Improving Knowledge Graph Embeddings through Contrastive Learning with Negative Statements

2025-10-13

AI Summary

Most existing knowledge graph embedding methods operate under the closed-world assumption, treating all missing triples as false—contradicting the open-world reality where absence of evidence is not evidence of absence. Moreover, explicit negative statements (i.e., human-annotated false triples) are scarce and underutilized, hindering reliable discrimination among true, false, and unknown triples. To address this, we propose the first contrastive learning framework that systematically incorporates explicit negative statements. Our method employs a dual-model parallel adversarial training architecture: the two models alternately serve as teachers to generate high-confidence hard negative triples, augmented by a scoring-function-guided hard negative mining mechanism for joint optimization of positive and negative triples. This is the first approach to formally integrate negative knowledge into embedding training while strictly adhering to the open-world assumption. Experiments demonstrate significant improvements over state-of-the-art models on link prediction and triple classification across both general and domain-specific knowledge graphs, validating the effectiveness and generalizability of negative knowledge modeling.

Technology Category

• Data Mining & Knowledge Management: Linked Open Data, Knowledge Graphs & KB Completion
• Knowledge Representation and Reasoning: Knowledge Acquisition
• Machine Learning: Graph-based Machine Learning

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Graph embeddings and representation learning for Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Knowledge graphs represent information as structured triples and serve as the backbone for a wide range of applications, including question answering, link prediction, and recommendation systems. A prominent line of research for exploring knowledge graphs involves graph embedding methods, where entities and relations are represented in low-dimensional vector spaces that capture underlying semantics and structure. However, most existing methods rely on assumptions such as the Closed World Assumption or Local Closed World Assumption, treating missing triples as false. This contrasts with the Open World Assumption underlying many real-world knowledge graphs. Furthermore, while explicitly stated negative statements can help distinguish between false and unknown triples, they are rarely included in knowledge graphs and are often overlooked during embedding training. In this work, we introduce a novel approach that integrates explicitly declared negative statements into the knowledge embedding learning process. Our approach employs a dual-model architecture, where two embedding models are trained in parallel, one on positive statements and the other on negative statements. During training, each model generates negative samples by corrupting positive samples and selecting the most likely candidates as scored by the other model. The proposed approach is evaluated on both general-purpose and domain-specific knowledge graphs, with a focus on link prediction and triple classification tasks. The extensive experiments demonstrate that our approach improves predictive performance over state-of-the-art embedding models, demonstrating the value of integrating meaningful negative knowledge into embedding learning.

Problem

• Addresses limitations of knowledge graph embedding methods under Open World Assumption
• Incorporates explicitly declared negative statements into embedding learning process
• Improves predictive performance for link prediction and triple classification tasks

Innovation

• Integrates negative statements into embedding learning
• Uses dual-model architecture for positive and negative training
• Generates negative samples via cross-model scoring selection