基于CRF的蒙古文人名自动识别研究*

摘要：该研究以蒙古文人名识别为目的，实现了基于条件随机场模型的人名自动识别。首先从蒙古语黏着性特点分析入手，研究了蒙古语语料库中人名的存在形式以及各类人名的特点，针对蒙古语语料库中人名的特点，在词汇特征、词性特征和指示词特征等基本特征基础上引入了汉语姓氏特征、人名词典特征、兼类人名特征以及双词根特征。以内蒙古大学开发的100万词规模的标注语料库为训练数据，该模型的人名识别性能达到了94.56%的准确率，90.60%的召回率和92.54%的F值。该方法比起以往的基于的规则的系统取得了较好的结果。

Recognition Method of Mongolian Person Names Based on Conditional Random Fields

Abstract: The paper presents a method to recognize Mongolian names based on conditional random fields (CRF). According to the characteristics of the Mongolian person names, we selected lexical features, part of speech, designation words, Chinese surname, names dictionary, category names and double roots as the features of the model. Using the 3rd-level annotated corpus with about 1000,000 words as the training data, this model achieves a accuracy of 94.56%, the recall rate of 90.60%, F score of 92.54%.the method had achieved good results compared with the previous rule based system.