Uncertainty-Aware Relational Graph Neural Network for Few-Shot Knowledge Graph Completion

TL;DR

Uncertainty-Aware FKGC addresses few-shot knowledge graph completion by modeling entities and relations as Gaussian distributions with means and variances. It introduces an uncertainty-aware relational GNN (UR-GNN) to propagate Gaussian features and employs variance-based attention, followed by multi-sample augmentation and a joint loss that combines completion, uncertainty mutual information, and KL regularization. The approach achieves state-of-the-art results on NELL and Wiki 5-shot tasks, demonstrating improved robustness to noise and limited data. This work provides a principled framework for incorporating uncertainty into FKGC, with implications for more reliable knowledge graph reasoning under data scarcity.

Abstract

Few-shot knowledge graph completion (FKGC) aims to query the unseen facts of a relation given its few-shot reference entity pairs. The side effect of noises due to the uncertainty of entities and triples may limit the few-shot learning, but existing FKGC works neglect such uncertainty, which leads them more susceptible to limited reference samples with noises. In this paper, we propose a novel uncertainty-aware few-shot KG completion framework (UFKGC) to model uncertainty for a better understanding of the limited data by learning representations under Gaussian distribution. Uncertainty representation is first designed for estimating the uncertainty scope of the entity pairs after transferring feature representations into a Gaussian distribution. Further, to better integrate the neighbors with uncertainty characteristics for entity features, we design an uncertainty-aware relational graph neural network (UR-GNN) to conduct convolution operations between the Gaussian distributions. Then, multiple random samplings are conducted for reference triples within the Gaussian distribution to generate smooth reference representations during the optimization. The final completion score for each query instance is measured by the designed uncertainty optimization to make our approach more robust to the noises in few-shot scenarios. Experimental results show that our approach achieves excellent performance on two benchmark datasets compared to its competitors.

Figures (5)

• Figure 1: An example of the few-shot KG completion task. (a) The task is to predict tail entities on KG. (b) The task utilizes only $K$ triples for the specific relation to predicting others.
• Figure 2: The framework of UFKGC. It contains three modules: Uncertainty Representation, UR-GNN, and Uncertainty Optimization. (a) The Uncertainty Representation module is used to convert the deterministic embedding into an uncertain embedding and evaluate the scope of uncertainty. (b) The UR-GNN is to learn embeddings in the uncertain space, and (c) Uncertainty Optimization is to evaluate the completion prediction.
• Figure 3: Discussions for different few-shot situations, compared to the partial outstanding few-shot completion baselines.
• Figure 4: Impacts of the UR-GNN layer on the NELL and Wiki dataset for the 5-shot situation.
• Figure 5: Impacts of the number of random samples on the NELL and Wiki dataset for the 5-shot situation.