Researchers develop algorithm capable of analysing and ‘tasting’ wine
Researchers from DTU, the University of Copenhagen and Caltech have shown that you can add a new, untraditional parameter to the algorithms that makes it easier to find a precise match to an individual’s taste buds.
“We have demonstrated here that by feeding an algorithm with data consisting of people’s taste impressions, the algorithm can make more accurate predictions of what kind of wine we each prefer,” says Thoranna Bender, who conducted the study under the auspices of the Pioneer Centre for AI at the University of Copenhagen, and who is a Master’s student at DTU.
The researchers held wine tastings, where a total of 256 participants were asked to arrange mugs of different wines on a piece of A3 paper based on which wines they think are most similar in taste. The greater the distance between the mugs, the greater the diversity in taste. It is a method that is widely used in consumer tests. The researchers then digitised the points on the sheets of paper by photographing them.
The data collected from the wine tastings has then been combined with hundreds of thousands of wine labels and reviews provided to the researchers by the company Vivino. Next, they have developed an algorithm based on the huge data set.
“This dimension of taste we have created in the model gives us information about which wines are similar in taste and which are not. So, for example, I can stand with a bottle of wine, which is my favourite wine, and say: I want to know which wine is most similar to it in taste – or both in taste and price,” says Thoranna Bender.
Professor and co-author Serge Belongie from the Department of Computer Science, who is also head of the Pioneer Centre for AI at the University of Copenhagen, adds: “We can see that when the algorithm combines data from wine labels and reviews with data from wine tastings, it makes more accurate predictions of people’s wine preferences than when it only uses the traditional types of data in the form of images and text. So teaching machines to use human sensory experiences results in better algorithms that benefit the user.”
According to Serge Belongie, there is a growing trend in machine learning of using so-called multimodal data, which usually consists of a combination of images, text and sound. But using taste or other sensory inputs as data sources is completely new. And it has great potential – eg in the food area, the professor believes: “Understanding taste is a central part of food science and essential for achieving healthy and sustainable food production. But the use of AI in this context is completely at the infant stage. This project shows the power of using human-based inputs in artificial intelligence, and I predict that the results will spur more research at the intersection of food science and AI,” says Serge Belongie.
The method that the researchers have demonstrated here can also easily be transferred to other types of food and drink, Thoranna Bender points out: “We have chosen wine as a case, but the same method can just as well be applied to beer and coffee. The approach can, for example, be used to recommend products and perhaps also food recipes to people. And if we know the taste similarities in food better, we can also use it in the healthcare sector to put together meals that are both right for the taste and nutrition of patients. Maybe it can even be used to develop food products tailored to different taste profiles,” says Thoranna Bender.
The researchers have published their data on an open server for free use.
“We hope that someone out there will want to build on top of our data. I’ve already had requests from people who have additional data that they want to add to our dataset. And I think that’s really cool,” Thoranna Bender concludes.