Training a Named Entity Recognizer on the Web

Abstract

In this paper, we introduce an approach for training a Named Entity Recognizer (NER) from a set of seed entities on the web. Creating training data for NERs is tedious, time consuming, and becomes more difficult with a growing set of entity types that should be learned and recognized. Named Entity Recognition is a building block in natural language processing and is widely used in fields such as question answering, tagging, and information retrieval. Our NER can be trained on a set of entity names of different types and can be extended whenever a new entity type should be recognized. This feature increases the practical applications of the NER.