A Context-Assisted, Semi-Automated Activity Recall Interface Allowing Uncertainty.

Area of Science:

• Ubiquitous Computing
• Human-Computer Interaction
• Personal Health Informatics

Background:

• Accurate activity and posture measurement is crucial for health informatics and HCI research.
• Self-reporting daily activities is a common data collection method, but precise temporal recall is challenging.
• Existing methods for activity recall lack efficiency and can compromise data validity.

Purpose of the Study:

• To introduce ACAI (Context-Assisted Activity Annotation Interface), a novel system for efficient and accurate self-reported activity annotation.
• To evaluate ACAI's usability and effectiveness in free-living conditions compared to standard recall methods.
• To assess ACAI's impact on annotation time, perceived effort, and data quality for activity recognition.

Main Methods:

• Developed ACAI, a context-assisted interface providing activity suggestions for user acceptance or adjustment.
• Conducted a usability study (11 participants) and a two-week free-living study (14 participants).
• Compared ACAI against 24PAR and ACT24, established methods for activity recall in health sciences.

Main Results:

• ACAI significantly reduced annotation time and perceived user effort.
• The system demonstrated improved data validity and fidelity compared to standard human-supervised and unsupervised recall methods.
• Participants could efficiently label activities, accepting or adjusting system suggestions and indicating temporal boundary uncertainty.

Conclusions:

• ACAI offers an efficient and effective solution for collecting high-quality self-reported activity data.
• The findings support the development of adaptive, human-in-the-loop systems for activity recognition.
• ACAI has significant implications for research in ubiquitous computing, HCI, and personal health informatics.