Adaptivity and Personalization in Persuasive Technologies

Shlomo Berkovsky, Maurits Kaptein, Massimo ZancanaroPpt@persuasive 2016

Abstract

Persuasive technologies (PT) are increasingly used to influence human behavior, yet most systems rely on a "one-size-fits-all" approach that ignores individual differences. This paper proposes a multidimensional framework for designing personalized persuasive systems that adapt to user traits, behaviors, and contexts. By analyzing three distinct case studies—ranging from fitness games to health reminders—the authors demonstrate that adaptive interventions significantly increase effectiveness and sustain user engagement longer than static alternatives.

Introduction: Beyond One-Size-Fits-All

Persuasive technologies are ubiquitous, from fitness apps nudging users to exercise to e-commerce platforms using scarcity tactics to drive sales. However, a strategy that motivates a competitive athlete might discourage a casual walker, and a "consensus" argument ("others are doing it") might fail for someone who values non-conformity. The authors address the lack of a unified design space for personalization in PT, identifying a gap between the potential of personalized persuasion and its actual implementation.

A Framework for Personalization

The authors propose a framework categorized into three principal dimensions to guide the design of adaptive systems:

  1. The Data (What facilitates personalization?): Systems can use domain-specific features (e.g., gaming skill) or general traits (e.g., personality). Data can be gathered explicitly via surveys or implicitly through behavioral logs, and the resulting user models can be static or dynamic.
  2. The Intervention (How is personalization achieved?): Systems can personalize the means (how a message is framed) or the ends (the target behavior itself). Persuasion can be direct or masked behind another activity (indirect).
  3. The Metrics (How is it evaluated?): Success is measured objectively (actual behavior change) or subjectively (user enjoyment), across short-term or long-term timeframes.

Case Studies and Methodology

The paper maps three experimental setups onto this framework:

  • Case 1 (Physical Activity Games): 90 players were evaluated using K-Nearest-Neighbors (KNN) to adjust game difficulty and rewards based on implicit player skill.
  • Case 2 (Health Reminders): 1,129 users received reminder emails to upload activity data. An adaptive algorithm selected between persuasion principles (Scarcity, Authority, Consensus) based on historical response rates.
  • Case 3 (Augmented Café Table): 24 visitors in a museum setting interacted with a table that used computer vision and audio sensors to adapt ambient visualizations, encouraging group conversation.

Key Results

Balancing the Playing Field in Gaming

In the gaming study, the "Tailored Reward" (TR) technique successfully leveled the playing field. High-skilled players, who typically performed less physical activity in a fixed setting, saw their jump count increase by 29% under the TR condition. Conversely, low-skilled players' counts decreased by 11%, resulting in a more balanced exercise load across the group.

Figure 1: Comparison of jumps captured between baseline (BL) and tailored rewards (TR) across skill levels.

Sustaining Engagement in Health Reminders

The health service study revealed that the "Adaptive" condition had a significantly higher success probability over time compared to random or "best pre-tested" messages. Crucially, the effectiveness of adaptive messages decayed much slower than non-personalized ones, suggesting that personalization is key to preventing user fatigue.

Figure 2: Probability of success for different reminder message strategies over time, showing the superior performance of adaptive messaging.

Enhancing Social Interaction

Adaptive stimuli in the museum study (visuals in a digital "Zen pond") were more effective at sparking conversation than random ones (26% vs. 14%). Users also spoke more frequently during the adaptive condition, accounting for 38% of the time compared to 34% in the control.

Discussion and Ethical Considerations

The authors highlight several critical insights for future PT design:

  • Implicit Data Collection: Observing behavior is often superior to questionnaires as it reduces user burden and avoids "social desirability bias."
  • Means vs. Ends: While most systems personalize the way they ask (means), long-term health management may require personalizing the goals themselves (ends).
  • The Disclosure Dilemma: In these studies, personalization was not disclosed to users to avoid "spoiling" the effect. This raises ethical concerns regarding transparency and whether users are being manipulated without consent.

Conclusion

While adaptive systems show great promise in the short term, the "holy grail" of persuasive technology remains long-term, stable lifestyle change. Future work must balance persuasive efficacy with user agency and privacy, ensuring that as systems become more invisible and indirect, they remain ethically grounded and transparent to the users they aim to help.

Summary updated

BibTeX
@inproceedings{berkovsky2016adaptivity,
  author = {Shlomo Berkovsky and Maurits Kaptein and Massimo Zancanaro},
  title = {{Adaptivity and Personalization in Persuasive Technologies}},
  booktitle = {Ppt@persuasive},
  year = {2016},
}
Figure