Salt is consumed at too high levels in the general population, causing high blood pressure and related health problems. In this paper, we present results of ongoing research that tries to reduce salt intake via technology and in particular from an interface perspective. In detail, this paper features results of a study that examines the extent to which visual and textual explanations in a search interface can change salty food preferences. An online user study with 200 participants demonstrates that this is possible in food search results by accompanying recipes with a visual taste map that includes salt-replacer herbs and spices in the calculation of salty taste.
replacer recipes (SR recipes) we test an approach in which the visual and textual explanations of food are altered Worldwide obesity levels are increasing [1]. A main issue in a search interface. The intention of our explanations is the high prevalence of unhealthy foods available, both is to boost SR recipes by educating the user about the ofline and online [ 2, 3]. On the other hand, technology healthier nutritional content. Such an intervention is has been shown to be useful to tackle obesity. Food referred to as a “Boost”, a type of nudging that promotes recommender technology, for example, has demonstrated small changes in behavior through education [7]. the potential to change people’s eating behavior [1, 4]. Main objective: We seek to show that visual and texYet, less explored is the principle of “search”, the main tual explanations of salt content and salt replacers (SRs) in means of finding information about food on the Web. a search interface can alter salty food preferences. Focus
To contribute to this little researched area, we have fo- ing on the online recipe domain, we posit the following cused on the extent to which search interfaces can change research questions: people’s eating preferences. One recent study shows healthy recipes can be boosted by presenting attractive • RQ1: To what extent do salty food preferences food images alongside them [5], overcoming possible change due to visual and textual explanations on innate preferences for unhealthy food [3]. a recipe’s salty taste?
In this work, we focus on salty food preferences and • RQ2: To what extent do other recipe and user how to change these during the search process. The main characteristics afect salty food preferences? principle is to replace food that has a high salt content with less salty food. The food items under investigation In the remainder of this paper, we first discuss relevant are online food recipes. To support possible changes in work related to this research (cf. Section 2), positioning user preferences, we emphasize the preference of salt our approach. Then, we describe the contents of our user replacers in food through visual and textual explanations. study, including recipe dataset and our visual and textual Most people who like salty taste in food are equally satis- explanations for salty taste. Finally, we describe our main ifed if salt is reduced slightly, as long as substitute herbs results and discuss their implications. or spices are added [6]. Such salt replacers (SRs) are ingredients like oregano and garlic, which mimic the taste 2. Background sensation triggered by consuming salt. To promote salt
Joint Proceedings of the ACM IUI 2021 Workshops, April 13–17, 2021, tions of consuming food high in salt content. Secondly, College Station, USA we present studies that have examined how food prefer(eVm.aVi.lS:jaørenng)e;iAr.lbaeirng.Set@arnkoer@ceuriebs.enaorc(Ah..nSota(Ark.eB);erge); vsj010@uib.no ences can be shifted towards healthier options (i.e. digital Christoph.Trattner@uib.no (C. Trattner) nudging), by focusing on how food is presented rather CPWrEooUrckReshdoinpgs IhStpN:/c1e6u1r3-w-0s.o7r3g ©CCo2Em02mU1oRCnospLWyicreigonhsrtekfAostrthrtihboiusptpioanpPe4rr.0obIynctieetsrenaaduttiihononragsl.s(CUC(seCBpYEer4Um.0i)Rt.te-d WundSe.roCrrega)tive than changing what is presented. Thirdly, we show how this work is based on earlier research in the field, and about the suggested changes at the same time [7]. We where it makes a contribution. believe that it is possible to both make it easier for users to select healthy recipes and increase their knowledge, 2.1. Salt in Food and Salt Replacers by calculating a wholesome salty taste score based on research about SRs, as well as educating users about High salt/sodium content in food is one of the fundamen- how selecting SRs can accomplish a salty taste with less tal indirect causes of disease and death [8, 9]. Cardio- sodium. vascular diseases can be attributed to increased blood pressure, which is found to be responsible for 62% of 2.3. Diferences to Previous Research strokes and 49% of coronary hearth diseases. Starting a downward spiral, a person’s blood pressure is increased The key diferences between this research and the works by high salt intake [8], which can often be attributed to mentioned above are the focus on the type of interface, high levels of consumption of preprocessed food [1]. the food preferences, and the type of nudges being used
The high prevalence of salt in food is also notable on- to change food preferences. line. Trattner et al. [1] show that most recipes on the With regard to user interface design, a lot of research world’s largest recipe website, Allrecipes.com, are rela- is devoted to recommender interfaces [3], while our retively unhealthy. To make matters worse, in terms of search is on food search. Another key diference is the health, most personalization algorithms tend to priori- type of preferences we consider. While previous research tize popular, yet unhealthy items in their search results, examined food preferences in general [2], we focus on which tend to surpass WHO and FSA guidelines for salt salty food preferences, as high salt intake is shown to be intake. the main cause for current cardiovascular diseases [8].
