Identification of Different Feature Sets for NER Tagging Using CRFs and Its Impact

This paper presents a study of the impact of different types of language modeling by selecting different feature matrices in the Conditional Random Fields (CRFs) learning algorithm for Named Entity tagging. We have come up with four different feature matrices and identified features at word, phrase and sentence level. It is identified that the language model which has the structural feature is better than the models with other features. Adapted from the source document