Abstract
Objectives
This study presents the design and implementation of Coco: The Conversational Calorie Counter, a spoken food logging application for iOS. The aim of this work is to reduce the burden on individuals wanting to monitor their food to support healthy eating, as well as individuals with obesity tracking food intake to manage their weight.
Methods
We built a server to predict the best matching foods in the most recent USDA database, given a user's natural language meal description. For each logged food, the user selects the correct USDA food and portion sizes from a list of ranked options provided by our server. We constructed convolutional neural network models to predict matching USDA foods automatically. We launched an iOS prototype of our system for 14 participants in a pilot study to log their meals for five days.
Results
Participants were an average of 23.2 ± 2.60 years, 92.3% female, and 14.3% actively trying to lose weight. The results of the patient satisfaction survey administered at the end of study indicated that, on a scale from 1–5 (1 is best, and 5 worst), Coco was rated 1.86 ± 0.53 for perceived accuracy of nutrition facts, 1.79 ± 0.70 for personalization, and 1.43 ± 0.65 for appealing interface design. Compared to existing food logging applications, on a scale of 1–3 (1 is better, 2 the same, and 3 worse), Coco was rated better or the same as existing methods on 8 out of 9 questions, with a score of 1.67 ± 1.16 for difficulty of food logging, 1.33 ± 0.58 for personalization, and 1.67 ± 0.58 for how fun it was. 14.3% of participants said they would definitely continue using Coco, and 28.6% said they would probably continue using the app.
Conclusions
This study demonstrates the acceptability of logging food intake with spoken natural language, with the potential to benefit individuals who find existing methods of tracking dietary intake too tedious and/or time-consuming for sustainable use.
Funding Sources
This research was sponsored by the NIH (grant # R21HL118347), the USDA under agreement no. 58-1950-4-003 with Tufts University, a grant from Quanta Computing, Inc., and by the DoD through the National Defense Science Engineering Graduate Fellowship (NDSEG) Program.