Insights from Unfiltered Twitter Without Geofiltering

A couple of posts ago we wrote about the loss of information when geo-filtering is used on Twitter. The other side of that coin is the insights to be gained when topics are viewed from a macro level without any geo-filtering.

What lossless data tells us

For example, without having to apply any geofilters we can define the topic of the upcoming Australian Federal election sufficiently accurately that there are almost no irrelevant mentions.

We can see (below) that over three months up to mid-April 2016 there were 84,768 tweets and 87.1% of those came from Australia. What’s interesting is that two-thirds of the people tweeting were male, and only 1/3 female. This begs the question of further research which is easily done by digging down into the data.

If we look at sentiment surrounding the election, we can see the trend (in online News over 6 months to mid-April 2016) has become increasingly more negative. The positive sentiment has trended down, the neutral sentiment has trended down, and the positive sentiment has trended up.

This could mean that our politicians need to be especially nimble and alert during this election campaign as the public may be unforgiving and short-tempered about ploys, stunts and point-scoring. The public hardly seems to be in a mood for an election.

If we just look at Twitter (the 84,768 tweets noted above) then it’s apparent that the incumbent Prime Minister Malcolm Turnbull is making far more proficient use of social media that his opponent Bill Shorten – the latter accounting for less than 1/6 as many mentions in tweets as the former.

We can see that of the topics selected that “tax” gets the most Twitter attention in relation to the election, while innovation ranks last among the chosen topics. So it seems that tax, education and health have the attention of the Twitter Nation in relation to the upcoming election.

Finally, if we compare sentiment around various topics we can see very different outcomes across selected topics (derived from online News over 6 months to mid-April 2016).

It seems that the vague and varied potential outcomes on tax reform have jaded people’s expectations, and on the other hand the economic sentiment is quite positive. The mood about innovation is very positive, while company tax raises mixed emotions especially around the global tax debate about multinationals and their various tax schemes.

In summary, none of this data has any geofiltering and therefore is a full “census” of activity around the defined topic, in this case the upcoming Federal election in Australia.

If geo-filters had been used to isolate data then 70 to 90% of this information would have been lost. In that case the answers would be in the form of a sample rather than a census of the available data. Of course the data itself in its entirety is only a sample of opinion, but a large sample is often better than a small sample when less than millions of data points are available.

The big issues unfiltered e.g. climate change

As a final illustration, if we move beyond the Australian election and take a global topic, such as climate change, by using just topic definitions without any geo-filtering we can get an instant view of the global heat map on this issue.

If we look back over the last year 87% of activity in social media was on Twitter comprising more than 14 million tweets. The USA accounted for 51.2% of those tweets and perhaps surprisingly 5.5% came from Australia (Australia accounting for more than 10% of that of the USA while having a population far less than 10% of the USA).  Oddly enough men accounted for 61% of the tweets.

In terms of popularity over the 12 months there were large peaks of activity, which can be seen to be associated with major conferences, events and agreements. 

The examples above perhaps all raise more questions than they answer, and that’s where skilled social media analysis comes into play. The point is that where topics can be defined relatively precisely and the use of geo-filters avoided there is much more detailed data to allow the deep dive and for specific insights to be analysed and developed. This post shows a couple of examples of very relevant topic data extracted without geo-filtering.