How to verify consumers’ age on social media
Quantitative insights derived from social media are rarely trusted. One of the key reasons is the lack of formal methodologies to target consumers’ in the right demographic group.
While with the current technology we can locate consumers’ provenance and gender segmentation on social media, age is still very problematic. In most cases, age is not specified or it is vaguely estimated ( eg Mums 25–44).
The consequence of this is that consumers’ panels on social media provide highly generic insights.
Even worse, social insights cannot be delivered for certain verticals such as teen products or alcoholic beverages where hard verification of age is a fundamental requirement (e.g. underage consumers discarded / included).
While age verification has been a daunting task, today, after more than 10 years of data accumulation on social media such as Facebook or Twitter, most of the answers are already available, in data spontaneously shared by consumers, including their age.
Find answers in the data
Artificial Intelligence, when efficiently combined with Human Intelligence, can reliably estimate age of social network users on a large scale. This is true because we have at hand a great deal of data such as:
- The face of the user on the profile picture and other images on the personal feed
- A short textual biography
- The list of friends and followers
The methodology
While data points such as consumers’ face or biography can be inconclusive to estimate age if taken on their own, we can make more reliable predictions if we use all the information available at the same time. In the picture below we describe a typical methodology involving 4 steps performed one after the other.
Step 1 — Biography analysis : social media users frequently disclose their birthday or their age in their biography. In this step, simple Natural Language Processing detects age related patterns and it provides a first prediction of age.
Step 2 — Visual recognition: profile pictures are scanned. Visual Recognition can predict how old you are by looking at your facial features. This step is useful for discarding False Positives resulting from Biographical analysis.
Step 3 — Social graph verification: your friends’ ages are used to predict yours. This step further eliminates spurious results.
Step 4 — Manual Curation: an analyst manually double checks the results filtered so far by looking at all the data at the same time. Error is further reduced.
An example : teens in Asia Pacific
In a recent research study about teens’ behaviors and attitudes towards skin care, we created a panel of more than 4,000 teens in 4 different countries.
Using the methodology above we reliably estimated their age and gender. In the graph below we display the panel age distribution and the geographical location.
The performances were benchmarked against a sample of 250 users.
Age Estimation accuracy : 95.2%
Gender Estimation accuracy : 98.7%
Conclusion
Reliable estimation of consumer age can be achieved with a smart combination of different technologies. This opens new perspectives for Market Research practitioners willing to hyper-target consumers for ad hoc briefs.
If you are interested in knowing more, download the full report and don’t hesitate to request a demo.
