Abstract:Methods for knowledge graph (KG) completion need to capture semantic regularities and use these regularities to infer plausible knowledge that is not explicitly stated. Most embedding-based methods are opaque in the kinds of regularities they can capture, although region-based KG embedding models have emerged as a more transparent alternative. By modeling relations as geometric regions in high-dimensional vector spaces, such models can explicitly capture semantic regularities in terms of the spatial arrangement of these regions. Unfortunately, existing region-based approaches are severely limited in the kinds of rules they can capture. We argue that this limitation arises because the considered regions are defined as the Cartesian product of two-dimensional regions. As an alternative, in this paper, we propose RESHUFFLE, a simple model based on ordering constraints that can faithfully capture a much larger class of rule bases than existing approaches. Moreover, the embeddings in our framework can be learned by a monotonic Graph Neural Network (GNN), which effectively acts as a differentiable rule base. This approach has the important advantage that embeddings can be easily updated as new knowledge is added to the KG. At the same time, since the resulting representations can be used similarly to standard KG embeddings, our approach is significantly more efficient than existing approaches to differentiable reasoning.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is how to capture and utilize semantic rules more effectively in Knowledge Graph (KG) completion tasks. Specifically, the existing region - based embedding models have limitations in capturing certain types of rules, especially when these rules involve closed - path rule sets. The author points out that the existing methods are mainly limited by the fact that they use the Cartesian product of two - dimensional regions to define relation regions, which restricts the types of rules that can be captured. To solve this problem, the paper proposes a new model named RESHUFFLE. By introducing a simple model based on ranking constraints, this model can more accurately capture a larger class of rule bases, thus overcoming some limitations of the existing region models. In addition, the embeddings in the RESHUFFLE model can be learned through Monotonic Graph Neural Network (GNN). This not only enables the model to efficiently update the embeddings when new knowledge is added, but also retains the efficiency advantage of standard KG embeddings, that is, the rationality of triples can be checked almost instantaneously. The main contributions of the paper are as follows: 1. Propose a new model. Although simple, this model can capture a large number of rule bases. In particular, when considering the conclusions that can be deduced using a limited number of reasoning steps, it can faithfully capture any set of closed - path rules. 2. Demonstrate how to use Monotonic GNN to learn KG embeddings. This GNN actually acts as an approximation of a differentiable rule base, acting on the initial representations of entities to ensure that they can capture the conclusions deduced from the KG. This enables KG embeddings to be efficiently updated when new knowledge is available, which is especially suitable for KG completion tasks in an inductive setting where links between entities not seen during training need to be predicted.

Differentiable Reasoning about Knowledge Graphs with Region-based Graph Neural Networks

Neural Axiom Network for Knowledge Graph Reasoning

Knowledge Reasoning via Jointly Modeling Knowledge Graphs and Soft Rules

River of No Return: Graph Percolation Embeddings for Efficient Knowledge Graph Reasoning

Iteratively Learning Embeddings and Rules for Knowledge Graph Reasoning

A Hybrid Model for Learning Embeddings and Logical Rules Simultaneously from Knowledge Graphs

Knowledge Graph Reasoning with Logics and Embeddings: Survey and Perspective

Knowledge Graph Reasoning with Relational Digraph

Neural Probabilistic Logic Learning for Knowledge Graph Reasoning

Explainable Sparse Knowledge Graph Completion Via High-order Graph Reasoning Network

ExKGR: Explainable Multi-hop Reasoning for Evolving Knowledge Graph

Relgraph: A Multi-Relational Graph Neural Network Framework for Knowledge Graph Reasoning Based on Relation Graph

From Concept to Instance: Hierarchical Reinforced Knowledge Graph Reasoning

Rule-enhanced Iterative Complementation for Knowledge Graph Reasoning

Location Sensitive Embedding for Knowledge Graph Reasoning

Generating Graph-Based Rules for Enhancing Logical Reasoning

Understanding Expressivity of Neural KG Reasoning from Rule Structure Learning

RulE: Neural-Symbolic Knowledge Graph Reasoning with Rule Embedding

Attention-based Exploitation and Exploration Strategy for Multi-Hop Knowledge Graph Reasoning

Reasoning Through Memorization: Nearest Neighbor Knowledge Graph Embeddings

EM-RBR: a reinforced framework for knowledge graph completion from reasoning perspective