面向方志类古籍的多类型命名实体联合自动识别模型构建

Construction of Automatic Recognition Model of Multi-Type Named Entities for Local Gazetteers

方志类古籍是特色典籍,是优秀中华文化的重要载体。命名实体识别是中文古籍文本挖掘与利用的重要环节,对理解古籍内容起到积极作用。文章以数字化的特色馆藏方志农业典籍《方志物产》为研究语料,邀请领域专家制定语料标注标准,并采用交叉互核的方式进行语料标注,构建实体的内外部特征模板,完成基于条件随机场的自动实体识别,实现多类型命名实体的自动抽取。实验结果证明,基于条件随机场的方志类古籍多类型命名实体自动识别模型发挥了较好的性能,其中别名、地名和引用名的识别率较显著,最高正确率分别达95.56%、98.28%和95.56%;人名和用途名的识别效果稍显不足,最高正确率分别为75%和74.04%,验证了条件随机场模型在方志类古籍多类型命名实体识别中的有效性。

Ancient local gazetteers(fangzhi)are important carriers of excellent Chinese culture with special characteristics.Named entity recognition is an important tool for text mining and utilization of ancient Chinese books,and which plays a positive role in understanding the contents of these books.Using digital local produces gazetteers as the research corpus,subject experts are invited to formulate the standard of corpus annotation and cross-checking method is adopted to annotate the corpus.Internal and external feature templates of these entities are then created.An automatic entity recognition model based on conditional random fields is designed to perform automatic extraction of multi-type named entities.Results of this experiment show that the multi-type named entity automatic recognition model for ancient local gazetteers based on conditional random fields has a good performance.The successful recognition rates of aliases,place names and citation books are significant,with the highest accuracy rates of 95.56%,98.28%and 95.56%respectively.The recognition of personal names and usage is slightly insufficient,with the highest accuracy rates of 75%and 74.04%respectively.To a certain extent,these results confirm the effectiveness of conditional random fields model in multi-type entity recognition of ancient local gazetteers.