Algorithms and personal taste

Determining what someone likes and satiating their everyday desires is a challenge that many content providers are now racing to tackle. From Netflix to Amazon, companies are using big data and artificial intelligence to scope out users’ tastes in order to deliver them customised recommendations that is on the button every time.

We recently undertook a project to help an online entertainment streaming service improve their on-boarding process. As a service with millions of users globally, our client wanted to identify the mental models people have of their personal preferences, how they communicate them, and the best mechanisms and tricks to learn about someone’s taste and know how to satisfy it.

Creating the dream machine

We recruited 16 diverse participants with different backgrounds and demographic profiles and ran user labs with them, using a mix of individual data collection and reflection, mixed with a group discussion.

During the lab, we gave participants several activities to make sense of their own taste for the type of content our client offers, including a collage and a timeline to map the evolution of their preferences. Next, we paired people up and asked them to explain their taste to each other for five minutes, before explaining their partner’s taste to the rest of the group.

This demonstrated the variations of language people use when expressing their taste, and how much information and nuance can be packed into five minutes, even when speaking to a stranger. We also noticed that taste is inherently difficult to verbalise, since it is wrapped with so many associated meanings, values and emotions.

Finally, we led a speculative exercise, asking participants to ‘build’ their own dream suggestion-generating machine. We asked them what questions it would ask, what they would like to tell it, and how it would work. This gave us insight into how much people would like to get from such services, and their expectations of them.

Context is important

We learned during this research that people want to be understood by content providers on a human level that makes recommendations meaningful. Often, when we first meet someone new, we reveal a bit about ourselves, often by talking about likes or dislikes. But to truly understand someone better, we need to learn about them over time: why do they watch the same movie when they’re sad, or why do they get nostalgic over a certain song? It’s a tall order to match the intricacies of human relationships with an algorithm, but we expect streaming services to keep improving their dialogues with users in this direction.