Choosing the perfect bottle of wine can be daunting, especially for those not well-versed in the intricacies of wine labels. The dilemma of what a particular wine tastes like and which one matches personal preferences can be overwhelming. In response to this challenge, wine apps like Vivino, Hello Vino, and Wine Searcher, powered by artificial intelligence (AI) algorithms, have emerged to assist consumers in making informed choices.
Now, researchers from the Technical University of Denmark (DTU), the University of Copenhagen, and Caltech have elevated this concept by incorporating a new parameter into the algorithms – people’s flavor impressions.
Thoranna Bender, a graduate student at DTU, led the study under the Pioneer Centre for AI at the University of Copenhagen. The researchers conducted wine tastings involving 256 participants who arranged shot-sized cups of different wines based on perceived flavor similarities. The resulting data, captured by photographing the placed cups on A3 paper, was combined with extensive wine label and user review datasets from Vivino, a global wine app.
The researchers developed an algorithm using this combined dataset, introducing a dimension of flavor derived from human sensory experiences. This innovative approach significantly improved the algorithm’s ability to predict individual wine preferences compared to traditional methods relying solely on images and text. Serge Belongie, co-author and head of the Pioneer Centre for AI, emphasizes the value of incorporating human-based inputs in AI, stating that using taste as a data source is a groundbreaking aspect of machine learning.
The study suggests that integrating taste as a sensory input has broader implications beyond wine selection. Professor Belongie sees potential applications in the food sector, where understanding taste is crucial for healthy and sustainable food production. While the use of AI in this context is in its early stages, it holds promise. The researchers assert their method is transferable to other food and drink categories, envisioning its application in recommending products and food recipes and tailoring meals to individuals’ taste preferences and nutritional needs.
The study, available on the arXiv preprint server, is a significant step in harnessing the power of AI and human sensory experiences to refine algorithms. It has potential implications for the intersection of food science and AI.
The researchers have made their data openly accessible, inviting further collaboration and expansion of their findings. Thoranna Bender expresses enthusiasm for the potential growth of research in this area, anticipating applications in diverse fields beyond the initial focus on wine.
