The third speaker in this WebSci 2016 session is Dongwon Lee, whose interest is in user activity patterns on Instagram. The major finding of this study is that teens and adults exhibit different activity patterns, but just as important a contribution here is the methodological contribution to the study of Instagram that this paper makes.
The project estimated the age of Instagram users based on profile descriptions, photos, tags, and account names, and deployed learned models to identify different user behaviours. Part of the underlying interest here is in understanding differences between teen and adult user behaviour, which could be used to alert teens to their behavioural patterns, and/or report irregularities back to parents or platform providers. Teens and adults are defined here as 13-19 and 30-39 age ranges, in order to have a clear separation between these groups; this may be a somewhat blunt distinction, however.
This proceeded by creating two independent datasets, using profile and tag data from Instagram. From these data the project extracted content-, interaction-, and relation-based patterns; user photos were also examined using facial recognition software that detects age and gender. Inactive accounts were excluded from this analysis.
Some of the key patterns that emerged from this were that adults exhibited more location diversity, but that teens were more active in engaging with other users, for instance. In this, tagging data appeared more useful in identifying the likely age of users than profile data. The overall accuracy in classifying users by age was 82%.