A Pluggable Common Sense-Enhanced Framework for Knowledge Graph Completion

TL;DR

This work proposes a pluggable common sense-enhanced KGC framework that incorporates both fact and common sense for KGC, and introduces common sense-guided negative sampling and a coarse-to-fine inference approach for KGs with rich entity concepts.

Abstract

Knowledge graph completion (KGC) tasks aim to infer missing facts in a knowledge graph (KG) for many knowledge-intensive applications. However, existing embedding-based KGC approaches primarily rely on factual triples, potentially leading to outcomes inconsistent with common sense. Besides, generating explicit common sense is often impractical or costly for a KG. To address these challenges, we propose a pluggable common sense-enhanced KGC framework that incorporates both fact and common sense for KGC. This framework is adaptable to different KGs based on their entity concept richness and has the capability to automatically generate explicit or implicit common sense from factual triples. Furthermore, we introduce common sense-guided negative sampling and a coarse-to-fine inference approach for KGs with rich entity concepts. For KGs without concepts, we propose a dual scoring scheme involving a relation-aware concept embedding mechanism. Importantly, our approach can be integrated as a pluggable module for many knowledge graph embedding (KGE) models, facilitating joint common sense and fact-driven training and inference. The experiments illustrate that our framework exhibits good scalability and outperforms existing models across various KGC tasks.

Figures (7)

• Figure 1: Illustration of the KGC task and three main challenges of the KGE technique. The top half shows a KG with triples and ontological concepts linked to entities as well as a KGC task of tail entity missing $(David, Nationality, ?)$. This figure highlights three challenges of KGE models on KGC tasks: (1) Lack of ready-made common sense that can be directly used for entity-specific KGC tasks. (2) Invalid negative sampling during the training procedure of a KGE model which would limit its performance. (3) Incorrect results of KGC due to the uncertainty of KGE at the inference stage.
• Figure 2: The brief structure of our proposed framework. The upper part is the explicit common sense-enhanced model, containing a simple yet effective automatic common sense generation module that could produce valuable explicit common sense and facilitate common sense-guided negative sampling as well as coarse-to-fine inference. Contrarily, the lower part exhibits the implicit common sense-enhanced model. Particularly, it introduces relation-aware concept embeddings for representing implicit common sense, and then conducts joint inference in the view of both common sense and fact.
• Figure 3: Illustration of the common sense generation module. Each entity is linked by a corresponding concept (in yellow) in the KG. Meanwhile, entities belonging to the same concept are shown in the same color. Then, the explicit common sense containing concepts with their relations could be generated and derived from the KG by an entity-to-concept converter.
• Figure 4: A case of common sense-guided negative sampling procedure specific to an N-1 relation $BirthPlace$. The common sense in set-form corresponding to the positive triple is presented. Besides, the arrow represents the direction of increasing value. The darker colors of negative triple candidates indicate larger weights for training KGE models.
• Figure 5: An illustration of factual triples along with their corresponding implicit common-sense triples. A factual triple comprises an explicit entity pair connected by a relation while a common sense triple involves implicit entity concepts along with an explicit relation. It is worth noting that The entity concepts in parentheses are intended to convey potential meanings rather than real textual representations.

Theorems & Definitions (10)

• proof
• proof
• proof
• proof
• proof
• proof
• proof
• proof
• proof
• proof