Algorithms are widely used to personalize the online experience. While recommending movies or products is relatively simple, suggesting foods or recipes remains a more complex exercise. The popularization of grocery delivery services and recipe sites have spurred developments in this area. They are largely based on algorithmic recommendation systems.
In this article we will deal with
- the use of these algorithms to facilitate food choices
- the trade-off between the “popularity” of a food and health
- the role that nudges can play in this area (and if they really work)
- specific challenges that still need to be explored
- This article is brought to you by Alain Starke, post-doctoral fellow specializing in algorithmic recommendation applied to food at the University of Wageningen, the Netherlands, and associate professor in recommender systems at the University of Bergen, Norway.
Table of Contents
- Recommendation algorithms applied to food
- The difficult balance between healthy eating and favorite foods?
- Applying Nudges to Food Choices
- Visual attractiveness, another form of nudge
- Do nudges really work?
- Nudging, algorithmic recommendations… what does the future hold for us?
The AllRecipes.com website hosts more recipes than you can cook in a lifetime. Every day, its users add almost 200 recipes to its already huge database. To help users find their way around, filtering the information is essential. How do I determine which type of ebook writing services to show to which user?
Recommendation algorithms applied to food
This is where recommendation algorithms come in. They allow personalization based on user data.
When searching for a recipe, these algorithms help find the most relevant content. They can show recipes that contain sweet potatoes if a user types in that search term. In addition, the recommendation algorithms take into account the user’s history on the website. If a user has previously searched for and liked other recipes that contain specific ingredients (egg, cauliflower), recipes that contain both sweet potatoes and cauliflower will be preferred.
Another approach is particularly useful when the user is not yet known or does not have a significant search history. It consists of prioritizing the most popular sweet potato recipes. We then show the recipes which have received the most positive evaluations and/or which have accumulated the most page views.
If more is known about the users and they are active on the site, collaborative methods can be used. For example, users who have liked similar recipes can be tapped to offer suggestions for relevant content.
Applying Nudges to Food Choices
Recommendation algorithms do not work in isolation. Influencing user behaviors also depends on how recommendations are displayed.
Algorithms determine what is shown. Interfaces determine how recommendations are displayed.
Although the term “nudging” is used indiscriminately in many publications, it refers to modifications that cause predictable changes in users. In general, nudges exploit human biases.
The most classic example of food nudge is to place the products you want to “push” within sight in a supermarket.
Recommendation algorithms and the nudging technique can be combined. For example, the recommendations generated can be explained to users. Netflix’s algorithm does this by suggesting a movie based on your viewing history.
For food recommender systems, a good example is a study by Catalo Must et al. in 2021. The study aimed to determine whether food recommendation systems could justify to users why a healthy option might be favorable. The explanation was based on the nutrients and food characteristics of the proposed recipe. Such justifications make recipes more appealing and could help overcome popularity biases.
To achieve this result, recommendation algorithms can rely on natural language processing (NLP) to automate this process. The explanations or justifications can be considered as an additional “nudge” which is superimposed on the recommendations of the algorithm.
Visual attractiveness, another form of nudge
A more “back door” way of enticing people to eat would be to make specific foods or recipes more appealing.
Pictures shown next to recipes have been shown to predict whether people enjoy trying those recipes. This method can be used to overcome food popularity bias, as demonstrated in this 2020 study. It is based on the following mechanism:
- make some recipe photos more appealing
- Deteriorate the visual appearance of others.
In this study, half of the users saw attractive images accompanied by healthier recipes, while the other half saw the original content. This “visual nudging” mechanism helps to mitigate the popularity bias, since the healthy nature of the chosen recipes increased by one point on a 9-point scale (see results above). It goes without saying that such a strategy raises ethical objections, since users are probably not aware of being influenced.
Do nudges really work?
The actual effectiveness of nudging has recently been the subject of discussion in psychology and marketing. Researchers have carried out meta-analyses. They make it possible to examine the effectiveness of nudging across a large set of studies. Although this has led some researchers to conclude that nudging might not work at all, the reality seems to be that effectiveness is highly dependent on the domain and type of nudge involved.
When it comes to food and healthy eating, previous analyzes have already shown that the specific type of nudge plays an important role in memoir ghostwriters. A 2020 study shows that behavior-oriented nudges work better than cognitive nudges.
For example, rearranging the order of shelves in a supermarket may be more effective in affecting shopping behavior than labeling a set of products with a high Nutria-score. It seems that many of our food choices are rather automatic and are therefore heavily influenced by how they are presented to the user. If specific options are not easily accessible, they may suffer.
Nudging, algorithmic recommendations.
What does the future hold for us?
The scope of most research on recommender systems is short-term or limited. Often different interfaces are compared in a kind of A/B test and then an analysis is performed to understand how the user’s decision-making is affected. It usually stops there.
Possibilities exist, however, to study how behavior actually changes, in the longer term. This would involve more algorithmic adaptation, in the sense that users might have a specific dietary goal and that might change over time. These recommendation algorithms should therefore “evolve” at the same time as the user’s preferences, so as to be able to accompany the “customer journey”.