Yesterday I performed a quick and dirty analysis of the Twitter sentiment during the Japanese and Hawaii tsunami. Although it wasn't very thorough, it certainly highlighted some interesting trends in the sentiment (the peaks and troughs).
This time around I've spent some more time analysing the Twitter sentiment from yesterday, and I've also gone and analysed the sentiment from today as well. The hope is that by comparing the two days, some insights can be made into the data.
Average sentiment per minute
The following graphs show the average sentiment on Twitter regarding the earthquake on a per minute basis. The timescales aren't identical, but it's interesting to see how stable the sentiment seems to be on Saturday, compared to Friday (the day of the earthquake). Also, notice how on both days the sentiment is trending upwards, meaning that both days ended happier than they begun.
Twitter sentiment surrounding the Japanese earthquake - Fri 11th March 2011
Twitter sentiment surrounding the Japanese earthquake - Sat 12th March 2011
Average sentiment per hour
In these graphs I lowered the resolution of the data and grabbed the same average sentiment, just this time across each hour instead of each minute. The peaks and troughs aren't so profound this time, but you can still see quite obvious differences between Friday and Saturday. One of the most noticeable differences is that Friday (the day of the earthquake) is much more unstable, perhaps as news spreads, aftershocks occur, and the tsunami travels across the pacific.
Twitter sentiment surrounding the Japanese earthquake - Fri 11th March 2011
Twitter sentiment surrounding the Japanese earthquake - Sat 12th March 2011
Taking this further
I hope to take this analysis a step further and dig a little deeper into the peaks and troughs. I'd like to tie in key events in the timeline of the disaster to see if they match up with dramatic changes in sentiment on Twitter.
Transmissions Received
Umar
One of our modules at uni, Interaction Design, we have to produce something interesting using Twitter data, for example a visualisation. I was wondering if you had any interest in open-sourcing the code you’ve used to collect the data and perform sentiment analysis.
If this isn’t possible would you mind providing any more information on how you approached this from a technical point of view, for example any libraries you used.
Thanks.
Bill
What the hell is wrong with you?
Rob Hawkes
@Bill: Do you mind expanding on that? I’m sure there’s plenty wrong with me, but I don’t know what exactly you’re referring to.
Jason
@Bill I don’t see the problem here.
The problem with this is that I assume you limit your checks to English tweets? The affected country uses Japanese so those directly affected wouldn’t appear in your analysis.
I am certain if you limited your analysis to those tweets from Japan it would be very different. Even then it wouldn’t be accurate as those who had lost someone or lost property etc wouldn’t be on twitter talking about their dismay at the event.
I would expect a graph showing dismay then rising as relief of being alive / finding love ones alive and property undamaged set in. Then falling again in days after the event as the nuclear crisis unraveled and the true extent of the event was revealed.
A more interesting study would be the sentiment towards the Japanese government by Japanese citizens.
Mike
@Bill - having followed Rob & his twitter sentiment measurements work for several months, I can tell you that his work is not attempting to do anything other than establish a somewhat objective view of people’s sentiment about local and world events. It’s a tool he’s simply used to see the effect on sentiment of this disaster.
I found this page quite encouraging as a fair number of people have commented negatively or made puerile jokes on twitter- this metric seems to indicate that on average, the twitter population are normal humans, capable of emphasising with a nation undergoing a enormous tragedy.
Phil
@Bill @Rob Hawkes - Making an assumption about what Bill is meaning - I understand it is an emotional issue, but Rob is not trying to profit from the disaster (as Microsoft did with it’s Bing advertising!). He is only making an observation on how people have reacted to the terrible event, and whether Twitter use reflects the “real world”.
@Rob Hawkes - I think it is a worthwhile analysis that doesn’t cross any lines (for me). It would be interesting to see whether the Twitter results are reflected in other media, and whether there is a similar pattern to the coverage of other events (in the sense of a strong reaction that levels off, whether getting happier or less happy).
dude
What is up the asses of these trolls? You’re simply showing a graph of average sentiment on twitter. Do they really expect the entire world to immediately get depressed when there’s an earthquake? Earthquakes happen every day. The big dip is pretty spectacular, anyway, but the point is, again, that it doesn’t matter. This is an objective view on the data, not a subjective one based on your (‘your’ here being the trolls in these comments) personal emotional viewpoint.
My tweets didn’t change _at all_ due to the events in Japan… why would they? This analysis looks spot-on to me.
Julio
Which tool you used to get those tweets?
Artesanias
The big dip is pretty spectacular, anyway, but the point is, again, that it doesn’t matter
Martin Varesio
Very nice article, and theTwitter use reflects the “real world”.