Social information can be used to enhance existing applications and services or can be utilized to devise entirely new applications. Examples of such applications include recommendation systems, peer-to-peer networks, opportunistic data dissemination in ad hoc networks, or mobile friend finder. Social information can be collected from either online or mobile sources.
This thesis focuses on identifying social groups based on data collected from mobile phones. These data can be either location or co-location traces. Unfortunately, location traces require a localization system for every mobile device, and users are reluctant to share absolute location due to privacy concerns. On the other hand, co- location can be collected using the embedded Bluetooth interface, present on almost all phones, and alleviates the privacy concerns as it does not collect user location.
Existing graph algorithms, such as K-Clique and WNA, applied on co-location traces achieve low group detection accuracy because they focus on pair-wise ties, which cannot tell if multiple users spent time together simultaneously or how often they met.
This thesis proposes the Group Discovery using Co-location (GDC) algorithm, which leverages the meeting frequency and meeting duration to accurately detect social groups. These parameters allow us to compare, categorize, and rank the groups discovered by GDC. This algorithm is tested and validated on data collected from 141 active users who carried mobile phones on our campus over the duration of one month. GDC received ratings that were 30% better than the K-Clique algorithm.