A simple semi-supervised algorithm for named entity recognition

We present a simple semi-supervised learning algorithm for named entity recognition (NER) using conditional random fields (CRFs). The algorithm is based on exploiting evidence that is independent from the features used for a classifier, which provides high-precision labels to unlabeled data. Such independent evidence is used to automatically extract high-accuracy and non-redundant data, leading to a much improved classifier at the next iteration. We show that our algorithm achieves an average improvement of 12 in recall and 4 in precision compared to the supervised algorithm. We also show that our algorithm achieves high accuracy when the training and test sets are from different domains.