Digital Nudges and Food Recommender Systems: Beyond Accuracy and Autonomy

Reason for the assigned topic: This topic was chosen because the dissertation focuses on integrating nudging techniques into food recommender systems but does not explicitly address the broader methodological and ethical tensions between recommendation accuracy, user autonomy, and systemic objectives, such as health promotion.
Accuracy remains the dominant evaluation criterion in recommender systems research, yet, in domains such as food, maximizing accuracy may inadvertently reinforce unhealthy preferences rather than supporting long-term well-being.

1. Digital Nudges and Food Recommender Systems: Beyond Accuracy and Autonomy Ayoub El Majjodi

2. 2 Outline Recommender Systems 01. Recommendation Accuracy 02. Nudging and Digital Nudges 03. Autonomy and Ethical Lens 04. Conclusion 05. Outline

3. 3 Human Daily Decisions ? Every day we make Decision about what to eat, what to watch, listen, read, where to travel … 35,000

4. 4 Which Book Should I Read? Choice Overload Results of too many choices being available. It can result in decision fatigue, sticking to the default option, or even avoiding making a decision altogether. [ pudding.cool ]

5. 5 Recommender Systems Recommender systems can tackle Choice Overload Tools that support decision making process They can learn the user preferences and suggest products that can be interesting and personalized for each user A recommender system is any system that produces individualized recommendations as output or guides the user in a personalized way to interesting or useful objects in a large space of possible options.

6. 6 Recommender Systems

7. Recommendation Approaches 7 Collaborative Filtering Generates recommendations by analyzing correlations in user preferences, assuming that users with similar behaviors will favor similar items. Content-based Suggests items similar to those a user has liked before, using item features such as ingredients, text, or descriptions. Knowledge-based Recommends items based on explicit domain knowledge or rules about how item attributes meet user needs or preferences.

8. 8

9. 9 Find good Items [Herlocker et al. 2004] What is a good item ? Good for whom? for the user or for the provider ?

10. Accuray Find good Items

11. 11 Recommendation Accuracy Recommendation accuracy is defined as the fraction of the correct recommendations out the total possible recommendations. Higher accuracy leads to proactively present items that match the user profile and preferences.

12. 12 Food Recommender System A food recommender system is any system with the aim of recommending food item that match user eating preferences or characteristics, based on traditional recommendations approaches. Recipe Recommenders: cooking a meal Grocery Recommenders: Food products Restaurant Recommenders Health Recommender

13. 13 Food Recommendation Process [ Elsweiler et al. 2012] Food choice theory User Characteristic, Mood Olfactory factors, Biological, factors, Cultural factors Food Food flavor Food Texture, Food Quality Food Composition , Environment Social environment, Social context ….. Recommendation Algorithms User Interface

14. 14 Food Recommender System Select Recipe Servings Serving Size 8 301 (g) Tasty Collard Greens Goat Cheese & Spinach Turkey Burger Select Recipe Servings Serving Size 8 236.80 (g) Fiery Fish Tacos with Crunchy Corn Salsa Select Recipe Servings Serving Size 8 300 (g) Recommendation list Preference elicitation phase [Starke et al. 2022]

15. 15 Current Food Recommendations Current food recommendations are unhealthy [Trattner et al. 2017]

16. 16 Current Online Food Dataset Popular and unhealthy Only 5% are healthy [Trattner et al. 2017]

17. 17 Accuray Is not Enough

18. 18 Find Healthy Food Items: Accuray Current Food Recommender Algorithms Current Popular and Unhealthy Dataset Echo Chambers, filter Babbles Build Food Recommender System with Higher Accuracy Unhealthy Behaviors

19. 19 Find Healthy Food Items: How ? Restart internet and remove all current online dataset Filter and always give healthy recommendations to all users Ignores user preference

20. 20 Find Healthy Food Items Nudging

21. 21 Nudging and Decision Making   System 1: fast, automatic, intuitive, and emotional, operating with little to no voluntary effort or sense of conscious control . System 2: is slow, deliberate, effortful, and logical, requiring attention and conscious mental exertion for complex calculations or deep problem-solving.

22. 22 Nudging and Decision Making  

23. 23 Any subtle design intervention that predictably guide people’s behavior without restricting options or significantly changing incentives. It stees individuals towards choices by making the preferred option easier and more silent. Nudging: Overcoming Biased Decisions Libertarian paternalism: helping people make better decision but preserving freedom of choice [Thaler et al 2008]

24. 24 Nudging: Overcoming Biased Decisions Elevator Poster [ wur.nl / ] 42% more used stairs than elevator 20% less food waste Plate Size [ GreeNudge.org ] Cigarette Bin 46% less cigarette butt litter [ theisleofthanetnews .com ]

25. 25 The use of user interface design elements of software applications to guide people's choices or influence user's inputs in online decision environment. Digital Nudging [Weinmann et al 2016]

26. 26 Digital Nudging Anchoring and informational Nudges Scarcity nudge

27. 27 Nudges and Recommender Systems Recommender systems to personalize and digital nudges to support final decisions.

