Automatically Building Training Examples for Entity Extraction

In this paper we present methods for automatically acquiring training examples for the task of entity extraction. Experimental evidence show that: (1) our methods compete with a current heavily supervised state-of-the-art system, within 0.04 absolute mean average precision; and (2) our model significantly outperforms other supervised and unsupervised baselines by between 0.15 and 0.30 in absolute mean average precision.