With the increasing prevalence of large scale online social networks, the field has evolved from studying small scale networks and interactions to massive ones that encompass huge fractions of the world’s population. While many methods focus on techniques at scale applied to a single domain, methods that apply techniques across multiple domains are becoming increasingly important. These methods rely on understanding the complex relationships in the data. In the context of social networks, the big data available allows us to better model and analyze the flow of information within the network.
The first part of this thesis discusses methods to more effectively learn and predict in a social network by leveraging information across multiple domains and types of data. We document a method to identify users from their access to content in a network and their click behavior. Even on a macro level, click behavior is often hard to obtain. We describe a technique to predict click behavior using other public information about the social network.
Communication within a network inevitably has some bias that can be attributed to individual preferences and quality as well as the underlying structure of the network. The second part of the thesis characterizes the structural bias in a network by modeling the underlying information flow as a commodity of trade.
Identifer | oai:union.ndltd.org:columbia.edu/oai:academiccommons.columbia.edu:10.7916/D8H99HNX |
Date | January 2017 |
Creators | Ramachandran, Arthi |
Source Sets | Columbia University |
Language | English |
Detected Language | English |
Type | Theses |
Page generated in 0.001 seconds