Man sleeping rough outside some gates

Homeless young encouraged to seek HIV tests via social algorithms

Image credit: Dreamstime

Researchers at the University of Southern California (USC) have developed two new algorithms which aim to identify the leaders within a homeless community most likely to encourage their peers to get tested for the HIV virus.

In the US, homeless youth are significantly more likely than their peers to be HIV positive, with seven per cent of rough sleepers affected.

Traditionally, social services attempt to spread public health information within homeless communities by guessing which individuals hold most influence among their peers and asking these influencers to share information.

Researchers from the USC Center for AI for Society – which brings together researchers in engineering and in social work – have developed new machine-learning tools which, like social workers, target the most influential individuals within a community, enabling peer-to-peer interventions, rather than falling back on the less personal spreading of public health information through social media.

These tools work on the principle of influence maximisation: optimising the spread of information among people. The researchers developed two different algorithms: Hierarchical Ensembling-based Agent which pLans for Effective Reduction in HIV Spread (HEALER) and Double Oracle for Social Influence Maximisation (DOSIM)

HEALER required the participants to rate their friends in terms of closeness and willingness to confide in them about their sexuality and HIV status, while DOSIM – based on the assumption that this information is hard to collect – works even if the participants do not rate their friends at all.

The two algorithms were tested in a network of 173 homeless young people over a seven month period. While current methods used by social services distribute health information to approximately 27 per cent of the intended population, with much overlap between social circles, HEALER and DOSIM were able to spread the same information to around 70 per cent of the population - an increase of more than 150 per cent.

Current methods used by social services do not appear to succeed in convincing individuals to get tested for HIV. HEALER convinced 37 per cent of the sample population to seek out testing; DOSIM convinced 25 per cent.

“It is really exciting because this is the first time algorithms for maximisation have been deployed in the real world,” said Amulya Yadav, a PhD student who led the study.

Next, Mr Yadav and his colleagues will look into testing the flow of information in a community of 900 homeless young people.

Recent articles

Info Message

Our sites use cookies to support some functionality, and to collect anonymous user data.

Learn more about IET cookies and how to control them

Close