Spatial properties of online social services : measurement, analysis and applications

Online social networking services entice millions of users to spend hours every day interacting with each other. At the same time, thanks to the widespread and growing popularity of mobile devices equipped with location-sensing technology, users are now increasingly sharing details about their geographic location and about the places they visit. This adds a crucial spatial and geographic dimension to online social services, bridging the gap between the online world and physical presence. These observations motivate the work in this dissertation: our thesis is that the spatial properties of online social networking services offer important insights about users' social behaviour. This thesis is supported by a set of re.sults related to the measurement and the analysis of such spatial properties. First, we present a comparative study of three online social services: we find that geographic distance constrains social connections, although users exhibit heterogeneous spatial properties. Furthermore, we demonstrate that by considering only social or only spatial factors it is not possible to reproduce the observed properties. Therefore, we investigate how these factors are jointly influencing the evolution of online social services. The resulting observations are then incorporated in a new model of network growth which is able to reproduce the properties of real systems. Then, we outline two case studies where we exploit our findings in real application scenarios. The first concerns building a link prediction system to find pairs of users likely to connect on online social services. Even though spatial proximity fosters the creation of social ties, the computational challenge is accurately and efficiently to discern when being close in space results in a new social connection. We address this problem with a system that uses, alongside other information, features based on the places that users visit. The second example presents a method to extract geographic information about users sharing online videos to understand whether such videos are going to become locally or globally popular. This information is then harnessed to build caching policies that consider which items should be prioritised in memory, thus improving performance of content delivery networks. We summarise our findings with a discussion about the implications of our results, debating potential future research trends and practical applications.

Name dc:type.qualificationname

Doctor of Philosophy (PhD)

Level dc:type.qualificationlevel

Doctoral

Grantor dc:publisher.institution

University of Cambridge

Year dc:date.issued

2012

Author dc:creator

• Scellato, Salvatore

Licence

• https://apollo8-f-pro.lib.cam.ac.uk/bitstreams/5b5bc739-76e4-4582-a4d3-99d6e2d9c52e/download
• https://www.rioxx.net/licenses/all-rights-reserved/

Language dc:language

eng

DOI dc:identifier.doi

https://doi.org/10.17863/CAM.11725

OAI identifier oai:identifier

oai:www.repository.cam.ac.uk:1810/265547