基于隐马尔可夫模型的引文信息提取

Citation Extraction Based on Hidden Markov Model

提出一种基于符号特征提取的HMM结构学习方法，并利用修改的Viterbi算法进行引文信息提取。在实验结果的评价上，除了常用的准确率(Precision)、查全率(Recall)、综合评价指标(F)之外，引入了两个评价标准：宏平均和微平均来综合评价算法的性能。试验表明该方法用于引文信息提取的正确率较高。

This paper puts forward a method of HMM structure learning which is based on the extraction of tokens' features, uses a variation of Viterbi algorithm to extract citation. For evaluating the method, it introduces two standards: Macro-averaging and Mico-averaging in addition to commonly used Precision, Recall and F.