MomentMeld: AI-augmented Mobile Photographic Memento towards Mutually Stimulatory Inter-generational Interaction

Title of the Paper

MomentMeld: AI-augmented Mobile Photographic Memento towards Mutually Stimulatory Inter-generational Interaction

Paper Information

• Subject Area: Human-Computer Interaction (HCI), Artificial Intelligence, Social Computing
• Keywords: Inter-generational interaction, mutually stimulatory memento, semantic similarity, deep neural networks, artificial intelligence, aging, mobile photography, smartphones

Research Background and Problem

• Background and Problem:
  As people age, their social interactions gradually decrease, which not only affects their life satisfaction but also leads to feelings of loneliness and social isolation, increasing health risks. Within families, interactions between elderly parents and adult children also tend to decline. Although the birth of grandchildren can somewhat increase interactions, these interactions are primarily centered around topics related to the grandchildren and are often one-sided, making it difficult for the elderly to actively participate or guide the interaction.

• Significance:
  The reduction in inter-generational interaction has a direct negative impact on the life satisfaction and mental health of the elderly. Therefore, exploring how technology can improve the frequency and quality of interactions between elderly parents and adult children is an important research topic.

• Research Motivation:
  The authors aim to develop an innovative AI-driven service embedded in everyday life scenarios to foster inter-generational connections among family members.

Solution

• Research Method and Innovation:
  The concept of Mutually Stimulatory Memento is proposed, which involves juxtaposing semantically related inter-generational photos to evoke associations, resonance, and emotional memories within families. This is a visual interaction aid designed to naturally integrate into users' daily activities, such as taking photos and chatting.

• Technical Implementation:
  A mobile cloud service called MomentMeld was developed, with core functionalities including:

  1. Automatically generating "mutually stimulatory mementos": The system uses deep learning models to extract semantically similar cross-generational photos from existing photo libraries.
  2. Providing instant sharing mechanisms: After taking a new photo, users receive recommendations for similar historical photos, which can be directly embedded into communication apps for sharing with older generations.
  3. Utilizing deep neural network (DNN) models to evaluate attributes such as scene similarity, facial expressions, poses, and background objects in photos.

• Key Implementation Steps:

  1. Data Processing and Semantic Matching: The server-side AI model analyzes key semantic attributes (e.g., scenes, expressions, poses) in selected photos and calculates similarity.
  2. User Guidance and Interaction: The mobile app notifies users and gradually optimizes the recommendation algorithm based on user choices.
  3. System Optimization and Deployment: Dynamic online learning is used to further enhance photo matching accuracy and user satisfaction.

Research Outcomes

• Experiments and Deployment:
  The research team developed and deployed a complete service prototype, conducting an 8-week real-world usage test with 6 families (33 members in total):

  • Increased Interaction Frequency (M1): Inter-generational interaction frequency increased by 90% post-usage (p = 0.012).
  • Extended Interaction Duration (M2): The length of each interaction session increased by 39% (p = 0.025).
  • Enhanced Balance in Inter-generational Participation (M3): The proportion of active participation by the older generation in interactions increased by 49% (p = 0.001).

• Advantages and User Feedback:

  • Efficient Interaction Trigger: Users generally reported that the paired mementos enhanced the fun and initiative of interactions.
  • Natural Integration into Daily Life: The system leveraged users' daily photo-taking habits, lowering the barrier to use.
  • Filial Piety and Emotional Connection: Users noted that the system helped overcome discomfort in contacting parents and strengthened emotional bonds across three generations.

• Limitations and Future Directions:

  • Incomplete Coverage of Family Subjective Attributes: The study did not integrate certain family-specific semantics (e.g., cultural background, history).
  • Short to Medium-Term Validation Period: Further exploration is needed for long-term usage and user retention studies.
  • Potential for Model Improvement: There is room for optimization in the intelligent matching model, such as better support for personalized family photo needs.

Conclusion

MomentMeld innovatively combines artificial intelligence technology with everyday family life, significantly enhancing the frequency and quality of inter-generational interactions through visual memory triggers and emotional resonance. This technology demonstrates the potential of AI in fostering intimate relationships and addressing aging-related challenges, while also providing valuable insights for designing interactions in broader social contexts.