The healthiness of online recipes can be improved by Finally, compared to previous work, this study is among
swapping salt content for replacement ingredients that the first to investigate the use of taste maps and textual
mimic the taste. Salt-replacer (SR) herbs and spices can explanations to change people’s salty food preferences.
do this and have health benefits in their own right,
sometimes even lowering blood pressure [
This section presents the methods used and applied in our research. In particular, we discuss our recipe dataset sample, reveal details about the visual and textual explanations and how we developed these and, finally, describe the design of our online study and how this study was carried out.
Besides changing what recipes are recommended, one
way to promote SR recipes is to highlight them in a choice
interface. To change the choice context without altering
what is presented can be referred to as (digital)
nudging. Cadario et al. [
A specific type of nudging is called “Boosting”. “Boosts” promote small changes in behavior, by educating users
To address our research questions, we selected main
courses from a recipe database, obtained from the recipe
website Allrecipes.com and provided by [1]. The dataset
comprised a sub-sample of 1031 recipes with detailed
information about ingredients, directions and nutritional
values. To perform our research we selected two times
six recipes corresponding to two search result sets in
our prototype, see Table 1. Six of the recipes answered a
search for “chicken” and the other six answered a search
for “pork”. Each of the search result sets had half-half
recipes with and without SR ingredients. When selecting
SR recipes we looked for SRs that appear in the
literature [
Table 1 shows the recipes for the two search result sets used in the online user study. For SR recipes we calculated an SR-boosting score. For Non-SR recipes we let the SR-boosting score be similar to the sodium score.
a taste map, see Figure 1. The taste map was inspired by other visual taste descriptions in the literature [15]. Our visual taste explanation is a pentagram with one axis for each of the five basic tastes. For this study, we only calculated values for the salty taste axis. Had values for all axes been calculated, a shape would emerge to show a complete taste profile that would enable users to look for a variety of visual shapes when browsing search results or recipe presentations.
The taste map covered a scale from 0 to 1 on the salty axis, with a granularity of 0.1. We anticipated that many of our participants would expect a recipe with a mean sodium level to be displayed with 0.5, the middle value of our scale. The sodium content per serving in the larger dataset was normally distributed. The mean was 0.9 grams and the 95 percentile was 2.0 grams. We therefore made it simple by regarding the maximum sodium level as the mean times two, which is 1.8 grams. Any serving with 1.8 grams sodium or more would turn up as 1 on our scale. A mean sodium level would turn up as 0.5 on our scale. In one of the experiment conditions, the value of the taste map was boosted when shown alongside SR recipes. The taste map functioned as a visual salty taste explanation for all recipes at all times in our experiment. In the SR-boosting condition, however, the taste map had a heightened explanatory role by showing visually that SRs contribute to the perceived salty taste. Calculating the contribution of salt replacers to salty taste is a subject of food chemistry and beyond the scope of this paper. To be able to perform our experiment, we took a pragmatic approach and added 20% to the sodium score to calculate an SR-boosting score for SR recipes.
In addition to the visual explanation, we designed a textual explanation in our search prototype to boost SR recipes, as depicted on the bottom right of Figure 2. The textual explanation had two parts: First, the headline “Healthy ingredients enhancing salty taste”, followed by an explanation like “Oregano and garlic joins the salt in enriching the flavors of this recipe.”
