=Paper=
{{Paper
|id=Vol-2903/IUI21WS-HEALTHI-11
|storemode=property
|title=Promoting Healthy Food Choices Online: A Case for Multi-List Recommender Systems
|pdfUrl=https://ceur-ws.org/Vol-2903/IUI21WS-HEALTHI-11.pdf
|volume=Vol-2903
|authors=Alain D. Starke,Christoph Trattner
|dblpUrl=https://dblp.org/rec/conf/iui/StarkeT21
}}
==Promoting Healthy Food Choices Online: A Case for Multi-List Recommender Systems==
https://ceur-ws.org/Vol-2903/IUI21WS-HEALTHI-11.pdf
Promoting Healthy Food Choices Online: A Case for
Multi-List Recommender Systems
Alain D. Starkea,b , Christoph Trattnerb
a
Marketing and Consumer Behavior Group, Wageningen University & Research, The Netherlands
b
Department of Information Science & Media Studies, University of Bergen, Norway
Abstract
Many food choices are made online. Interactive, personalized interfaces, such as recommender systems, can help users to
find new products to eat or recipes to cook, but they tend to promote unhealthy alternatives. In this position paper, we argue
that better algorithms are not the only way forward. We blend algorithms and user interface design to present a multi-list
recommender interface that presents multiple lists of personalized items in a single interface, where each list is optimized
for a specific feature (e.g., ‘less fat’). We argue how multi-list recommenders can be used to support healthier food choices.
Keywords
Recommender Systems, Food, Digital Health, Interfaces, Algorithms
1. Introduction content is recommended (cf. [9]). We propose a multi-list
recommender interface to support healthy food choices,
Recipe websites have become increasingly popular. This based on the UI of Netflix [10]. This interface combines
has spurred the development of food recommender sys- different ‘single lists’ of algorithms that use a specific op-
tems, which help users to navigate the thousands of timization (e.g., ‘Drama movies’ or ‘Because you watched
recipes found online, by presenting personalized content Frozen’) to come up with a new subset of similar recom-
[1]. However, online users struggle to identify healthy mendations, which are combined into a comprehensive
recipes, due to a lack of knowledge and misleading cues, multi-list UI. We describe how this can be applied in the
and because popular recipes tend to be unhealthy [1]. food domain to promote healthier alternatives, taking as
An increasing number of algorithmic approaches to a starting point an interface that suggests alternatives
personalization have surfaced [2]. For example, recom- for a recipe that a user had searched for, either through
mender systems present food (e.g., meal plans) that is a look-up task or implicitly in an exploratory search.
similar to what a user liked in the past [1]. Yet, there has
been little attention for health [1, 3]. It has been shown
that computing similarity between recipes (cf. [2]) is 2. Case: Multi-list Recommenders
not enough to shift preferences towards health-related
behaviors [3, 4]. In fact, it is suggested that suitable rec- Most recommender systems optimize their content to be
ommendations for one’s current preferences could even similar towards a user’s past preferences. Algorithmi-
be counterproductive [5, 6, 7], especially if one’s current cally speaking, this leads to “more like this” recommen-
lifestyle is rather unhealthy [3]. dations [1]. For example, in the food domain, if a user has
Most users stick to familiar recipes [8]. For example, in bookmarked several recipes that contain potatoes, then
the context of sustainability, users might swap their ham- a recommender system will present more potato-based
burger’s beef-based patty for a plant-based one, relying recipes. The downside of this approach is that if a user
on familiar substitutes. Although this is consistent with is currently unhealthy, she is reinforced into her current
recommendation strategies on recipe websites, which preferences through more unhealthy content [3, 4].
typically present a ‘more like this’ set of recommenda- A few solutions have been proposed to alleviate this.
tions alongside each recipe [2], such similar recommenda- For example, imposing health constraints on the algo-
tions are unlikely to produce healthy recommendations. rithm might boost the healthiness of chosen items [11],
In this position paper, we argue that personalization but this might leave users dissatisfied if they have no
approaches should go beyond only changing what is rec- healthy eating goals. Moreover, one could show a much
ommended, by also focusing on the decision context: how larger list of recommendations, but this will lead to a
sharp increase in choice difficulty, or choice overload,
Joint Proceedings of the ACM IUI 2021 Workshops, April 13–17, 2021, when not explained properly to a user [12].
College Station, USA We argue that a multi-list recommender system can
" alain.starke@wur.nl (A. D. Starke); christoph.trattner@uib.no
overcome algorithmic biases towards unhealthy foods
(C. Trattner)
© 2021 for this paper by its authors. Use permitted under Creative Commons and mitigate choice overload. It comprises novel algo-
License Attribution 4.0 International (CC BY 4.0).