28. 28 Nudges and Recommender Systems ?

29. 29 Nudges and Recommender Systems After eliciting user preferences, personalized items are displayed with nutritional labels as a nudge.

30. 30 What are the key ethical considerations in designing and deploying such systems? How do food recommender systems and digital nudges influence user autonomy and decision-making? Ethical Lens

31. 31 User autonomy refers to the individual’s capacity to make informed and self-directed choices. In the context of recommender systems, autonomy is preserved when users can understand the rationale behind recommendations and reflectively decide whether to follow them. User Autonomy Informed choices Why choices are recommended [Bartmann et al 2023]

32. 32 Digital Nudges and User Autonomy Informed choices Digital Nudges Libertarian Paternalism Healthy Recipes How the choice is presented not What is presented ?

33. 33 Providing healthy and unhealthy recipes, but changing how they are presented. Support the user the ability to judge the recommendations. Digital Nudges and User Autonomy

34. 34 Digital nudges preserve user autonomy only when designed to support practical reasoning, helping users weigh differences between options (recipes) without covertly shaping what is presented. Why choices are recommended Digital Nudges and User Autonomy

35. 35 Why choices are recommended Recommenders and User Autonomy Transparency Beyond Accuray Diversity User Centric Eval Explainability

36. 36 Transparency refers to the extent to which users can understand how and why recommendations are generated, highlighting explainability, interpretability, and understandability. It is crucial for supporting trust, satisfaction, and effective decision-making. Transparency [Vorm et al. 2018]

37. 37 Transparency Guidelines Quality of The data 01. Using data from well trusted and validate sources for example Allrecipes.com Recruit users with higher approval rating

38. 38 Transparency Guidelines Options and recommended Items 02. Generate options are based on user preferences

39. 39 Transparency Guidelines User Representations 03. How user (user data) are represented in the system

40. 40 Transparency Guidelines System inner logic 03. Why items are recommended

41. 41 Transparency Guidelines System inner logic 03. Explaining the nudge

42. 42 Transparency Guidelines System inner logic 03. Explaining the nudge Explained nudge leads to more healthier choices.

43. 43 Transparency Guidelines Options and recommended Items Quality of The data User Representations 01. 02. 03. System inner logic 04. Explainability

44. 44 Diversity ensures that users are exposed to a wide range of content, enhancing their experience and preventing the narrowing of their informational and cultural horizons. Diversity [Neidhardt. 2024]

45. 45 Diversity Personalization may reinforce unhealthy preferences, while random recommendations can increase diversity and lead to healthier choices.

46. 46 Diversity To what extent should food recommender systems diversify their suggestions enough to promote healthier and more sustainable choices, but without overwhelming or alienating users? Too little diversity Reinforces habits and potentially unhealthy patterns Too much diversity May reduce satisfaction, trust, or perceived personalization. The challenge lies in finding a balance between user satisfaction , and health outcomes .

47. 47 User Centric Evaluation An approach that examines how users perceive and experience their interactions with recommender systems, going beyond accuracy metrics to consider the user's subjective experience, including perceived recommendation quality, diversity, satisfaction, and the cognitive and emotional processes involved in using the system. [ Knijnenburg . et al 2010]

48. 48 User Centric Evaluation User centric approach illustrates how different system components collectively shape users’ perceptions, experiences, and interactions with the recommender. Objective aspects User experience Perception aspects User characteristics

49. 49 In The Long Run Our Hypothesis …

50. 50 In The Long Run Integrating digital nudges into recommender systems to should include human in the loop to ensure autonomy, diversity, satisfaction, and overall well-being.

51. https://edepot.wur.nl/455083 https://www.dailymail.co.uk/sport/football/article-3220919/Cristiano-Ronaldo-Lionel-Messi-butt-new-quirky-scheme-reduce-cigarette-litter-streets-London.html https://www.sciencedirect.com/science/article/pii/S0165176513001286 https://greenudge.org/alle-prosjekter/ https://theisleofthanetnews.com/2017/08/08/stub-it-out-and-pick-your-option-with-thanets-new-cigarette-butt-voting-bins/ https://gemini.google.com https://www.behavioraleconomics.com/be-academy/courses/behavioral-science-ethics/lessons/nudge-ethics-part-1/topic/autonomy-of-the-nudged/ https://www.accio.com/c/7c68c817-7c85-4af2-baea-029b610964c5 References Slides

52. for Your Attention