Cumulative Path-Level Semantic Reasoning for Inductive Knowledge Graph Completion

TL;DR

This paper tackles inductive knowledge graph completion for emerging entities by introducing CPSR, a framework that jointly filters noisy structural information through a query-dependent masking module and captures long-range semantic dependencies via a global path-aware scoring mechanism. It combines path-level reasoning with Top-$k$ path-based entity embeddings and optimizes using a multi-class log-loss, reporting state-of-the-art results on eight inductive datasets. Key contributions include the design of a data-driven masking strategy, a path-global semantic scoring scheme, and an effective integration of structural and semantic information for inductive KGC. The approach improves robustness to noisy graphs and enhances the capture of long-range dependencies, with practical impact on dynamic knowledge graphs reliant on unseen entities and relations.

Abstract

Conventional Knowledge Graph Completion (KGC) methods aim to infer missing information in incomplete Knowledge Graphs (KGs) by leveraging existing information, which struggle to perform effectively in scenarios involving emerging entities. Inductive KGC methods can handle the emerging entities and relations in KGs, offering greater dynamic adaptability. While existing inductive KGC methods have achieved some success, they also face challenges, such as susceptibility to noisy structural information during reasoning and difficulty in capturing long-range dependencies in reasoning paths. To address these challenges, this paper proposes the Cumulative Path-Level Semantic Reasoning for inductive knowledge graph completion (CPSR) framework, which simultaneously captures both the structural and semantic information of KGs to enhance the inductive KGC task. Specifically, the proposed CPSR employs a query-dependent masking module to adaptively mask noisy structural information while retaining important information closely related to the targets. Additionally, CPSR introduces a global semantic scoring module that evaluates both the individual contributions and the collective impact of nodes along the reasoning path within KGs. The experimental results demonstrate that CPSR achieves state-of-the-art performance.

Figures (3)

• Figure 1: The motivation of the conventional inductive KGC (left) and the proposed CPSR (right). Conventional inductive KGCs traverse nodes directly over KGs. CPSR sequentially masks the noise structure and captures the long-range semantic dependencies over the whole reasoning path.
• Figure 2: Overview of CPSR. Specifically, the query-dependent masking module filters noisy structural information, while the global semantic scoring module captures long-range dependencies across the reasoning path. Finally, an MLP computes candidate scores. Here, m and n denote the number of paths and nodes per path, respectively.
• Figure 3: The performance with different $p_{e}$ values on WN18RR(V1) and FB15k237(V1).