CAJ | 학술논문

中文嵌套命名实体识别是自然语言处理中一个比较困难的问题。针对传统的序列化标注方法的不足，本文提出了一种新的基于联合模型的中文嵌套命名实体识别方法，该方法将嵌套命名实体识别看作是一种联合切分和标注任务。联合模型用一种改进的beam search算法作为系统的解码算法，并采用一种在线学习算法平均感知器算法作为训练算法，获得了较快的收敛速度和较好的识别效果。实验结果表明基于联合模型的方法对嵌套命名实体识别取得了更好的效果。
중문감투명명실체식별시자연어언처리중일개비교곤난적문제。침대전통적서렬화표주방법적불족，본문제출료일충신적기우연합모형적중문감투명명실체식별방법，해방법장감투명명실체식별간작시일충연합절분화표주임무。연합모형용일충개진적beam search산법작위계통적해마산법，병채용일충재선학습산법평균감지기산법작위훈련산법，획득료교쾌적수렴속도화교호적식별효과。실험결과표명기우연합모형적방법대감투명명실체식별취득료경호적효과。
Chinese nested named entity recognition is a very difficult problem in natural language processing. This paper presents a novel method based on a joint model,which treats the recognition of Chinese nested named entity as a task of joint word segmentation and labeling. The proposed method exploits an improved beam search algorithm as decoding algorithm,and uses the averaged perceptron algorithm as training algorithm,attaining fast convergence during training. The experimental results show that the joint model achieves better performance than two baseline systems using the traditional sequence labeling models.