Human Behavior Analysis by Means of Multimodal Context Mining

Area of Science:

• Computer Science
• Human-Computer Interaction
• Behavioral Science

Background:

• Negative lifestyle choices significantly impact health and wellness.
• Current behavior monitoring methods are often subjective (questionnaires) or limited (activity counters).
• Accurate, continuous user behavior description is crucial for health interventions.

Purpose of the Study:

• To present an innovative multimodal context mining framework for holistic human behavior inspection and inference.
• To overcome limitations of existing methods by integrating diverse contextual information.
• To enable a more comprehensive understanding of user conduct for improved health outcomes.

Main Methods:

• Utilized machine learning techniques to identify low-level contexts (activities, emotions, locations) from heterogeneous sensory data.
• Employed ontological mechanisms to combine low-level contexts into a high-level contextual representation.
• Developed and implemented a real-time context identification framework.

Main Results:

• The framework successfully integrates diverse levels of context (low-level and high-level).
• Demonstrated prominent context-aware capabilities in realistic scenarios and on-the-go data.
• Evaluated using a novel multimodal context open dataset.

Conclusions:

• The proposed multimodal framework offers a holistic approach to understanding human behavior.
• This method enhances self-awareness and provides richer data for healthcare professionals.
• The system shows significant potential for real-world applications in health and wellness monitoring.