Self-Supervised Chinese Ontology Learning from Online Encyclopedias

Abstract EN

Constructing ontology manually is a time-consuming, error-prone,and tedious task.

We present SSCO, a self-supervised learningbased chinese ontology, which contains about 255 thousand concepts,5 million entities, and 40 million facts.

We explore the three largest onlineChinese encyclopedias for ontology learning and describe how totransfer the structured knowledge in encyclopedias, including article titles,category labels, redirection pages, taxonomy systems, and InfoBoxmodules, into ontological form.

In order to avoid the errors in encyclopediasand enrich the learnt ontology, we also apply some machinelearning based methods.

First, we proof that the self-supervised machinelearning method is practicable in Chinese relation extraction (at leastfor synonymy and hyponymy) statistically and experimentally and trainsome self-supervised models (SVMs and CRFs) for synonymy extraction,concept-subconcept relation extraction, and concept-instance relation extraction;the advantages of our methods are that all training examplesare automatically generated from the structural information of encyclopediasand a few general heuristic rules.

Finally, we evaluate SSCO intwo aspects, scale and precision; manual evaluation results show thatthe ontology has excellent precision, and high coverage is concluded bycomparing SSCO with other famous ontologies and knowledge bases; theexperiment results also indicate that the self-supervised models obviouslyenrich SSCO.