Multilingual Knowledge Graph Completion via Efficient Multilingual Knowledge Sharing

TL;DR

This work tackles MKGC by exploiting multilingual shared knowledge through a dual-component framework: KL-GMoE for efficient knowledge-level routing across language groups, and Iterative Entity Reranking (IER) to progressively refine entity predictions using cross-lingual signals. The authors construct a multilingual Wikidata5M-derived dataset and tailor a prompt-based training regime to integrate KGE candidates with LLM decoding. Empirical results show substantial gains over SOTA MKGC methods in Hits@1, Hits@3, Hits@10, and MRR, with robust performance under language imbalance and strong generalization to unseen languages. The approach reduces parameter and computation demands relative to competing methods, offering practical benefits for scalable, multilingual knowledge completion.

Abstract

Large language models (LLMs) based Multilingual Knowledge Graph Completion (MKGC) aim to predict missing facts by leveraging LLMs' multilingual understanding capabilities, improving the completeness of multilingual knowledge graphs (KGs). However, existing MKGC research underutilizes the multilingual capabilities of LLMs and ignores the shareability of cross-lingual knowledge. In this paper, we propose a novel MKGC framework that leverages multilingual shared knowledge to significantly enhance performance through two components: Knowledge-level Grouped Mixture of Experts (KL-GMoE) and Iterative Entity Reranking (IER). KL-GMoE efficiently models shared knowledge, while IER significantly enhances its utilization. To evaluate our framework, we constructed a mKG dataset containing 5 languages and conducted comprehensive comparative experiments with existing state-of-the-art (SOTA) MKGC method. The experimental results demonstrate that our framework achieves improvements of 5.47%, 3.27%, and 1.01% in the Hits@1, Hits@3, and Hits@10 metrics, respectively, compared with SOTA MKGC method. Further experimental analysis revealed the properties of knowledge sharing in settings of unseen and unbalanced languages. We have released the dataset and code for our work on https://github.com/gaoxiaofei07/KL-GMoE.

Figures (12)

• Figure 1: This figure depicts the problems encountered when applying LLMs directly to the MKGC task. (a) illustrates that existing PEFT is not suitable for the MKGC task. (b) Indicates a discrepancy between the task paradigms LLMs excel at and MKGC tasks.
• Figure 2: The figure illustrates our proposed framework. Figure (a) depicts the architecture and workflow of the KL-GMoE, where the matrices $A_{i}$ and $B_{i,j}$ highlighted in red represent the currently activated expert. Figure (b) illustrates the workflow of the IER method. After $N_t$ iterations, we can obtain a reranked list of entities.
• Figure 3: This figure shows the variation in Hits@1 scores of our framework under training data settings with five different language proportions.
• Figure 4: The figure illustrates the Hits@1 performance of our method on five languages under three different training language settings.
• Figure 5: The figure illustrates the impact of the number of iterations of the IER method on performance.