多特征知识下的食品安全事件实体抽取研究

发布时间：2018-04-28 23:21

本文选题：特征知识 + 条件随机场模型　；参考：《数据分析与知识发现》2017年03期

【摘要】：【目的】从大规模食品安全事件当中抽取食品安全事件实体。【方法】基于已发生的食品安全事件,结合情报学数据获取、标注和组织的方法,融合食品安全事件实体的多种分布特征知识,通过条件随机场模型,构建食品安全事件语料并从中抽取相应的实体。【局限】在食品安全事件实体抽取过程中所制定的特征模板在领域化迁移上具有一定的局限性。【结果】在已有1 500万字经过标注的食品安全事件语料的规模上,通过统计食品安全事件实体的内部和外部特征,基于条件随机场机器学习模型,构建了食品安全实体的抽取模型,该模型最高的F值达到91.94%。【结论】通过对食品安全事件实体抽取结果的分析,在食品这一领域化的语料上,基于条件随机场进行实体抽取是可行的。
[Abstract]:[objective] to extract food safety event entities from large-scale food safety events. [methods] based on the food safety events that have occurred, the method of obtaining, labeling and organizing information science data is used. Integrating multiple distribution features of food safety event entities, using conditional random field model, To construct the food safety event corpus and extract the corresponding entity from it. [results] the feature template developed in the process of food safety event entity extraction has some limitations in domain migration. 1 5 million words annotated on the scale of food safety event corpus, Based on the internal and external characteristics of food safety event entities, the extraction model of food safety entities is constructed based on conditional random field machine learning model. The maximum F value of this model is 91.94. [conclusion] based on the analysis of the results of entity extraction of food safety events, it is feasible to extract entities based on conditional random field in the field of food.
【作者单位】：南京农业大学信息科学技术学院;南京农业大学领域知识关联研究中心;
【基金】：2011湖北省协同创新中心项目“面向应急推演平台的海量突发事件知识库与模型库构建研究”(编号:JD20150101) 国家自然科学基金项目“基于CSSCI的句法级汉英平行语料库构建及知识挖掘研究”(项目编号:71303120) 地震科技星火计划项目“面向地震应急的空间智能决策方法研究”(项目编号:HX15019)的研究成果之一
【分类号】：TP391.1

【参考文献】