Graph-Theoretic Approaches to Minimally-Supervised Natural Language Learning

A0e65af9a6baff8efb7e632212f5eec3?s=47 Mamoru Komachi
September 09, 2013

Graph-Theoretic Approaches to Minimally-Supervised Natural LanguageĀ Learning

Slides presented at the Web & AI Seminar, National Institute of Informatics. Parts of the slides are taken from my dissertation defense (March 2010). We investigated the root of so-called "semantic drift" in Espresso-style bootstrapping algorithms using graph-theoretic approaches. It turned out that semantic drift is a parallel to the topic drift in the well-known link analysis algorithm, HITS (Kleinberg, 1999). We also showed that the regularized Laplacian reduces the effect of semantic drift, and is easy to use compared to the state of the art bootstrapping algorithm, Espresso.

A0e65af9a6baff8efb7e632212f5eec3?s=128

Mamoru Komachi

September 09, 2013
Tweet