A Hybrid Recommender System for Diet Improvement Based on Health and Taste

Abstract

Recommender systems are used everywhere today, such as for online shopping or Netflix videos. The use of these systems gives us the ability to predict the taste of food effectively. In addition, other food concerns, such as food safety and nutritional value have become more important than before. With these thoughts, a novel diet improvement system is proposed in this research, which is based on a recommender system using hybrid matrix decomposition and K-nearest neighborhood algorithms to help people meet their health goals. There are many existing programs to recommend a healthy diet. Unfortunately, most systems recommend foods that are radically different from a person’s typical diet, making it difficult to adjust to and likely that the person will reject the diet and not meet their health goals. The results of an additional psychological study of induced motivation are incorporated into the system to generate a model enabling each user to gradually accept healthier foods into his/her diet. The need to minimize error in predicting taste requires the evaluation of a combination of different algorithms. In addition, it also requires optimization of different parameters of the algorithms to determine the most efficient and accurate way to recommend diet changes. The system contributes a new application of the hybrid optimized recommender system on gradually diet improvement in order to satisfy user’s health goals.

To view full PDF

Personal Project @ Lyndon Institute

Github Source Code. MIT License, Copyright © 2017 Yiren Qu