Research On Instance Extraction Technology In Chinese Linking Open Data

Constructing knowledge bases is an essential and important part in the development of semantic web. Recently, researches on the knowledge base construction are more and more popular. At the same time, the instance extraction technologies play an important role as well, which are used for linking instances with categories in knowledge bases. However, the existing approaches of extracting instances have some drawbacks, which either are language dependent, like in YAGO, or only can extract limited general categories, like in DBpedia. In addition, the data source of extracting instances includes unstructured Internet resource and structured LOD. While the Internet resource contains tons of noisy data, the content of LOD is useful. Besides, there is little Chinese instance data in current knowledge bases. In a word, the instance extraction research on Chinese LOD is valuable and helpful for the more complex situation of Internet resource.Based on those described above, technologies on instance extraction are studied deeply and a novel approach to extracting instances from Chinese LOD is proposed. The contributions are described as follow:(1) The instance extraction algorithm, which is based on entity attributes and category attributes, generates lots of high qualified InstanceOf relationship triples.(2) The attribute propagation algorithm makes more than 42 thousand categories from Chinese encyclopedia online generate attributes.(3) The evaluation of generated category attributes and InstanceOf triples shows that their correct rates are more than 88% and 91% respectively. What's more, compared with knowledge base DBpedia, YAGO and BabelNet, the data set obtained in (1) and (2) contains more Chinese InstanceOf relationship triple with more appropriate granularity.