Improving the Performance of a Named Entity Recognition System with Knowledge Acquisition

Abstract

Named Entity Recognition (NER) is important for extracting information from highly heterogeneous web documents. Most NER systems have been developed based on formal documents, but informal web documents usually contain noise, and incorrect and incomplete expressions. The performance of current NER systems drops dramatically as informality increases in web documents and a different kind of NER is needed. Here we propose a Ripple-Down-Rules-based Named Entity Recognition (RDRNER) system. This is a wrapper around the machine-learning-based Stanford NER system, correcting its output using rules added by people to deal with specific application domains. The key advantages of this approach are that it can handle the freer writing style that occurs in web documents and correct errors introduced by the web's informal characteristics. In these studies the Ripple-Down Rule approach, with low-cost rule addition improved the Stanford NER system's performance on informal web document in a specific domain to the same level as its state-of-the-art performance on formal documents.