together and enable the study of users’ physical behaviours by continuously tracking their loca- tion. Previously, it was conflicting to study and relate the behaviour of physical social network with the behaviour of online social network. Until recently, researchers found conflicting res- ults while comparing the two worlds together. Some studies argue that online social networks contribute in isolating people in physical world [20].
The growing ubiquity of location-enabled smart phone reduces the difference between online and offline social networks [20]. Foursquare is one of the most mobile applications that sup- ports new means for online interaction based on the physical location of their users. Although applications such as Foursquare have unique properties that support different research, it always requires human interaction (the user must check-in for every new location he/she has moved to). This will affect the accuracy of the research as the user may intentionally or unintentionally check-in at certain places.
2.3
Analysing Data through Social Network Structures
Pervasive computing technologies for mobility
& location detection
Data on mobility & location
Social Network Analysis
Opportunistic Network & Correlation with Human
Social network
Figure 2.2: The Third Step of Methodology Flow.
2.3.1
Extract Social Network Structure
Aron et al.[22] is the first paper that operates an end-to-end system that integrates information extraction and social network analysis together. The system builds up the social network links by extracting mentioned people from web pages and creating a link between the page owner and the extracted person. A recursive extraction and process builds larger networks that contain
18 2.3 Analysing Data through Social Network Structures
“friends of friends of friends“. It has been noticed that Bluetooth proximity data can be used for building a social network structure of the users [27].
An analysis of a Bluetooth proximity network has been presented in [19]. Pairwise links between people have been defined in [29] by using Bluetooth and phone call data instead of using questionnaire-based self reported data. In [62] an algorithm proposed group discovery that focused on fully connected sub networks of the Bluetooth proximity network. Some draw- backs emerging from this method are the noise effects and group size that grows gradually. Some researchers focus on the face-to-face aspect in order to overcome Bluetooth drawbacks by using specific mobile devices [41]. Using this method, the actual physical interaction has been sensed but it requires people to carry an additional device.
As every mobile phone is now equipped with Bluetooth, the research in individual/group beha- viour mining [102, 43] has received more attention recently. The Reality Mining project utilizes data from the mobile phone to observe and characterize the social behaviour of individual users and organizations [30]. Roue et al [85] extracted a social hierarchy by analysing the flow of emails within organizations. They focused on the users’ behaviour patterns to determine the strength of communication links between users. Corresponding to the relation between com- munication media and relationship strength between users, it was found that strong links (strong friendships or close friends) communicate more frequently using more media than weaker links [48].
To sum up, most of the recent research uses mobile phone data to study users’ mobility pattern and internet data such as emails, Facebook to learn about the users’ physical social network. However, they do not explore indoor mobility datasets to understand users’ mobility pattern which are challenging. These relationships can be used for information dissemination. Another limitation with previous research is that they do not record and utilize the physical interaction to understand the opportunistic networks between users and analyse their structure. These chal- lenges have been overcome through this research.
2.3.2
Infer Relationship Strength
Social network structures are important in order to analyse and produce some inferences from the collected mobility data. As seen in Figure 2.2, it is the third step in the research methodology of this study. Social network analysis can be applied to different types of data, such as email, phone logging and proximity, since the data represents interaction between people. Onnela et al [76] analyse the structure and tie strength of social and communication networks by utilizing the recorded calls of mobile phones. The results from the social network analysis provide
2.3 Analysing Data through Social Network Structures 19
social information that represents the frequency of interaction between people and the pattern of communication between them.
Humans are social by nature and social interaction is a common activity in daily lives [33]. Generally, there is always a link between people and each other, and this link is the evidence to show how people communicate in different ways. Recently, research has focused on identifying these kinds of links from peoples’ mobility patterns. This is to explore the extent to which new communication technology can maintain, expand or decrease existing relationships. In addition, it explores how these relationships differ or support the physical interaction relationship (face- to-face).
One of the challenging problems in this area is inferring the characteristics of users’ social behaviour from their location. Eagle et al [32, 31] and Li et al [58] develop and utilize the measure of users’ similarity to infer the social structure between users. This inference is very challenging where co-location users are defined as being in the same place at the same time, but there is no evidence of having a relationship between them. Cranshaw et al [20] introduce a set of features that focus on the social context of the locations that users visit. Then they evaluate these features by predicting both whether two co-located users are friends on Facebook, as well as the number of friends each has in the social network.
Dong et al [28] propose to explore individual friendship patterns from cellular phone call logs. On the other hand, Gho et al [18] study the relation between human geographic movements, temporal dynamics, and social network ties. They analyse the effect of geography on daily routine and social ties on the human mobility pattern. Tang et.al [92] infer social ties from heterogeneous networks and find that 70% of the online social networks (such as Facebook, Twitter, and LinkedIn) have not been well labelled.
Yu [101] used sensing data that was captured by the MIT Reality Mining project to explore the social relationships of evolution. Firstly, they define friendship as a directed link. Secondly, they recognize human friendship from a supervised learning perspective. They adapted an inference model that is able to predict friendship from a variety of features extracted from mobile phone data, including proximity, outgoing calls, outgoing text messages, incoming calls, and incoming text messages. This approach is called Support Vector Machine (SVM). Finally, they use social balance theory to demonstrate the social relation evolution.
In another study focusing on on-line communication methodologies, it was found that strong links use email and instant message more frequently than weak links [47]. The discovered rela- tionship between mobile users can be used for information dissemination in different network environments [52]. [53] proposed a novel paradigm that focuses on behaviour-driven commu- nication, enabling a new class of services in mobile societies.