Named Entity Recognition Based On Machine Learning Approach

Named Entity Recognition techniques have been the focus of recent Natural Language Processing research. NER is a form of information extraction in which we seek to classify every word in a document as being a person-name, location, organization, date, time, number or none of above. NER as a subtask of Information Extraction has been applied on many compute linguistics tasks, such as machine translation.In this thesis, two machine learning methods are applied on named entity recognition. One is maximum entropy method and another is boosting algorithm. The machine learning methods are both robust and portable compared with rule based approach. The system based on machine learning approach can be ported a new domain or language with minimal expense. At first a character-based model and a word-based model are constructed which only some basic features are employed. We compare the performance of two models on the task. In order to use the advantages of the two models, we decode the word segmentation information into the character-based model. Moreover, some complex linguistics knowledge features are employed in the model. The result of experiment shows that the performance of the model is better than the others mentioned above. Meanwhile, we compare the performance of the classifiers under the same conditions.Finite State Machine is employed to recognize date, time and number and extract the candidates of foreign person-name in a document. As a result, we concentrate on three types in machine learning framework, which can decrease the complexity of these algorithms. Finally global information is utilized to increase the performance of NER system.