Search the Blog
All our latest data stories and insights straight to your inbox
You’re now subscribed to the
Brandwatch React newsletter.
Find it in your inbox every
Product Update: Demographics Now With Editing Product
Back in February, we launched Demographic Insights – a suite of useful data and components all about helping you better understand online audiences.
It tells our users the gender, interests, profession, location and account type (whether it’s an individual or a business account) of Twitter authors within their data.
It was a big step forward in terms of giving our users even more information to better analyse and understand the online audiences discussing their brand or topic of interest (we used it ourselves to analyse Valentine’s Day conversation, and even House of Cards chatter).
Here’s more on the feature in case you missed it:
This weekend, we released the next major update: editing.
This allows users to edit a classification for any of our demographic categories if they don’t agree with it, or to add a category if it is classified as ‘unknown’, making our demographics data even better, for all our users.
For example, our gender classification is based on first names – which makes it somewhat tricky when we find a Jamie or a Sam, or when a user doesn’t put in their real first name.
In that case, we usually leave the author or mention unclassified, preferring quality over quantity when it comes to classifications and accuracy.
Now, you can add those classifications yourself. If you see that Jamie or Mr no-name is clearly a man from his avatar, then you can add that in.
This is also useful for those who are tracking data in other languages, as currently we only classify Twitter bios that are in English. Now, if you need demographics data for other languages, then you an add this in manually.
When a user changes or adds a classification to an author, this will be reflected in the data for that author going forward for all users (subject to some vetting, of course), so over time the wisdom of the crowd means the data gets better and better.
The impact on accuracy
Our demographics classifications are already pretty accurate – check out these stats:
- Account type: All authors are classified as either an individual or an organisation – over 91% of the time this classification is correct.
- Gender: Over 70% of authors are assigned a gender (the rest are left as ‘unknown’). Accuracy on those assigned is around 97%.
- Interests: Just over half of authors are assigned one or more interests – bearing in mind that not all tweeters will list interests in their bio. Those interests are over 90% accurate.
- Profession: About 10-20% of authors are assigned a profession, as many tweeters do not list their profession in their Twitter bio. Professions are around 91% accurate.
- Location: All posts are assigned a location, based on various different methods of calculation. Of those, around 65% is accurate, with a further 30% indeterminable (e.g. there is not sufficient data to classify a location).
However, allowing our users to classify the unclassified, and to update any that have sneaked through the net and been classified incorrectly means that we can improve accuracy even further.
And who wouldn’t want that?
The problem with Artists
Since launching Demographic Insights, one thing we and our users have noticed is that the Artist category in professions more often than not dominates.
This is because this category is more broad than some of the others, and includes everyone from DJs to photographers and comedians. Whilst not incorrect as such, understandably users have expressed concern that this skews the data, and we agree.
Listening is important to us, so that’s exactly what we did.
We are now working on breaking down the Artist category further into new, smaller categories. That will be coming later this year and will, we hope, make the Profession breakdown even more useful and improve its accuracy.
You’ll see that if you hover over a profession classification in any table it will tell you what the term in the bio was that meant it was classified as that particular profession.
If the profession has been manually assigned (by yourself or another user) the tip will tell you so. That way, you can always understand why an author has been classified as they have, so you know you can trust the data.