基于BP-HMM的词性标注方法的研究

Study on part of speech tagging method based on BP-HMM

针对传统的HMM模型在词性标注中具有对兼类词及其对应词类标注能力差和模型差异大的缺点,提出一种利用混合BP-HMM词性标注模型进行词性标注算法。该算法通过BP网络优秀的甄别能力有效的弥补了HMM在对兼类词进行标注方面的不足,同时也利用HMM增强了BP网络的建模能力。实验结果表明,该模型相比传统的HMM以及BP网络模型,建模能力、分类性以及适应性都得到很大的增强,准确率也得到了2%~7%的提高。

The traditional HMM model has bad performance in tagging double words and the corresponding pos with huge differ- ence between models, a kind of annotation model for part-o^speech tagging algorithm using hybrid BP-HMM part of speech is put forward, the algorithm through the BP network good discrimination ability effectively compensates for the HMM in the multi-category word tagging problems, while also using HMM to enhance the modeling ability of BP network. The experimental results show that, the model is compared with the traditional HMM and BP network model, modeling, classification and adapta- bility have been greatly enhanced, the accuracy rate of 2%-7% has also been improved.