissertation was defended on twitter topics in the late fall 2007 at the London School of Economics and Political Science. Read more http://enzaac.wordpress.com/2008/04/21/msc-thesis-on-140-characters-or-less .
Posted by: iVenus | May 14, 2008 7:14 AM
Josh Catone: sorry to bother, but there seems to be some problem with my trackbacks. I had this recently and now again I linked to your post and tried to ping you, but either you didn't get it or I did something wrong and you rejected it.
Would be very obliged if you told me what I'm doing wrong. The post is here: http://topify.wordpress.com/2008/05/13/what-people-say-when-they-talk/
Posted by: Martin | May 14, 2008 9:02 AM
@Martin: Sorry about that. There's a fourth option here... MovableType trackbacks are just strange. Sometimes they just don't seem to work right. But thanks for leaving the link in the comments. :)
Josh: Oh, no probs :) Thanks for the reply
Posted by: Martin | May 14, 2008 9:14 AM
I am in love with Summize. Great design, great API, great labs apps. So excited to play with this. Thanks for the thorough, data-rich report, RWW!
Posted by: Robin | May 15, 2008 6:03 PM