Towards Automatic Construction of Knowledge Bases from Chinese Online Resources

Automatically constructing knowledge bases from online resources has become a crucial task in many research areas. Most existing knowledge bases are built from English resources, while few efforts have been made for other languages. Building knowledge bases for Chinese is of great importance on its own right. However, simply adapting existing tools from English to Chinese yields inferior results. In this paper, we propose to create Chinese knowledge bases from online resources with less human involvement. This project will be formulated in a self-supervised framework which requires little manual work to extract knowledge facts from online encyclopedia resources in a probabilistic view. In addition, this framework will be able to update the constructed knowledge base with knowledge facts extracted from up-to-date newswire. Currently, we have obtained encouraging results in our pilot experiments that extracting knowledge facts from infoboxes can achieve a high accuracy of around 95%, which will be then used as training data for the extraction of plain webpages.