TwistLens: A Docent-Informed Image Transformation to Create Previews That Prompt Anticipation and Interpretive Experiences Before Museum Visits

Paper Title

TwistLens: A Docent-Informed Image Transformation to Create Previews That Prompt Anticipation and Interpretive Experiences Before Museum Visits

Publication Info

• Topic area: AI-driven visual transformation for museum preview design.
• Keywords: anticipation, interpretive learning, museum previews, AI transformation, EchoLens, DecoyLens, semantic analysis, spoiler prevention, curiosity, docent-informed design.

Background and Problem

• Problem / challenge: Traditional museum previews either overwhelm visitors with text or spoil the surprise with full visual exposure. Existing spoiler-management techniques focus on text or privacy, neglecting the balance between anticipation and interpretive learning in visual contexts.
• Significance: Anticipation enhances curiosity and enjoyment, while interpretive learning deepens engagement. Addressing this balance can transform pre-visit materials into tools for richer museum experiences.
• Motivation and related work: Prior research highlights the importance of ambiguity and curiosity in fostering engagement but lacks methods to apply these principles to visual previews. Current approaches to visual obfuscation focus on privacy rather than anticipation or learning, leaving a gap in designing previews that stimulate curiosity without spoiling the experience.

Solution

• Proposed approach: TwistLens, an AI-supported, docent-informed image transformation system that generates semantically twisted previews to preserve anticipation and foster interpretive learning.
• Novelty:
  1. Introduction of two transformation strategies: EchoLens (preserves meaning, alters representation) and DecoyLens (distorts meaning, maintains coherence).
  2. Development of a structured Information Taxonomy to guide semantic transformations.
  3. Empirical demonstration of how semantic transformations enhance anticipation, curiosity, and learning in museum contexts.
• Procedure and key techniques:
  1. Input artwork and docent text.
  2. Analyze text using Information Taxonomy to identify key elements.
  3. Segment image regions corresponding to key elements using multimodal analysis.
  4. Apply EchoLens or DecoyLens transformations to segmented regions.
  5. Generate previews with clear visual markers indicating altered regions.

Results

• Concrete findings:
  • TwistLens increased pre-visit anticipation (M = 5.30 vs. 4.40, p < .01) and curiosity (M = 5.05 vs. 3.70, p < .001).
  • Post-visit, it enhanced enjoyment (M = 5.50 vs. 4.15, p < .001) and surprise (M = 5.00 vs. 2.25, p < .001).
  • DecoyLens was more effective for spoiler prevention (M = 4.75 vs. 1.65, p < .001), while EchoLens excelled in visual comfort and preserving interpretive clarity.
• Advantage over baselines: Compared to traditional previews, TwistLens better preserved anticipation, triggered curiosity, and supported active learning without spoiling the original visuals.
• Experiments / evaluation:
  • Co-design study with 21 art enthusiasts refined strategy preferences across eight information categories.
  • Controlled evaluation with 20 participants compared TwistLens to baseline brochures in virtual exhibitions, measuring anticipation, curiosity, learning, and spoiler prevention.
• Limitations and future work:
  • Limited artist and curator input; future iterations should involve these stakeholders.
  • Current system operates as a pre-processing tool; future work could explore real-time or interactive transformations.

Summary

TwistLens is an AI-driven system that creates docent-informed, semantically transformed previews to balance anticipation and interpretive learning in museum contexts. Through two strategies, EchoLens and DecoyLens, it selectively alters visual elements to preserve curiosity and prevent spoilers. Empirical studies demonstrated its effectiveness in enhancing anticipation, curiosity, and learning compared to traditional previews. Future work aims to involve more artist-curator collaboration and explore real-time, interactive applications in broader domains like education and promotional media.