Personalizing Food Recommendations with Data Science

At Zipongo, our mission is to make healthy eating easier and more convenient. Food is one of the top drivers of chronic disease risk, so we focus on personalized and realistic ways to improve our users’ food choices.

Everyone needs to eat, but we all have different health concerns, lifestyles, habits and food preferences. These differences are shown in our users’ behaviors. For example, our male users are 20% more likely to choose recipes with the word “creamy” in them than our female users.

We also know that various diets or ways of eating work better for different people depending on health conditions, genetics and behavioral habits. This means we can’t give generic nutrition advice like “eat more kale.” In order to have an impact at the population level, we have to provide personalized and actionable recommendations for individuals.

Data Science at Zipongo

Providing the best food recommendations to our users is an exciting challenge. It’s also complex, because neither food nor people are simple. To tackle this issue, we gather information about our users and information about food, which we harmonize and feed into our personalized relevancy engine.

We gather our users’ health data from biometric screenings and onsite clinics, which are arranged through their employer or health plan. We then use proprietary questionnaires to obtain data on their nutrient intake, lifestyle habits and food preferences. We’ve optimized these questionnaires to gather the most useful information with as few questions as possible. Using the data we collect, we model complex attributes like nutrient and food intake in order to understand any gaps in the user’s diet. We use both the health data and food intake data to create targeted, personalized recommendations.

Similarly, understanding a user’s food environment is an important part of our recommendation engine. Through data gathered from our integrations with food service vendors, recipes developed in-house, grocery store discounts and local restaurant menus, Zipongo strives to understand available food options and rank their healthfulness.

We develop robust mechanisms to automatically learn as much as possible about a large collection of meals, utilizing the expertise of our in-house nutritionists and clinicians. Our algorithm can then classify new meals, ultimately extracting a set of features for each one. At this point, we can rank meals for each individual user, displaying the highest-ranked options and providing truly personalized recommendations.

The algorithm learns from feedback, including what the user cooks, what the user buys at the grocery store, which Zipongo features the user takes advantage of most often, and how the user’s food consumption and health changes over time. We can apply this algorithm to foods in a corporate cafeteria, recipes users can cook, and even restaurant dishes and delivery services to recommend the best option for a user in any situation.

We’ve seen significant health improvements so far. Among one of our larger employee populations, we found statistically significant increases in consumption of fruits and vegetables. We also saw significant changes in health outcomes like reduction of blood pressure and waist circumference. Results aren’t limited to one population; diverse users from different locations, socioeconomic status and industries have shown improvements. By enabling small, daily changes in food choices, we can significantly impact health and reduce the risk of chronic disease.

Come join Caryn and the data science team at Zipongo to make a positive impact on real users!

Leave a Comment