Paper accepted to IEEE VIS 2026

Jul 2026 Our full paper entitled “KeySI: An Interaction Framework for Tuning Text Embeddings Based on Human Feedback” has been accepted to IEEE VIS 2026.

KeySI enables feature-level feedback for tuning text embeddings through keyword-based concept specification. Instead of labeling individual documents, users organize extracted keywords into conceptual groups, which the system translates into document-level supervision for model tuning. We evaluate KeySI through a user study, usage scenarios, and quantitative experiments.

The paper is available on arXiv. Congratulations to all co-authors: Yan Zhu and Rebecca Faust!