Dual Graph Embedding for Object-Tag LinkPrediction on the Knowledge Graph

TL;DR

This paper introduces Dual Graph Embedding (DGE), a novel method that models both first-order and high-order proximities in knowledge graphs to improve object-tag link prediction, outperforming existing methods.

Contribution

The paper proposes a dual graph embedding approach using auto-encoding to explicitly capture high-order proximities for better link prediction in knowledge graphs.

Findings

DGE outperforms state-of-the-art methods on three real-world datasets.

Explicit modeling of high-order proximities improves link prediction accuracy.

Dual graph approach effectively captures complex entity relationships.

Abstract

Knowledge graphs (KGs) composed of users, objects, and tags are widely used in web applications ranging from E-commerce, social media sites to news portals. This paper concentrates on an attractive application which aims to predict the object-tag links in the KG for better tag recommendation and object explanation. When predicting the object-tag links, both the first-order and high-order proximities between entities in the KG propagate essential similarity information for better prediction. Most existing methods focus on preserving the first-order proximity between entities in the KG. However, they cannot capture the high-order proximities in an explicit way, and the adopted margin-based criterion cannot measure the first-order proximity on the global structure accurately. In this paper, we propose a novel approach named Dual Graph Embedding (DGE) that models both the first-order and high-order proximities in the KG via an auto-encoding architecture to facilitate better object-tag relation inference. Here the dual graphs contain an object graph and a tag graph that explicitly depict the high-order object-object and tag-tag proximities in the KG. The dual graph encoder in DGE then encodes these high-order proximities in the dual graphs into entity embeddings. The decoder formulates a skip-gram objective that maximizes the first-order proximity between observed object-tag pairs over the global proximity structure. With the supervision of the decoder, the embeddings derived by the encoder will be refined to capture both the first-order and high-order proximities in the KG for better link prediction. Extensive experiments on three real-world datasets demonstrate that DGE outperforms the state-of-the-art methods.