La Trobe University researchers develop technology to support recovering alcoholics
La Trobe University researchers have developed technology which quantifies alcohol exposure in electronic images, potentially allowing parents, or people recovering from alcoholism, to be better aware of the number of alcohol images they are exposed to via social media.
Alcohol in media has been demonstrated to increase alcohol craving, impulsive decision-making and hazardous drinking, but there is little known about how often we see alcohol images on social media.
Led by PhD student Abraham Albert Bonela, who was supervised by Professors Emmanuel Kuntsche and Zhen He, from La Trobe University, the research was published in Alcoholism: Clinical and Experimental Research.
They describe the development of an improved version of their Alcoholic Beverage Identification Deep Learning Algorithm (ABIDLA), called ABIDLA2.
Professor Kuntsche said exposure to digital alcohol marketing, in both adolescents and adults, is known to be associated with drinking initiation amongst non-drinkers, increased levels of consumption among drinkers, and binge or hazardous drinking behaviour.
“A recent study found that 56 per cent of young adults between the age of 18 and 25 report alcohol use, and 37 per cent report binge drinking and are frequently exposed to alcohol images on social networking sites,” Professor Kuntsche said.
“Because of its impact there needs to be a way to monitor and control just the extent to which we are exposed to alcoholic images.”
The ABIDLA2 algorithm was trained on 191,286 images downloaded from Google and Bing image search results. In one task, ABIDLA2 automatically identified images as containing one of eight beverage categories (beer/cider cup, beer/cider bottle, beer/cider can, wine, champagne, cocktails, whiskey/cognac/brandy, other images).
In the second task, ABIDLA2 made a binary classification between images containing an ‘alcoholic beverage’ or ‘other’. The algorithm had an accuracy of 77.0 per cent for the first task and 87.7 per cent for the second.
The algorithm was most accurate in identifying: Whiskey/Cognac/Brandy (88.1 per cent); Beer/Cider Can (80.5 per cent); Beer/Cider Bottle (78.3 per cent) and Wine (77.8 per cent). According to the researchers, even the identification of the least accurate beverage category (Champagne, 64.5 per cent) was more than five times higher than random chance.
Mr Bonela said that, because the use of social media is growing exponentially, with 424 million new users in the 12 months to January 2022, it’s important to have a way to monitor and exclude references to alcohol on social media platforms.
“There is the possibility to integrate ABIDLA2 into a web browser (as a plugin) or a mobile application to quantify or limit exposure to alcohol exposure depending on individuals’ preferences,” Mr Bonela said.
“For example, people in alcohol-related rehabilitation or recovery may want to limit the alcohol exposure to reduce craving to a minimum, whereas others may want to decrease the amount of alcohol advertisements. Such a plugin or application may also be useful for parents who wish to restrict alcohol-related content for their children when browsing online.”