| 摘 要: 互联网中的HTML表格蕴含着丰富的结构化或半结构化知识，是知识库构建与扩充的重要数据资源。然 而如何对HTML表格进行正确解析并获得三元组知识用于扩充知识库，则是一个很有挑战的问题。首先，HTML表格的 结构各有不同。其次，表格与知识库中的实体和属性的表示不同，需要统一，即实体链接与属性对齐。本文首先提出了 一个基于知识库的在线百科表格解析与知识融合框架，该框架可针对不同类别的表格进行知识抽取；并提出了基于知识 库的表格实体链接和属性对齐方法，用以将表格中的知识与知识库进行匹配与融合。实验使用了126万在线百科表格数 据为CN-DBpedia扩充约1000万三元组。 |
| 关键词: HTML表格；知识抽取；知识融合 |
|
中图分类号: TP391
文献标识码: A
|
| 基金项目: 国家自然科学基金(No.61632016,61572336,61572335,61772356,61872258)；江苏省高校自然科学研究项目(No.17KJA520003,18KJA520010). |
|
| Knowledge Extraction and Fusion over Online Encyclopedia Tables for Knowledge Base Augmentation |
|
SONG Xiaozhao,ZHENG Xin,LI Zhixu,XU Jiajie1,2
|
1.( 1.School of Computer Science&Technology, Soochow University, Suzhou 215006, China;2.
2.Suzhou Institute, iFLYTEK, Suzhou 215000, China)
|
| Abstract: HTML tables in WWW have been flooded with (semi-)structured knowledge,which is an important source for knowledge base augmentation.However,it is a challenging problem to parse and extract triples in a correct way for knowledge base augmentation.Firstly,HTML tables have different types.Secondly,the descriptions of entities and attributes in different tables may be inconsistent with knowledge base,which needs to be matched and fused,i.e.,entity linking and property alignment.This paper first designs a table parse and knowledge fusion framework for the knowledge base,which is able to parse and extract knowledge in different types of tables.Additionally,an entity linking and property alignment method is proposed based on the knowledge base,to match and fuse the RDF triples with knowledge base.1.26 million tables in online encyclopedias are used in the experiment to augment 10 million triples for CN-DBpedia. |
| Keywords: HTML table;knowledge extraction;knowledge fusion |
