Dual Graph Embedding for Object-Tag LinkPrediction on the Knowledge Graph
This work addresses tag recommendation and object explanation in web applications such as E-commerce and social media, but it is incremental as it builds on existing graph embedding techniques by explicitly incorporating high-order proximities.
The paper tackles the problem of predicting object-tag links in knowledge graphs for applications like tag recommendation and object explanation, proposing a Dual Graph Embedding (DGE) method that models both first-order and high-order proximities, and it outperforms state-of-the-art methods in experiments on three real-world datasets.
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.