Knowledge graph edits often displace correct answers in ranking lists
When Successful Knowledge Graph Edits Displace Correct Answers: Rank-Level Locality beyond Parameter Support
Artificial Intelligence
Summary
Making a change to a knowledge graph model to promote a certain answer can push other correct answers down or out of the list of results. The authors show that tests focused only on parameters directly affected by the edit miss this effect. They introduce new ways to check for these unintended side effects across several types of related answers. They also find that while some editing methods can keep the promoted answer high without causing harm, those cases are rare. This work highlights the importance of reporting how editing a knowledge graph affects not just one answer but the overall ranking of correct answers.
What this means in practice
- •For data engineers: Improve knowledge graph update tools by measuring and controlling harmful rank shifts when editing models for corrected facts.
- •For search algorithm developers: Enhance ranking robustness by auditing how targeted knowledge graph edits affect the positions of other correct results in response lists.
Authors
Yi-Cheng Lai, Jerry Wang, Hsin-Ling Hsu, Li-Chu Chi, Ya-Wen Teng, Hen-Hsen Huang
Abstract
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. On FB15k-237 with DistMult and ComplEx, direct promotion always moves the target into the top ten, but does so without damage in only 23.0--23.2\% of edits. Strict preservation causes no measured damage, yet succeeds in only 1.3--1.4\%. Support-regularized entity editing gives the highest joint success, 36.3--37.7\%, while rank-truncated preservation reaches 32.8--34.7\% and reduces the mean number of displaced answers from about 14 to 1.2. Experiments across dimensions, scorers, ranking conventions, and a learned editor show that locality depends on both the protected scope and the editing mechanism. KGE editing should therefore report correction success together with the incidence and severity of rank displacement.