Rule-based pattern extractor and named entity recognition: A hybrid approach

Name Entity Recognition (NER) is one of the important tasks in Information Extraction (IE) research that has been flourishing for more than fifteen years ago. NER enables an IE system to recognize and classify information units in an unstructured text. This paper presents a Rule-based pattern extractor and a Semi-Supervised NER approach to automatically generate extraction pattern from a limited corpus and label the pre-defined entities in a collection of accident documents. Link Grammar parser and Stanford Part-of-Speech tagger are used in the pattern extractor to identify named entity and construct extraction pattern. The extraction pattern then feed to Semi-Supervised NER to categorize the entities into some predefined categories. Performance is evaluated using Exact Match evaluation and tested on two different entities-DATE and LOCATION. Using only two features, the system shows promising result.