Hybrid Approach for Named Entity Recognition

This paper proposes the Named Entity Recognition (NER) system for Punjabi language using a hybrid approach in which rule based approach and machine learning approach i.e. Hidden Markov Model (HMM) is combined. With no Dataset available, the Named Entities (NEs) were manually tagged which led us to the creation of training and testing dataset, under the linguistic supervision. Using hybrid approach, the proposed system is able to recognize Name of person, Location, Time, Date, Designation, Organization, Titleperson, Event, Abbreviation, Facility, Number, Artifact, Relation and Measure. This paper presents two versions of NER for Punjabi language, the first version is designed with HMM only and the second version is designed hybrid approach in which HMM is used in combination with handcrafted rules. NER system with proposed hybrid approach is able to achieve the precision of 72.92%, Recall of 76.27%, F-measure of 74.56% with hybrid approach and Precision, Recall and F-measure of 47.57%, 48.98%, 48.27% respectively has been achieved by using HMM only. This paper has also compared proposed method with simple HMM and observed that proposed NER system performs better. General Terms Named Entity recognition, Natural Language Processing (NLP), Hybrid approach.