Source-linked AI summary
Digital Nudging with Recommender Systems: Survey and Future Directions
Mathias Jesse, Dietmar Jannach
TL;DR
Recommender systems shape digital choice environments, but their relationship with digital nudging has been studied only to a limited extent. This paper systematically reviews nudging mechanisms, organizes 87 mechanisms into a taxonomy, surveys their use in recommender systems, and identifies opportunities for integrating additional nudges. It concludes that recommender systems inherently implement several nudging mechanisms and that future systems could incorporate more.
Problem
The relationship between digital nudging and recommender systems has been insufficiently explored, although recommendations shape users’ choice environments and decisions.
Method
The paper conducts a systematic literature review, organizes 87 nudging mechanisms into a taxonomy, and surveys studies applying them in recommender systems.
Results
The review finds that recommender systems inherently implement various nudging mechanisms, while existing recommender-system research has investigated only a small part of the identified mechanisms.
Takeaways & Limitations
The paper identifies future opportunities to incorporate additional nudging mechanisms into recommender systems, including explicit nudges alongside personalized recommendations.
Takeaways & Limitations
Some nudging mechanisms lack strong theoretical or empirical backing, and terminology can create confusion.
Abstract
from arXiv · showhide
Recommender systems are nowadays a pervasive part of our online user experience, where they either serve as information filters or provide us with suggestions for additionally relevant content. These systems thereby influence which information is easily accessible to us and thus affect our decision-making processes though the automated selection and ranking of the presented content. Automated recommendations can therefore be seen as digital nudges, because they determine different aspects of the choice architecture for users. In this work, we examine the relationship between digital nudging and recommender systems, topics that so far were mostly investigated in isolation. Through a systematic literature search, we first identified 87 nudging mechanisms, which we categorize in a novel taxonomy. A subsequent analysis then shows that only a small part of these nudging mechanisms was previously investigated in the context of recommender systems. This indicates that there is a huge potential to develop future recommender systems that leverage the power of digital nudging in order to influence the decision-making of users. In this work, we therefore outline potential ways of integrating nudging mechanisms into recommender systems.
1. Introduction
The paper frames recommender systems as digital nudges because their automated selection and presentation of content shape users’ choice environments. It surveys nudging mechanisms, their use in recommender systems, and opportunities for combining personalized recommendations with explicit nudges.
- Motivation: Recommender systems influence users’ choices by selecting, ranking, and presenting accessible information while helping reduce choice overload.They also support provider goals such as increased sales or user engagement.
- Motivation: Digital nudging applies behavioral-economics nudges to online interfaces, guiding users toward desired options without restricting the choice space.Recommendations can function as nudges because their presentation implements mechanisms such as hiding options.
- Research gap: The relationship between digital nudging and recommender systems has not been explored in depth, despite recommendations’ role in helping users make decisions.The paper positions this underexplored relationship as its central research gap.
- Contributions: A systematic review collects previously identified nudging mechanisms and organizes 87 of them into a novel taxonomy with associated psychological phenomena.The paper also surveys studies implementing nudging mechanisms in recommender systems.
- Contributions: The paper highlights future ways to combine recommender systems with additional nudging mechanisms, including explicit cues alongside personalized recommendations.One example highlights healthier food choices or options that broaden musical interests while preserving ordinary recommendations.
- Caveat: Visual highlighting in recommendation lists may be a nudge, but it is not always clear whether the highlighted option is most beneficial for the user.Examples include marking discounted items or “Amazon’s Choice.”
2. Concepts, Terminology, and Methodology
The paper distinguishes nudging, digital nudging, and related interventions, while noting unresolved ethical and conceptual boundaries. It then uses systematic database searching and citation snowballing to identify research on nudges in recommender systems.
- Concepts and terminology: Nudging uses choice architecture and libertarian paternalism to influence decisions while preserving users’ freedom to choose.A choice architecture is the environment in which people make decisions, and defaults are one example of a nudge.
- Concepts and terminology: Digital nudging uses user-interface design elements to guide behavior in digital choice environments without restricting the available choices.Online decision environments are commonly represented by application interfaces.
