As part of my ongoing research into sentiment on Twitter, I've spent much of my time refining the data visualisations to highlight patterns and trends in a much better way. In this entry I pick out a couple of the newest visualisations that I think are interesting.
Plain old average sentiment
This is how I originally visualised the sentiment. The concept is simple; you just plot the average sentiment for each hour of the day on the graph in the position that it's supposed to be. There is nothing intelligent happening with this method, and the result is a graph that looks quite erratic and hard to read.
Fancy moving average sentiment
The most interesting and recent progression has been with the shift to a moving average for the sentiment values. By utilising a moving average I've been able to smooth out the data so erratic fluctuations don't dramatically affect the average sentiment. For example, the following graph is for the same period as the graph above, just with smoothing added. Notice how the general trend of the sentiment is more apparent as the week goes on – the happiness drops like a stone on Monday, then slowly picks up again by Saturday.
This has had a pretty profound effect on the way I look at the sentiment data. Another cool example is the week during the run-up to Easter; notice the happiness levels increasing dramatically as a weekend of chocolate approaches.
It's all pretty cool. I'm hoping to take this further and really push the data to look for interesting trends and patterns with sentiment on Twitter. Who knows what I'll find!