• View
• Edit
• History
• Print

Social Interaction Network Extraction From Text

Abstract: Language is the primary tool that people use for establishing, maintaining and expressing social relations. People talk about who they meet, talk to, think about etc. using language. Though part of the social interactions that lead to stronger social relations may be captured by meta-data such as sending/receiving emails, a large part of the social interaction network is manifested in contents of such communications.

In this talk, I will present a system capable of extracting an interaction network from unstructured text such as emails and novels. We are using convolution kernels and Support Vector Machines (SVMs) to build a supervised model to detect interaction links between people. I will give a brief overview of convolution kernels, how they fit SVMs, their advantages and limitations. I will present some results on how extracting this social interaction network from Enron emails helps in predicting a better organizational hierarchy, and finally how this network gives us insight into the roles of characters in a work of fiction like Alice in Wonderland.

Bio: Apoorv Agarwal is a fourth year PhD candidate at Columbia University, NY. He received his M.S. in Computer Science from Columbia University in 2008, and BTech in Computer Engineering from NSIT, Delhi University, India, in 2006. He interned with the IBM DeepQA team that built Watson in Summer 2011 and Jan-Aug 2012.