A new tutorial led by Dr Benjamin Riordan, (Research Fellow, Centre for Alcohol Policy Research) and Dr Joshua Millward (Lecturer, Department of Computer Sciences and Information Technology), is helping researchers make the most of AI tools to analyse online imagery.
“Thanks to high-quality smartphone cameras and social media, people are sharing more online imagery than ever before, creating a rich source of data,” Dr Riordan says.
“This has the potential to provide unprecedented insights into daily life and can be used to help answer research questions. For example, at the Centre for Alcohol Policy Research, we are interested in how often alcohol appears in online imagery.”
Until now, it was impossible to manually analyse such large volumes of visual data.
To address this, Dr Riordan and the team turned to a method known as zero-shot learning, where a pre-trained AI model is used to recognise and categorise objects without needing to be trained on a specific dataset.
The approach was successful, enabling the team to detect alcohol-related imagery at a large scale. The team have now published a practical guide to help other researchers apply the same techniques to their work.
“Our paper provides a step-by-step guide for social science researchers who want to analyse large image datasets using zero-shot learning,” Dr Riordan says. “It doesn't require any technical expertise and the code can be run without downloading new software.”
"It is exciting to be able to provide a practical tutorial that helps researchers without a computer science background use zero-shot learning in their work,” added Dr Millward.
“It opens new possibilities for analysing large-scale data that would be impossible to process manually. Importantly, the tutorial also covers key evaluation considerations, so researchers understand what to expect from these models when applying them in real-world studies.”
Dr Riordan says AI has enormous potential in the analysis of visual data, and he hopes the guide will empower social scientists to use it more confidently.
“Our next step will be to continue applying zero-shot machine learning to our alcohol image research,” he says. “We plan to analyse social media posts and popular films to better understand how often alcohol appears across different forms of media.”
