AI for behaviour understanding
Billions of people are daily using social networks to connect, talk, and share their stories with the public or with their connected friends
AI for behaviour understanding
As such, to some extent, social media reflects an individual’s health status, both mental and physical, via mood, thinking, activities and communication. This is a massive scale and real-time data, valuable to Online Human Behaviour Analytics (OHBA).
The primary goal of this work is to utilise social media as a means for behaviour analytics, gaining new understandings of new forms of behaviours emerging from online settings. Specifically, we aim at:
- Understanding individual behaviours in online settings
- Understanding crowd/community behaviours in online settings
Our key focus is on:
Individual behaviours
Individual behaviours are derived from each individual when then participate online. We build machine learning and data mining models to capture the digital footprints or breadcrumbs left by the individuals. Relevant projects include a three-year NHMRC Project Grant APP1165233, 2019–2021: “Using social media data to identify markers of depression risk among individuals: A longitudinal cohort study”. This study proposes that social media data, that which is generated naturally and in the daily course of people’s lives, can be used to identify individuals who are at risk of depression. This study will correlate individuals’ social media data with their depression levels to identify the social media ‘markers’ of depression. This has the potential to significantly improve our ability to identify and treat those with depression.
Crowd/community behaviours
Community behaviours can be derived from geocoded social media data, which can provide an alternative reflection of local health trends. In this line of research, we build scalable machine learning algorithms to deal with large scale datasets generated by people in a community, either a county or state. Tweeting behaviours, as well as tweet content, have been found to be strong predictors of county-level health indices. Similarly, Internet searching data help estimate the prevalence of non-communicable diseases like stroke, heart disease or cancer.