- Ethical considerations: Digital nudges are not always used for users’ benefit, creating ethical concerns when interfaces encourage purchases of non-essential options or more products.The paper notes that ethical discussions have not reached consensus.
- Ethical considerations: The paper does not primarily focus on ethical questions and acknowledges that distinguishing nudging from persuasion can be difficult.It notes that recommender systems are often considered persuasive technology, while coercion or deception is not acceptable under a common definition of persuasive technology.
- Methodology: The literature review began with systematic searches of major electronic libraries for studies applying nudges in recommender systems, followed by citation snowballing.The search combined nudge-related and recommender-system-related terms across multiple databases.
3. Taxonomy and Catalog of Nudging Mechanisms
The paper develops a taxonomy and catalog of nudging mechanisms, refining prior categorizations to classify 87 mechanisms and distinguish mechanisms from their underlying psychological effects.
- Taxonomy: The taxonomy consistently uses “nudging mechanism” for ways of nudging and separately identifies the underlying psychological phenomena.The paper maps mechanisms to psychological phenomena in a later section.
- Taxonomy: The taxonomy has four highest-level categories: Decision Information, Decision Structure, Decision Assistance, and Social Decision Appeal.The first three categories derive from prior work, while Social Decision Appeal was added to reflect the catalog’s mechanisms.
- Taxonomy: Decision Information changes or emphasizes presented information, whereas Decision Structure alters option arrangement through mechanisms such as defaults, effort, or ordering.These categories modify information or choice structure without necessarily changing the available options.
- Catalog: Decision Assistance supports users’ goals, while Social Decision Appeal uses emotional and social implications of information, including others’ decisions.The catalog contains eight Decision Assistance mechanisms and 11 Social Decision Appeal mechanisms.
- Catalog: The second taxonomy level contains 13 subcategories, including additional categories for salience, messenger reputation, and insights from human behavior.The catalog is organized by taxonomy category, with category overviews in Figures 2–5 and detailed mechanisms in appendix tables.
- Taxonomy: 87 nudging mechanisms are organized into a novel taxonomy grounded in a systematic literature review.The taxonomy is intended to classify the variety of mechanisms identified in the literature more finely than previous proposals.
4. Underlying Psychological Phenomena
The review links nudging mechanisms to underlying psychological phenomena, identifying a broad set of effects but finding that explicit theoretical grounding is uncommon. It emphasizes that mechanisms, implementations, and user characteristics interact, so their effects require whole-system evaluation.
- Identified phenomena: Examples include anchoring and adjustment, availability, similarity, loss aversion, hyperbolic discounting, priming, and decision fatigue.The literature also reports effects such as choice aversion, confirmation bias, present bias, and reciprocity bias.
- Identified phenomena: The review identified 58 psychological phenomena mentioned across the nudging literature.These phenomena are summarized in Table 1.
- Theoretical grounding: Only 22 of 87 nudging mechanisms were explicitly linked to underlying decision-related phenomena in the reviewed literature.The review characterizes this as less than one third of the identified mechanisms.
- Theoretical grounding: Nudging mechanisms and psychological effects are sometimes treated interchangeably, creating terminological confusion and leaving some mechanisms without strong theoretical or empirical backing.The review specifically notes examples such as anchoring and adjustment or framing being used as mechanisms while referring directly to psychological phenomena.
- Interactions and evaluation: Mechanism effects depend on implementation, interactions among multiple psychological phenomena, and demographic characteristics such as age.One cited example found that older participants were less likely to be affected by nudges.
- Interactions and evaluation: Because the same mechanism can have multiple implementations and rely on several phenomena, nudging systems should be designed and evaluated as wholes.The review states that resulting effects cannot be directly connected to mechanisms without observing multiple variables together.
5. Nudging with Recommender Systems
The paper shows that recommender systems inherently implement multiple nudging mechanisms, while most identified mechanisms remain unexplored in recommender-system research. It reviews existing studies and proposes additional ways to integrate nudges into recommender systems.
