Discovering Social Community Structures Based on Human Mobility Traces

Abstract

We consider a community detection problem in a social network. A social network is composed of smaller communities; that is, a society can be partitioned into different social groups in which the members of the same group maintain stronger and denser social connections than individuals from different groups. In other words, people in the same community have substantially interdependent social characteristics, indicating that the community structure may facilitate understanding human interactions as well as individual’s behaviors. We detect the social groups within a network of mobile users by analyzing the Bluetooth-based encounter history from a real-life mobility dataset. Our community detection methodology focuses on designing similarity measurements that can reflect the degree of social connections between users by considering tempospatial aspects of human interactions, followed by clustering algorithms. We also present two evaluation methods for the proposed schemes. The first method relies on the natural properties of friendship, where the longevity, frequency, and regularity characteristics of human encounters are considered. The second is a movement-prediction-based method which is used to verify the social ties between users. The evaluation results show that the proposed schemes can achieve high performance in detecting the social community structure.

Predicting Human Location Using Correlated Movements

“…Table 5 presents the setup for the mobility prediction framework. Specifically, the two proposed PCM selection methods consisting of community interaction similarity-based (CISB) and behavioral similarity-based (BSB) are evaluated and compared with three recent selection approaches: encounter frequency-based method (EFB) [18,23,24], spatial closeness (SC) [6], and spatiotemporal closeness (STC) [6]. Table 6 shows a list of acronyms which are used in the manuscript.…”

“…A number of studies were conducted to detect social communities using graph clustering [24,37] where a network was divided into disjoint communities by using clustering techniques. There were several studies on community detection based on the contact history of members in the network, e.g., encounter frequency and duration [38] and the total number of past encounters of a person [39].…”

“…Hence, this work uses the dataset of cellular traces in order to evaluate the movement prediction model. The subjects in the dataset participated in the experiment in different time periods, and some subjects have no data or very little data [24]. Therefore, by considering overlapping periods and available data, 43 people with sufficient mobility data were selected.…”

“…We draw a comparison between the designed PLP model and the baseline one, called most frequent location (MFL), in which the most visited position is predicted as the location of the person-of-interest. Two proposed PCMs selection methods are also compared with three recent existing approaches: encounter frequency-based (EFB) [18,23,24], spatial closeness (SC) [6], and spatio-temporal closeness (STC) [6]. Specifically, EFB is rooted in encounter frequency, which is one of the natural characteristics of a friendship, i.e., two members who more frequently meet each other are likely to be closer friends.…”

Discovering Individual Life Style From Anonymized WiFi Scan Lists on Smartphones

“…Zhao et al [9] discovered individual's life regularity from anonymized WiFi logs, such as the visiting orders of different places. Nguyen et al [10] measured the social similarity among mobile phone users by analyzing their cell tower logs and Bluetooth proximity traces, and determined social groups among individuals in human society.…”