Indexing Multivariate Mobile Data through Spatio-Temporal Event Detection and Clustering.

Area of Science:

• Computer Science
• Human-Computer Interaction
• Data Science

Background:

• Mobile and wearable devices generate rich contextual sensor data for user behavior quantification.
• Challenges in indexing and annotating raw, multivariate, and sparse sensor data hinder advanced human-centric search.
• Existing methods struggle with the complexity of sensor data, limiting effective information retrieval.

Purpose of the Study:

• To propose a novel pipeline of algorithms for efficient indexing and annotation of mobile contextual sensor data.
• To enhance the accuracy and efficiency of human behavior quantification using sensor data.
• To improve information retrieval capabilities for human-centric search applications.

Main Methods:

• A three-algorithm pipeline including spatio-temporal event detection and mobile contextual data clustering.
• Development of a spatio-temporal clustering approach for annotating raw sensor data.
• Implementation of an algorithm to identify contrasting events within data clusters for refined behavior analysis.

Main Results:

• The proposed spatio-temporal clustering approach effectively annotates raw sensor data, reducing search space.
• Algorithms demonstrated utility and resource efficiency in evaluating two large-scale real-world smartphone datasets.
• The pipeline significantly improves information retrieval by enabling searches within relevant data clusters.

Conclusions:

• The developed pipeline offers an effective solution for indexing and annotating mobile sensor data.
• This approach enhances human behavior quantification and facilitates more efficient human-centric search.
• The algorithms show promise for advancing the capabilities of mobile and wearable device data analysis.