KGRED: Knowledge-graph-based rule discovery for weakly supervised data labeling

Wenjun Hou, Liang Hong, Ziyi Zhu

Abstract: Highlights•Rule knowledge graph (KG) utilizes multi-dimensional semantic associations among rules to alleviate semantic drift in rule discovery.•Label-aware rule generation approach realizes attentive semantic information propagation based on rule KG to decrease misclassified rules.•Cross-attention-based semantic matching mechanism adaptively refines semantic information of sentences while enriching that of rules to reduce wrongly matched sentences.•Inconsistency-directed active learning strategy verifies inconsistent rules in rule generation and matching to improve the overall quality of rule discovery.