Skip to content
Book Open access

Self-Evolving Ontology via Cumulative Definition Enrichment for Entity Canonicalization

Nov 2026 · Proceedings of the 35th ACM International Conference on Information and Knowledge Management · 0 citations · 33 references

Abstract

Ontology grounding pipelines rely on expert-curated vocabularies, but citizens describe the same facilities in their own words. The Korean Safety e-Report platform, which routes over one million citizen complaints annually to responsible agencies, suffers from exactly this problem: its keyword-matching pipeline fails to recognize many citizen expressions, causing misrouted reports and delayed field response. We present Cumulative Definition Enrichment (CDE), which continually discovers synonym expressions from incoming complaints and incorporates them into entity definitions, producing embeddings that improve as the ontology absorbs real usage. Experts design the ontology structure, while CDE learns the vocabulary from the complaint stream. We define two evaluation metrics: Direct Link Rate (DLR), the fraction of complaints resolved through automatic ontology grounding, and Qualified DLR (Q-DLR), which combines DLR with synonym mapping accuracy. DLR alone can be misleading because adding incorrect synonyms still increases coverage; Q-DLR captures whether the added synonyms are actually correct. On 100K real complaints across two ontology versions and two initialization settings, CDE achieves the highest Q-DLR in every condition. Scalability experiments on 600K complaints confirm consistent performance. All processing runs on-premise with limited GPU resources under strict data sensitivity constraints, and the system is planned for government integration starting in 2027.

Read PDF

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.