A Hybrid Approach for Chinese Named Entity Recognition in Music Domain

The amount of music information available on the Web is rapidly increasing. There is a pressing need for music information extraction. To extract useful information from natural language text, we must recognize music named entities first. This paper introduces a hybrid method to identify the Chinese named entities in music domain. Recently, machine learning approaches are frequently used to solve name entity recognition (NER). So our method uses a hidden Markov model (HMM) as the underlying method. Since HMM has innate weaknesses, we incorporate it with rule-based method for pre-processing and post-processing. The combination of machine learning method and rule-based method results in a high precision recognition. And we improve both training and recognizing process of HMM for music named entity recognition (MNER). In this paper, a novel and convenient musical name entity (MNE) tagging method to generate training data is proposed, which makes HMM method practically usable. In addition, we present an effective method of unknown words tagging in recognition. The experimental results show that our framework brings significant improvements for solving MNER.