Configuration and evaluation of a constrained nutrition dialogue system

Abstract

Logging daily food intake and nutritional information is a proven way to lose weight. However, research shows that existing approaches for the prevention and treatment of excess weight gain are ineffective, burdensome, and often times, inaccurate. Thus, we have developed a natural language processing based nutrition application that translates spoken food diaries into nutritional database entries. These food diaries are compiled from a comprehensive list of specific food items from various databases, including the United States Department of Agriculture (USDA), that sum to more than 10,000 food items. In this thesis, we describe the main components of the current nutrition system, the overall ow of the system architecture, and modifications of the previous system. We also discuss the creation of a constrained nutrition system specifically designed for user testing of 100 specific food items and the user studies that will assess the accuracy and eciency of the application. We include an analysis of the modifications made for the user study, such as an alternate method for storing images, and an evaluation of the additional features added, such as the ability to save audio recordings. The work described in this thesis is part of an ongoing data collection, language understanding experiments, and updated user interface design conducted as part of a larger effort to create a nutrition dialogue system that can support the larger database of food items.