# Lightweight Inference-Time Personalization for Frozen Knowledge Graph Embeddings

arXiv: https://arxiv.org/abs/2512.22398 · HTML: https://mlsanity.com/paper/2512.22398/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 6 of its 13 references that are in the map.
- First posted on arXiv 26 Dec 2025.

## Builds on (most cited first, 6 of 6)

- [KGAT: Knowledge Graph Attention Network for Recommendation](https://mlsanity.com/paper/1905.07854/) (2019)
- [RippleNet: Propagating User Preferences on the Knowledge Graph for Recommender Systems](https://mlsanity.com/paper/1803.03467/) (2018)
- [Probability Calibration for Knowledge Graph Embedding Models](https://mlsanity.com/paper/1912.10000/) (2019)
- [Explainable Reasoning over Knowledge Graphs for Recommendation](https://mlsanity.com/paper/1811.04540/) (2018)
- [Fast and Continual Knowledge Graph Embedding via Incremental LoRA](https://mlsanity.com/paper/2407.05705/) (2024)
- [In-n-Out: Calibrating Graph Neural Networks for Link Prediction](https://mlsanity.com/paper/2403.04605/) (2024)

## Abstract (arXiv)

Foundation models for knowledge graphs (KGs) achieve strong cohort-level performance in link prediction, yet fail to capture individual user preferences; a key disconnect between general relational reasoning and personalized ranking. We propose GatedBias, a lightweight inference-time personalization framework that adapts frozen KG embeddings to individual user contexts without retraining or compromising global accuracy. Our approach introduces structure-gated adaptation: profile-specific features combine with graph-derived binary gates to produce interpretable, per-entity biases, requiring only ~300 trainable parameters. We evaluate GatedBias on two benchmark datasets (Amazon-Book and Last-FM), demonstrating statistically significant improvements in alignment metrics while preserving cohort performance. Counterfactual perturbation experiments validate causal responsiveness; entities benefiting from specific preference signals show 6--30× greater rank improvements when those signals are boosted. These results show that personalized adaptation of foundation models can be both parameter-efficient and causally verifiable, bridging general knowledge representations with individual user needs.