3.3.1. Participants To have a diverse pool of participants, we used two crowdsourcing platforms. We recruited a total of 200 participants: 100 on Prolific, 100 on Amazon MTurk. We only sampled U.S. nationals, as ingredient amounts were denoted in U.S. metrics. As is common in research, we had a diferent required approval rate for the two platforms [16], 90% and 98% respectively. Prolific participants were reimbursed with 2 GBP for participation as we assumed that our study took 14 minutes, while “MTurkers” were compensated with 0.5 USD when we found out that the mean completion time was much shorter (7 minutes). The eventual sample comprised 54.8% males, with a mean age of 37.6 years ( = 14.16). 3.3.2. Prototype
before and after the experiment. More notably, it had web pages for search samples (see upper part of Figure 2). Figure 2: Top: One of the two search result sets. Bottom left A search bar prefilled with a search term was shown and right: A single participant would see this recipe in only together with six search results. To the right of each one of the left/right presentation states. Since the recipe had recipe title was a miniature taste map, so that users could salt-replacer ingredients, it could be presented in two states: quickly scan the taste of each recipe in the results. The On the left with a taste map based on sodium score and on prototype also had separate web pages to present the the right with a textual explanation and a taste map based on full recipes that appeared in the search results (see lower SR-boosting score. Non-SR recipes could only have the state part of Figure 2). Each page contained the recipe title, shown on the left-hand side. an image of the meal, ingredients, cooking directions and nutritional info, in addition to a taste map section.
Figure 2 also shows side-by-side the two ways SR recipes were presented. On the left-hand side the salty taste value in the taste map was based on sodium score. On the right-hand side, the salty taste value in the taste map was boosted visually and accompanied by a textual explanation. levels from the taste map, the ingredients, as well as the nutritional info section’s sodium value and percentage of daily recommended intake.
Baseline Condition. This was one of two main conditions in the experiment. In this condition a search for either “chicken” or “pork” was shown, the opposite term from that in the other main condition. Search results were 3.3.3. Research Design ranked according to the recipes’ sodium score (as opposed to SR-boosting score), displayed on the miniature taste Recipe search results and individual presentations. maps. The participants were then guided through the In the two following main conditions, users were first recipe presentations. Three of the six recipes contained shown a sample search with a list of six search results SR ingredients, but participants were not given any texcontaining a mixture of three Non-SR recipes, and three tual explanation about this, and there was no boosting SR recipes. The search results were ordered from high of the taste map value in search results or on the recipe to low value on the miniature taste maps. Subsequently, pages. the participants were guided through recipe presenta- SR-Boosting Condition. This was the other main tion pages for each search result in the list. On top of condition. A search for either “chicken” or “pork” was each recipe, the participants were asked: “Please rate this presented, the opposite term from that in the baseline recipe according to how it fulfills your salty food pref- condition. Search results were ranked according to the erence on a scale from 1 to 7 (1 means very poorly and recipes’ SR-boosting score as shown in Table 1. Three 7 means very well).” Participants could glean salty taste of the search results contained SR ingredients and were
Initially in the procedure, participants were asked about characteristics including cooking experience and healthy eating habits. They were then shown a static page with an example of how a search in our recipe site could look. Either the search for “chicken” or “pork” was shown together with the corresponding recipe set consisting of six search results. Participants were asked to go through each recipe, read the recipe presentation, and rate it according to how well it satisfied their salty food preference. A second search sample showed the remaining search term, either “chicken” or “pork”, and the resulting recipe set. Once more, participants were asked to go through the recipes of the search results and rate them. Finally, participants filled out a short questionnaire where they were asked about their attitude towards SRs. 3.3.4. Procedure Figure 4 shows that after initial questions the participants were randomly assigned to four branches, ending up with a questionnaire. There were two main conditions with randomized order. These were paired with the two search samples, which also had randomized order. The visual scores in the taste maps were based on sodium scores in the baseline condition and SR-boosting scores in the SR-boosting condition. The latter condition also showed a textual explanation accompanying a boosted taste map value when a recipe had SR ingredients. 3.3.5. Measures
before the experiment were used as independent variables later in the analysis. We inquired about their age and gender (Male, Female, Other). Then, about what their highest completed education was: Less than high school, High school or equivalent, Bachelor degree (e.g. BA, BSc), Master degree, Doctorate, or Prefer not to say. They were also asked about their cooking experience and eating habits on a 5-point scale from “Very low / Very unhealthy” to “Very high / Very healthy”.
In the experiment we collected user preference data used as dependent variable in the analysis. Each participant was shown a total of 12 full recipe presentations. For each recipe the user was asked to read the presentation and rate the recipe according to how it fulfilled their salty-food preference on a 7-point Likert scale from “Very poorly” to “Very well”.
