A Cluster Based Assessment of Meal Similarity for Meal Recommendation
Video
Team Information
Team Members
Pooja Desai, Graduate Student, Biomedical Informatics, CUIMC
Tara Anand, Graduate Student, Biomedical Informatics, CUIMC
Faculty Advisor: Lena Mamykina, Associate Professor of Biomedical Informatics, Vagelos College of Physicians and Surgeons
Abstract
Nutrition management is a considerable challenge in diabetes self-care, as each individual needs to identify meals that algin with their preferences and lifestyle and at the same help to improve their blood glucose management. Self-monitoring technologies can help individuals to collect data on their lifestyle behaviors and indicators of health. In this research we explore computational data analysis techniques that can help to generate recommendations for future meals that align with individuals’ eating habits (similar to an individual’s past meals) and meet their healthy eating goals. Specifically, in this preliminary work, we leverage user-entered free text meal descriptions to (1) group meals on similarity using different clustering approaches and (2) train a binary classifier to assess if a meal meets or does not meet a nutrition goal. Meaningful meal clusters were achieved through average word2vec ingredient-list embeddings and k-means clustering approaches. We were able to achieve high performance for binary classification using multi-hot encodings to represent ingredient lists and logistic regression. This work is an important first step to the delivery of just-in-time personalized meal modification feedback.
Team Lead Contact
Pooja Desai: pmd2137@cumc.columbia.edu
