Your Brain on YouTube: Predicting Clicks and Watch Times with Neural Data
Post by Anastasia Sares
What's the science?
With the advent of the internet and social media, many barriers to communication are broken down. This means that we can access all kinds of content almost instantaneously; however, there is more information than we can ever hope to digest. Each person must choose how they will spend their time and which voices they will listen to. The competition is fierce: from Instagram influencers to national news to companies running ads. This is known as an attention market. Many people would like to predict how users will engage with different types of content, and some are turning to the brain for clues. This week in Proceedings of the National Academy of Sciences (PNAS), Tong and colleagues used functional MRI (fMRI) to examine Youtube video engagement.
How did they do it?
The authors used videos from the Youtube channels National Geographic, Animal Planet, and Discovery, whose usage data was publicly available. They analyzed the metadata from these videos (number of views, average watch time, etc.) and they also had a group of participants rate the thumbnails of these videos for emotional content.
In a functional magnetic imaging (fMRI) task, the authors selected videos that had the most variability in engagement and emotion ratings. The goal was to have videos that generated a large range of responses. While undergoing fMRI, participants first performed a video selection task, where they viewed the thumbnails of each video and indicated whether they would be interested in watching the video. In a second task, a series of videos was played and the participants could watch as long as they liked, pressing a button if they wanted to skip the video. The goal of the analysis was to see whether brain data from a small group of individuals could predict the number of views and watch time in the general population.
What did they find?
The authors found expected responses in well-known brain areas that deal with emotion and decision-making. Activity at the onset of the video in the nucleus accumbens (an area known for dopamine signaling, learning, and motivation) was positively correlated with viewing frequency and watch time for the videos. On the other hand, a significant decrease in activity in the anterior insula (another region in the evaluation circuit), was correlated with these metrics. Activity in the medial prefrontal cortex over the whole video also predicted watch time in the fMRI sample but failed to predict online behavior. In some cases, neural activity was a better predictor of the video’s online performance than ratings from people explicitly evaluating engagement.
The first two aforementioned brain regions were probably an anticipatory emotional response, reflecting the viewer’s expectations and excitement about watching the video. The final region (medial prefrontal cortex) was involved in the ongoing evaluation of the video to see whether it lived up to these expectations. In the case of videos, the anticipatory effect seemed to be the most robust in predicting online engagement.
What's the impact?
Understanding and using neural data may help people predict what kinds of content will go viral, among other applications. Although fMRI is very expensive, some groups may find it worth the price to gain an edge in a fiercely competitive attention market. Of course, ethics are important— the way in which this information is leveraged is up to us, and it can be used to have both positive and negative social effects.
Tong et al. Brain activity forecasts video engagement in an internet attention market. Proceedings of the National Academy of Sciences (2020). Access the original scientific publication here.