CEUR
CEUR Workshop Proceedings (CEUR-WS.org)
Workshop
Proceedings
http://ceur-ws.org
ISSN 1613-0073
rithmic and interface components. In terms of content,
a multi-list recommender uses multiple algorithms that
each retrieve similar content, but differ in terms of what
specific attributes are optimized for. For example, simi-
larity could be computed across multiple attributes (e.g.,
ingredients, nutrients) to optimize utility (cf. [13]), of
which the importance weights can be adapted in line
with a user’s goals (e.g., putting more weight on calorie
content for users who wish to lose weight). In terms of
the interface, the most important contribution is to high-
light the differences between the different ‘single lists’, by
explaining them to the user. For example, by highlighting
the similarity of new recommendations with items liked
in the past [10] or by emphasizing health benefits in a
food recommender system [6].
A recommender study on computers shows how this
can be achieved, by using a ‘critiquing’ approach [14].
The recommender system would present an initial item
that fits a user’s past preferences, for which alternatives
are presented underneath it across various attribute cat-
egories, such as ‘Cheaper and Heavier’ or ‘Higher Pro-
cessor Speed, but More Expensive’. This way, each ad- Figure 1: Example of a multi-list food recommender system.
ditional attribute category forms a new list of similar Depicted at the top is a recipe that a user may have searched
recommendations, which are diversified based on differ- for (i.e., reference recipe). Based on that recipe, the recom-
ent attributes, as well as explained clearly to a user. mender presents three (or more) ‘single lists’ of personalized
recipes that optimize for a specific attribute: ‘less fat’ and
An example of a multi-list recommender interface is
‘more protein’. Depicted here are only recipe photos, but also
depicted in Figure 1. A reference recipe is depicted at
more information could be shown, e.g., names, ingredients.
the top, which a user may have searched for earlier. To
explore alternatives, similar item recommendations are and user evaluation, for it is currently unclear to what
presented underneath it, in separate ‘single lists’, where extent they are effective [10]. For example, does a multi-
each list is optimized towards a certain attribute. In our list representation reduce choice overload? Second, we
example, the first row optimizes for similarity with the propose to investigate what types of explanatory labels
reference recipe, explained as “Recipes that contain sim- in a multi-list representation are the most persuasive to
ilar ingredients”. This could be achieved by using sim- shift a user’s food preferences. For example, should they
ilarity functions and metrics, such as cosine similarity highlight health prevention (‘Similar, but less fat’), or
and RMSE [2]. In contrast, the second row focuses on health promotion attributes (‘Similar, but more fiber’)?
‘Similar recipes with less fat’, which could be achieved Another interesting avenue of research is the use of vi-
by either putting more attribute weight on a recipe’s fat sual cues, which can affect consumer preferences for food
content or determining the top-100 similar recipes and selection [15, 16]. For one, product packaging color may
subsequently re-ranking that list on fat content. This way, be adapted to invoke certain emotions in supermarkets
multi-list recommender interfaces can be used to support [16]. In an online context, recipe websites could exploit
a variety of user goals, recommending recipes that people findings that recipes tend to be rated more favorably if
like, yet supporting variations that some users might find they are accompanied by visually attractive photos [15].
particularly interesting. Above all, we argue to evaluate multi-list recommender
systems through a user-centric approach (cf. [17]). Not
2.1. Directions for Future Research only should be examined what recipe is chosen, but
also how users perceive and evaluate a multi-list rec-
The effectiveness of multi-list recommenders to support
ommender interface, for instance, compared to a single
changes in user preferences and behavior has yet to be
list approach. An important measure would be the per-
tested. The food domain is an excellent starting point, for
ceived choice difficulty (cf. [12]), since it is currently
multi-list interfaces provide algorithmic diversity that is
unclear whether presenting many sub-lists (5+) is pos-
needed to improve current unhealthy approaches in food
sible without affecting user satisfaction. Moreover, it
recommenders [3, 4]. We propose two lines of research
would be interesting to examine which types of ‘single
to examine visual UI design in recommender systems.
lists’ are the most effective in supporting health food
First, different list representations (i.e., single list vs multi-
choices, which could be a interdisciplinary field of study
lists) should be compared in terms of choice behavior
between computer science and nutrition science. In do- of the 28th ACM Conference on User Modeling,
ing so, it would be important to striking the right balance Adaptation and Personalization, 2020, pp. 333–337.
