Multi-view knowledge graph embedding for entity alignment
Proposed MultiKE, a new entity alignment framework that uses multi-view KG embedding. Existing embedding-based entity alignment methods have two limitations: 1. They only exploit one or two types of features of entities in KGs but they are many more 2. They rely heavily on abundant seed entity alignment as training data and this is not always accessible and costly MultiKE aims to solve this by embedding entities based on multiple views such as entity names, relations, attributes. The entity embeddings can be learned from each of this view and jointly optimised using different combination strategies (experimented with three) to improve alignment performance.
Investigate more feasible views such as entity types
Study cross-lingual entity alignment
Entity alignment between knowledge graphs using attribute embeddings
Proposed a new entity alignment framework that consists of a predicate alignment module, an embedding learning module, and an entity alignment module. The new framework aims to learn embeddings that can capture similarity between entities in DIFFERENT KGs. The novel embedding model first generates attribute embeddings from attribute triples and then use it to shift entity embeddings of two KGs into the same vector space. The paper also uses a transitivity rule to further enrich the number of attributes of an entity to enhance attribute character embedding.