When Successful Knowledge Graph Edits Displace Correct Answers: Rank-Displacement Audit for KGE Models

Forum topic · 小凯 · 2026-09-15

Summary

A new paper on arXiv (2609.12116) by Yi-Cheng Lai, Jerry Wang, Hsin-Ling Hsu, Li-Chu Chi, Ya-Wen Teng, and Hen-Hsen Huang introduces a rank-displacement audit for editing knowledge graph embedding (KGE) models. Editing a KGE model to promote a desired answer can silently push other correct answers out of the returned list, and locality tests based only on facts reusing the edited parameter can miss this ranking effect. The authors audit three scopes: facts supported by the edited parameter, other correct answers to the target query, and correct answers across queries sharing the same relation, and derive dimensional and geometric conditions for updates that improve the target while exactly preserving selected scores. On FB15k-237 with DistMult and ComplEx, direct promotion always reached the top ten but was damage-free in only 23.0-23.2% of edits; strict preservation caused no damage but succeeded only 1.3-1.4% of the time; regularized entity editing achieved the highest joint success rate of 36.3-37.7%. The paper recommends that KGE editing report both correction success and rank-displacement incidence and severity.

Paper Overview

Field: Machine Learning Authors: Yi-Cheng Lai, Jerry Wang, Hsin-Ling Hsu, Li-Chu Chi, Ya-Wen Teng, Hen-Hsen Huang Published: 2026-09-15 arXiv: 2609.12116

Key Findings

Editing a knowledge graph embedding (KGE) model to promote a desired answer can displace correct answers from the returned list. Locality tests based only on facts that reuse the edited parameter can miss this ranking effect.
The authors introduce a common rank-displacement audit at three scopes: facts supported by the edited parameter, other correct answers to the target query, and correct answers across queries with the same relation.
They derive dimensional and geometric conditions for an update to improve the target while exactly preserving selected scores.

Experimental Results (FB15k-237, DistMult and ComplEx)

Direct promotion: always moves the target into the top ten, but only 23.0–23.2% of edits are damage-free.
Strict preservation: causes no measured damage, yet succeeds in only 1.3–1.4% of cases.
Entity editing with regularization support: achieves the highest joint success rate of 36.3–37.7%.
Truncated rank preservation: reaches 32.8–34.7%, reducing the average number of displaced answers from about 14 to 1.2.

Implications

Experiments across dimensions, scorers, ranking conventions, and a learned editor show that locality depends simultaneously on the protected scope and the editing mechanism. KGE editing should therefore report both correction success rate and the incidence and severity of rank displacement.

Original Abstract (excerpt)

> Editing a knowledge graph embedding (KGE) model to promote a desired answer can displace correct answers from the returned list. Locality tests based only on facts that reuse the edited parameter can miss this ranking effect. We introduce a common rank-displacement audit at three scopes: facts supported by the edited parameter, other correct answers to the target query, and correct answers across queries with the same relation. We also derive dimensional and geometric conditions for an update to improve the target while exactly preserving selected scores...

--- *Auto-collected on 2026-09-15*