- Inherent nudging mechanisms: Typical recommender systems tailor information, simplify search, structure complex choices, suggest alternatives, and shape perception through ordering, hiding, partitioning, and visual presentation.They also provide information such as titles, tags, and prices.
- Research coverage: 87 nudging mechanisms were identified, but only 18 had been studied in recommender systems; 69 remained unexplored.“Setting defaults” and “Using visuals to increase salience” were the most researched mechanisms.
- Study characteristics: The reviewed studies covered nine application domains and used surveys, interviews, laboratory studies, and field tests.Route planning and e-commerce were the most common domains, while laboratory studies and field tests were most common experimentally.
- Study characteristics: Only a few studies documented their duration, and those studies lasted one or two months.Eleven studies measured behavioral impact, while other papers addressed usability or conceptual and mathematical models.
- Existing evidence: 8 out of 11 user-study papers reported that at least one implemented nudging mechanism affected user behavior.Three studies reported no significant nudging effects.
- Limitations and future research: Long-term effects remain difficult to assess because nearly all reviewed research examined immediate effects.The paper also notes that effectiveness can depend on circumstances, application domain, or implementation.
- Existing evidence: A recommendation can influence choices indirectly even when users do not select the highlighted recommendation itself.In one study, recommending a particularly large backpack led users to choose larger backpacks on average.
- Research agenda: The research agenda includes visuals, defaults, warnings, proactive suggestions, social influence, and messenger effects.Examples include highlighting healthier food options and using avatars to increase recommender effectiveness.
6. Conclusion
The paper connects recommender systems with digital nudging and argues that recommendations inherently implement multiple nudging mechanisms. It identifies research gaps and outlines ways to incorporate additional mechanisms into recommender systems.
- Conclusion: Recommender systems and digital nudging mechanisms both influence online users’ behavior in desired ways.The paper emphasizes that recommender systems inherently implement various nudging mechanisms.
- Conclusion: The study identifies research gaps and future opportunities for incorporating additional nudging mechanisms into recommender systems.These opportunities are intended to further increase recommender-system effectiveness.
- Conclusion: The paper elaborates on the relationship between recommender systems and digital nudging.Its conclusion treats the two concepts as closely related rather than isolated topics.
Appendix A: List of Categorized Nudging Mechanisms
The appendix categorizes digital nudging mechanisms by how they present, phrase, and structure information or influence attention, evaluation, and choice. The mechanisms include informational aids, salience effects, cognitive shortcuts, and scarcity-related interventions.
- Information: Information-oriented mechanisms reduce ambiguity, tailor information, disclose relevant details, provide comparisons, and offer explanations or feedback.Examples include customized information, disclosure, informing, providing explanations, and providing feedback.
- Information: Some mechanisms make external information, multiple viewpoints, alternatives, goals, or progress visible to support decision assessment.These mechanisms include making external information visible, providing multiple viewpoints, suggesting alternatives, visible goals, and simplification.
- Comprehension: Understanding mapping explains difficult information through familiar concepts, analogies, or visual aids.This mechanism connects unfamiliar information to forms users can more readily understand.
- Salience and attention: Attention-related mechanisms attract users to selected information, hide undesirable options, or increase the salience of attributes and incentives.Examples include highlighting, hiding information, and emphasizing attributes such as weight, price, or color.
- Salience and attention: Warnings emphasize problems through visual or other means, while visuals can increase an option’s salience or make incentives more prominent.Visual effects may include colors, pictures, signs, or fonts.
- Phrasing and judgment: Phrasing mechanisms influence judgment through anchoring, selective attention, availability, memory effects, decoys, endowment, and framing.They alter starting points, remembered experiences, perceived value, ownership effects, or descriptions of options.
- Phrasing and judgment: Other mechanisms reduce cognitive effort, decouple payment from purchase, exploit time preferences, or shape responsibility and interpretation.The appendix includes hyperbolic discounting, framing, image motivation, and related mechanisms.
- Scarcity and valuation: Scarcity-oriented mechanisms use limited time windows or resources to make options seem more important and increase commitment probabilities.The appendix also describes loss aversion and mental accounting as mechanisms affecting valuation and spending interpretation.