Named Entity Recognition Based On Stacking Framework

The aim of named entity recognition (NER) in Chinese language is to recognize phrases that indicate the names of entities. NER is the basic task of many applications such as machine translation, text classification, information retrieval, automatic abstracting and automatic question answering. As a fundamental problem in information extraction (IE), named entity recognition has always been a hot spot in natural language processing (NLP). The research on NER in Chinese is a significant task.NER in Chinese language mainly focuses on the recognition of person names, locations and organization names. In this thesis, combined multiple classifiers based on a stacking framework are used to recognize named entities, the definitions and evaluation measures are invoked from SIGHAN. We combined global and local features to train the learning model. Experiment results show that adding word list information into the system can improve recognition performance.In this thesis, we described the stacking framework and its definition, model structure, training method and selection of model algorithm. Different feature templates are proposed for person names, locations and organization names respectively. We used charters instead of phrases in feature selection, eliminating the pos-tagging, to achieve a better precision and to keep our system independent. Local features of an entity are selected within the sentence it appears, and all instances of an entity in the corpora are used to extract the global features. Investigating these methods, our system performed in Chinese NER and proved its effectiveness.