基于Bootstrapping的本体标注方法

实现语义Web目标的一个重要前提是利用本体词汇标注Web资源。为此,提出一种基于弱监督(Bootstrapping)的本体标注方法。对给定的本体进行解析,生成规则文件,通过文本分类筛选出领域文档。采用Bootstrapping的方法进行信息标注抽取和本体推理,经过几次循环后,只利用少量的训练文本就能达到较好的标注效果。实验证明,该方法实体识别准确率高,标注效果好。

Ontology Annotation Method Based on Bootstrapping

A very important premise to achieve the Semantic Web aim is using ontology vocabulary to annotate Web resource. This paper presents a new annotation method which is based on Bootstrapping and rules. Ontology should be parsed to generate the rule files, and domain documents should be filtered by text classifier. Several loops using Bootstrapping to annotate and extract information and infer ontology are maked, so a good annotation result can be achieved by a few straining documents. Experiments show that the method with high recognition rate is effective.