特征值区间的有效划分对CRF识别效果的影响

Influence of Effective Eigenvalue Extent Partition on CRF's Performance

与其他统计机器学习方法相比,条件随机场(CRF)算法更适合应用到命名实体的识别工作中来。在对中文化学物质名称进行识别的研究中,通过实验发现,有效的特征值区间划分能提高CRF的识别效果。另外,对词一级序列标注和字一级序列标注在不同特征值区间划分下的识别效果进行比较。

Comparing with other machine learning method, CRF is suitably to be applied to the research on NER （named entity rec- ognition）. In the course of recognizing Chinese chemical substance names, the authors find that effective eigenvalue extent partition can distinctly improve the performance of CRF. In addition, the authors also conduct an experiment for comparing the recognition performance labeled on word with that labeled on char under different eigenvalue extent partition.