基于条件随机场多特征融合的中文地名、机构名实体识别  

Chinese Named Entity Recognition Based on Conditional Random Fields Multi-Feature Fusion

作　　者：马孟铖 艾斯卡尔.艾木都拉[1] 吐尔地.托合提 MA Meng-cheng;Askar Hamdulla;Turdi Tohti(School of Information Science and Engineering,Xinjiang University,Urumqi 830046)

机构地区：[1]新疆大学信息科学与工程学院,乌鲁木齐830046

出　　处：《现代计算机》2019年第12期13-17,共5页Modern Computer

基　　金：国家自然科学基金资助项目(No.61562083);国家重点研发计划(No.2017YFC0820603)

摘　　要：命名实体识别是自然语言处理的基础任务,其中通用领域语料下的复杂地名、机构名实体识别效果还有待提高。针对这一问题,提出一种多特征融合的中文地名、机构名实体识别方法,该方法以条件随机场为框架,结合实验语料的特点,选取统计特征,将局部特征、复合特征与规则知识库相融合,对中文语料进行命名实体识别。实验结果表明,1998年1月《人民日报》语料上的测试结果,地名实体的F1值提高2.2%,达到97.70%,机构名实体的F1值提高6.2%,达到92.80%。Named entity recognition is a basic task of natural language processing, among which the recognition effect of complex place names and in-stitutional name entities in common domain corpus needs to be improved. To solve this problem, proposes a multi-feature fusion method for entity recognition of Chinese place names and institutions. This method takes conditional random field as the framework, combines the char-acteristics of experimental corpus, selects statistical features, integrates local features and compound features with rule knowledge base, and identifies named entities of Chinese corpus. The experimental results show that the F1 value of place-name entities are increased by 2.2% to 97.70% in the test results on the corpus of People's Daily in January 1998. The F1 value for institutional entity names are in-creased by 6.2% to 92.80%.

关 键 词：命名实体 多特征融合 自然语言处理 条件随机场 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]