Twitter is a truly massive information sharing platform. While a lot of that can consist of people's minutia of their day, occasionally we can pull some interesting and useful from the noise. As an example of the former, I recently spoke to a researcher who had attempted to grab some data on how the top trends can vary across time. His main finding was that Twitter has such a high noise to signal ratio that it's essentially a pointless tool for people to share anything meaningful. Second, people seem very keen on discussing whatever a Bieber might be.
In contrast, a collaborative effort from some researchers at Northeastern University and Harvard seem to have had a more fruitful experience. They have pulled some 300 million tweets from across the USA in a bid to visually represent mood variation across geographical location across time. Tweets appear to have been analysed using the ANEW system, which is a rating scale for words based on three dimensions of arousal (calm/excited), valence (positive/negative) and dominance (degree of control). For example, fear has a high arousal, low valence, low dominance; fork has neutral arousal, valence and dominance; pride has high arousal, neutral valence and high dominance. The below video shows a time-lapse of a typical day's fluctuations in mood (negative mood, red to positive mood, green). The state geography is represented in cartogram form, displaying tweet density rather than land area.
The high temporal resolution of the data allows them to examine how universally moods change across days. Further to this, as the data was collected for a whole month, weekly trends can be picked out. Mislove et al. indicate that the moods described on Twitter follow a cyclical pattern here. The results are remarkable in their familiarity: in a typical week our moods are fairly constant from Monday to Friday (with the lowest point coming on a Thursday), mood picks up at the weekend, daily mood can be approximated by a sine wave with peaks around morning & evening and troughs late at night & early afternoon. All of these mood dynamics are well documented (e.g. in the book Changing Moods).
One consideration though, how accurately do people present themselves online? Is it not tempting to present yourself as and idealised version of you who always has a super exciting enjoyable life? Firstly, as the results linked above appear to match typical data so well, it doesn't look like that everyone is being inauthentic. Secondly, recent research indicates that despite the reputation online personas have for being projections of ideal or imaginary selves, there is a pretty decent relationship between our crafted Facebook profiles and our genuine 'real-life' selves.
An interesting, if not immediately plausible followup would be to try to pick out 'follower networks' within the tweets and determine whether this has any impact on moods. Can moods ripple through virtual networks as they are argued to do so though offline interaction?
In contrast, a collaborative effort from some researchers at Northeastern University and Harvard seem to have had a more fruitful experience. They have pulled some 300 million tweets from across the USA in a bid to visually represent mood variation across geographical location across time. Tweets appear to have been analysed using the ANEW system, which is a rating scale for words based on three dimensions of arousal (calm/excited), valence (positive/negative) and dominance (degree of control). For example, fear has a high arousal, low valence, low dominance; fork has neutral arousal, valence and dominance; pride has high arousal, neutral valence and high dominance. The below video shows a time-lapse of a typical day's fluctuations in mood (negative mood, red to positive mood, green). The state geography is represented in cartogram form, displaying tweet density rather than land area.
The high temporal resolution of the data allows them to examine how universally moods change across days. Further to this, as the data was collected for a whole month, weekly trends can be picked out. Mislove et al. indicate that the moods described on Twitter follow a cyclical pattern here. The results are remarkable in their familiarity: in a typical week our moods are fairly constant from Monday to Friday (with the lowest point coming on a Thursday), mood picks up at the weekend, daily mood can be approximated by a sine wave with peaks around morning & evening and troughs late at night & early afternoon. All of these mood dynamics are well documented (e.g. in the book Changing Moods).
One consideration though, how accurately do people present themselves online? Is it not tempting to present yourself as and idealised version of you who always has a super exciting enjoyable life? Firstly, as the results linked above appear to match typical data so well, it doesn't look like that everyone is being inauthentic. Secondly, recent research indicates that despite the reputation online personas have for being projections of ideal or imaginary selves, there is a pretty decent relationship between our crafted Facebook profiles and our genuine 'real-life' selves.
An interesting, if not immediately plausible followup would be to try to pick out 'follower networks' within the tweets and determine whether this has any impact on moods. Can moods ripple through virtual networks as they are argued to do so though offline interaction?
Posts
I presume the results are aggregated across every user and there's no way to pick them out, even anonymously by previous tweet mood? It might be interesting to see if there are seperate groups of users who are more happy morning people, and some who hate mornings, and whether they have different mood profiles.
ReplyDelete