# Half a Link can Be Enough to Predict a Whole Link: Understanding Generalization in Knowledge Graph Foundation Models

arXiv: https://arxiv.org/abs/2606.18001 · HTML: https://mlsanity.com/paper/2606.18001/

## Key facts

- Cited by 1 papers in the map (data through 14 Oct 2026).
- Builds on 23 of its 36 references that are in the map.
- First posted on arXiv 16 Jun 2026.

## Builds on (most cited first, 23 of 23)

- [Complex Embeddings for Simple Link Prediction](https://mlsanity.com/paper/1606.06357/) (2016)
- [RotatE: Knowledge Graph Embedding by Relational Rotation in Complex Space](https://mlsanity.com/paper/1902.10197/) (2019)
- [Convolutional 2D Knowledge Graph Embeddings](https://mlsanity.com/paper/1707.01476/) (2017)
- [Neural Bellman-Ford Networks: A General Graph Neural Network Framework for Link Prediction](https://mlsanity.com/paper/2106.06935/) (2021)
- [DeepPath: A Reinforcement Learning Method for Knowledge Graph Reasoning](https://mlsanity.com/paper/1707.06690/) (2017)
- [Inductive Relation Prediction by Subgraph Reasoning](https://mlsanity.com/paper/1911.06962/) (2019)
- [Knowledge Transfer for Out-of-Knowledge-Base Entities: A Graph Neural Network Approach](https://mlsanity.com/paper/1706.05674/) (2017)
- [CoDEx: A Comprehensive Knowledge Graph Completion Benchmark](https://mlsanity.com/paper/2009.07810/) (2020)
- [Knowledge Graph Reasoning with Relational Digraph](https://mlsanity.com/paper/2108.06040/) (2021)
- [Commonsense Knowledge Base Completion with Structural and Semantic Context](https://mlsanity.com/paper/1910.02915/) (2019)
- [InGram: Inductive Knowledge Graph Embedding via Relation Graphs](https://mlsanity.com/paper/2305.19987/) (2023)
- [A*Net: A Scalable Path-based Reasoning Approach for Knowledge Graphs](https://mlsanity.com/paper/2206.04798/) (2022)
- [Relation Prediction as an Auxiliary Training Objective for Improving Multi-Relational Graph Representations](https://mlsanity.com/paper/2110.02834/) (2021)
- [Improving Knowledge Graph Embedding Using Simple Constraints](https://mlsanity.com/paper/1805.02408/) (2018)
- [Dynamic Anticipation and Completion for Multi-Hop Reasoning over Sparse Knowledge Graph](https://mlsanity.com/paper/2010.01899/) (2020)
- [An Open Challenge for Inductive Link Prediction on Knowledge Graphs](https://mlsanity.com/paper/2203.01520/) (2022)
- [A Multi-Task Perspective for Link Prediction with New Relation Types and Nodes](https://mlsanity.com/paper/2307.06046/) (2023)
- [TRIX: A More Expressive Model for Zero-shot Domain Transfer in Knowledge Graphs](https://mlsanity.com/paper/2502.19512/) (2025)
- [Saga: A Platform for Continuous Construction and Serving of Knowledge At Scale](https://mlsanity.com/paper/2204.07309/) (2022)
- [Link Prediction with Attention Applied on Multiple Knowledge Graph Embedding Models](https://mlsanity.com/paper/2302.06229/) (2023)
- [R-GCN: The R Could Stand for Random](https://mlsanity.com/paper/2203.02424/) (2022)
- [Flock: A Knowledge Graph Foundation Model via Learning on Random Walks](https://mlsanity.com/paper/2510.01510/) (2025)
- [SEMMA: A Semantic Aware Knowledge Graph Foundation Model](https://mlsanity.com/paper/2505.20422/) (2025)

## Built on by (most cited first, 1 of 1)

- [MERIT: Scaling KG Foundation Models with Relation Transformers](https://mlsanity.com/paper/2610.15638/) (2026)

## Abstract (arXiv)

Knowledge graph (KG) foundation models (KGFMs) are zero-shot generalizers: trained once, they can predict links on unseen graphs without retraining. However, understanding when and how they can robustly generalize across KGs is still an open question. In this paper, we shed some light on their generalization mechanisms highlighting how their performance on unseen KGs is not uniform when it comes to partially seen links, which we call half-links. In fact, we show that to predict a test triple (h,r,t) it might suffice in practice to have observed the half-link (h,r) or (r,t) in the inference graph. This yields a taxonomy of four scenarios when combinations of these half-links are observed or not. In a rigorous stratified analysis over these scenarios, we reveal that SoTA KGFMs use seen half links for predictions, while unseen half-links pose different challenges. As such, our finer-grained taxonomy can be a diagnostic protocol for robust KGFM generalization and highlights where novel KGFMs can improve.