[Colloquium] Seminar Announcement: Mining and Proposing Links Among Academics

[Ninfa Mayorga]

Computation Institute Presentation - Data Lunch Seminar (DLS)

Speaker:  Aaron Gerow, Knowledge Lab, Computation Institute, University of Chicago
Host:  Tanu Malik 
Date:  August 22, 2014
Time: 12:00 PM - 1:00 PM
Location: University of Chicago, Searle 240A, 5735 S. Ellis Ave.

Mining and Proposing Links Among Academics 

Abstract: 
Nearly all professional relationships are multi-faceted, consisting of connections like sharing a department, an office, common research interests, acquaintances or supervisor. With increasingly robust methods of extracting relational information -- particularly from text -- network analysis can examine not only what actors are connected, but also how they are connected. Using faculty CVs from the University of Chicago, we find that co-occurring named entities (people, places, and organizations) provide links that can be used for accurate departmental classification. Building on this simple result, we develop a network-based method for proposing new relationships that prioritize similar sets of links between intermediaries. The new ties are proposed with a strength proportional to the aggregate amount and similarity of links between common neighbors. This task, equivalent to the weighted transitive closure, is trivial for unipartite graphs with untyped edges, but in a hyper-graph where link-types out-number nodes, a measure of link similarity is crucial. Using two socially-grounded validation schemes, we show that 1) a distributional measure of link similarity yields better proposals than treating links uniformly, and 2) set-based methods out-perform spatial methods when comparing link-sets. These results support the network-based approach and highlight the usefulness of link similarity in complex networks.
Bio: 
Aaron Gerow is a postdoc in the Knowledge Lab, lead by Prof. James Evans. He holds an MA in cognitive science and a PhD in computer science, having specialized in natural language processing and the statistical semantics of metaphor. His current work includes projects in graph-based NLP, text mining, topic modeling and term extraction.
Information: Lunch will be provided


-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mailman.cs.uchicago.edu/pipermail/colloquium/attachments/20140815/99a9a21c/attachment.htm