The questionnaire after the experiment notably asked a control variable Likert scale question: “Research shows that many people can be just as satisfied if salt is reduced a little and garlic, oregano or rosemary added. To what extent do you think that you would be satisfied with this in the future on a scale from 1 (to no extent) to 7 (to a high extent)?”
changed due to our taste map value boost and textual explanation (RQ1). Overall, 200 participants gave 2x 600 ratings for Non-SR recipes and 2x 600 ratings for SR recipes across the two main conditions. Since the boosting-explanations were only presented alongside recipes that actually contained SR ingredients, we performed a dependent t-test that compared the 2x 600 ratings for SR recipes. The t-test showed that ratings given in the SR-boosting condition ( = 4.54, = 1.77) were significantly higher than in the baseline condition ( = 4.33, = 1.79): (1198) = 2.08, < 0.05.
It is easier to understand this result by inspecting Figure 5. While there was no diference across conditions for Non-SR ingredients, the rating given for SR recipes increased around .20 on a scale from 1 to 7. This either suggested that explanations boosted preferences for SR recipes or that they stood out from the larger list of recipes. We further examined whether other recipe features and user characteristics afected users’ food preferences with regard to salty taste. To do so, we predicted the rating given by users to the presented recipes, based on a recipe’s taste value, a user’s self-reported health with regard to her eating habits, the SR-boosting condition, cooking experience and a user’s attitude towards SR ingredients. Table 2 describes the two random efects models. Model 1 shows that recipes for which a higher taste map value is reported, whether based on sodium score or SR-boosting score, are also more likely to receive a higher rating: = 1.34, < 0.001. This confirmed our expectations that a recipe’s salty taste is an important predictor of expected food enjoyment and food preferences.
Table 2, Model 2 expands our baseline model by adding user characteristics, as well as by controlling for our SRboosting condition. We again observed a positive relation between the presented taste map value and the given rating ( = 2.68, < 0.001), further confirming our expectations. In addition, Table 2 shows that users with self-reported health were more likely to give a higher rating: = .46, < 0.01, which suggested that users who perceived themselves as having a healthy lifestyle still preferred foods that contained higher amounts of salt. However, we also observed an interaction efect between the taste map value and self-reported health, which was understood best by inspecting Figure 6. While both higher levels of self-reported health and a recipe’s taste map value positively afected the rating given, the diferences in slopes suggested that the increase was stronger for users with lower self-reported health.
Table 2 also reports on other user characteristics. We found no relation between a user’s cooking experience and the given rating ( > 0.05), nor did we observe an 2 includes more user characteristics and the efects due to the SR-boosting condition. * 5. Summary & Future Work • We have examined to what extent food preferences change due to visual and textual explanations on salty taste. We have found that SR recipes can achieve a higher satisfaction of salty food preferences when boosted in the search interface. • We have also examined to what extent other recipe and user characteristics can afect salty food preferences. Recipes’ salty taste appears to be an important predictor of expected food enjoyment, since recipes with a high displayed salty taste value received the highest ratings for fulfilling salty food preferences. • Users with a high level of self-reported health and those with a positive attitude towards SRs provided higher ratings across all recipes, indicating a higher overall expected enjoyment of food. • We have found evidence that participants with low levels of self-reported health can benefit the most from SR explanations, since they had the largest increase in given ratings.
A number of limitations have to be noted. Due to the novelty of the research domain, our findings have to be interpreted with some caution. Hence, this is one of the ifrst studies on SR ingredients and salty food preferences in a search interface. First of all, there were only three SR ingredients available in our dataset sub-sample, namely garlic, oregano and rosemary. Also, we conducted our study on a very limited number of recipes. A larger study is obviously needed that captures a bigger variety of recipes and SR ingredients.
Furthermore, we did not personalize the search results or provide results based on specific salty food preferences of users. As a first efort, we concentrated on studying the overall efect of boosting SR recipes for the participant population, regardless of initial salty food preferences of individuals. Further research is needed to also understand the efect of personalization [4].
goals/dietary restrictions. Sweetness is, for example, another interesting axis that may be worthwhile to explore further, as high sugar intake also increases health risks.
that contained such SR ingredients, we observed no inter- taste map as presented in this paper and especially user
and the Research Council of Norway (grant numbers 309339 and 310468).