between algorithmic accuracy (i.e., reducing RMSE) and [5] M. D. Ekstrand, M. C. Willemsen, Behaviorism
interface design (e.g., nudges). is not enough: better recommendations through
listening to users, in: Proceedings of the 10th ACM
Conference on Recommender Systems, ACM, 2016,
3. Conclusion pp. 221–224.
[6] C. Musto, C. Trattner, A. Starke, G. Semeraro, Ex-
We have proposed how healthy food choices could be
ploring the effects of natural language justifications
supported by combining adaptations in the presented
on food recommender systems, in: Proceedings of
content (i.e., algorithms) and the decision context (i.e.,
the 29th ACM Conference on User Modeling, Adap-
the interface), in a new approach. We have argued how
tation and Personalization, 2021.
to enable users to find and to select healthy content in a
[7] A. D. Starke, M. C. Willemsen, C. Snijders, With a
recommender system. Fundamental to our approach is
little help from my peers: Depicting social norms
that users are still given the freedom to choose what they
in a recommender interface to promote energy con-
want, in line with research on nudging [9], but that the
servation, in: Proceedings of the 25th International
use of the interface would not trigger choice overload or
Conference on Intelligent User Interfaces, IUI ’20,
increase choice difficulty [12].
2020, p. 568–578.
For future research, we stress that interfaces and algo-
[8] Y. M. Asano, G. Biermann, Rising adoption and
rithms are not two mutually exclusive categories of re-
retention of meat-free diets in online recipe data,
search. Our multi-list recommender systems case nicely
Nature Sustainability 2 (2019) 621–627.
illustrates how ‘similar content’ and ‘healthy content’
[9] R. H. Thaler, C. R. Sunstein, Nudge: Improving de-
can go hand in hand, by pointing out what each single
cisions about health, wealth, and happiness, 2009.
list of recommendations represents. We expect that diver-
[10] C. A. Gomez-Uribe, N. Hunt, The netflix recom-
sifying the different types of recommendations presented,
mender system: Algorithms, business value, and
rather than only focusing on algorithmic optimization
innovation, ACM Transactions on Management
will be more effective in supporting healthy eating habits.
Information Systems (TMIS) 6 (2015) 1–19.
For instance, should these lists always be fully person-
[11] C. Trattner, D. Elsweiler, Investigating the healthi-
alized, or can they be less personalized in terms of past
ness of internet-sourced recipes: implications for
preferences and optimized to a user’s eating goals?
meal planning and recommender systems, in:
Proc. of WWW ’17, 2017, pp. 489–498.
4. Acknowledgments [12] D. Bollen, B. P. Knijnenburg, M. C. Willemsen,
M. Graus, Understanding choice overload in rec-
This work is in part funded by MediaFutures partners, ommender systems, in: Proceedings of the fourth
the Research Council of Norway (grant number 309339), ACM conference on Recommender systems, ACM,
and the Niels Stensen Fellowship. 2010, pp. 63–70.
[13] B. P. Knijnenburg, M. C. Willemsen, Understanding
the effect of adaptive preference elicitation meth-
References ods on user satisfaction of a recommender system,
in: Proceedings of the third ACM conference on
[1] C. Trattner, D. Elsweiler, Food recommender
Recommender systems, 2009, pp. 381–384.
systems: Important contributions, challenges
[14] P. Pu, L. Chen, Trust-inspiring explanation inter-
and future research directions, arXiv preprint
faces for recommender systems, Knowledge-Based
arXiv:1711.02760 (2017).
Systems 20 (2007) 542–556.
[2] C. Trattner, D. Jannach, Learning to recommend
[15] A. Starke, M. Willemsen, C. Trattner, Nudging
similar items from human judgments, User Model-
healthy choices in food search through visual at-
ing and User-Adapted Interaction (2019) 1–49.
tractiveness, Frontiers in Artificial Intelligence.
[3] A. Starke, Recsys challenges in achieving sustain-
Preprint (2020).
able eating habits, in: HealthRecSys’19: Proceed-
[16] I. Vermeir, G. Roose, Visual design cues impacting
ings of the 4th Workshop on Health Recommender
food choice: A review and future research agenda,
Systems, ACM, 2019, pp. 29–30.
Foods 9 (2020) 1495.
[4] C. Musto, C. Trattner, A. Starke, G. Semeraro, To-
[17] B. P. Knijnenburg, M. C. Willemsen, Evaluating
wards a knowledge-aware food recommender sys-
recommender systems with user experiments, in:
tem exploiting holistic user models, in: Proceedings
Recommender Systems Handbook, Springer, 2015,
pp. 309–352.