Sequential Semantic Knowledge Graph Embedding

Knowledge graph embedding is aimed at representing entities and relations of knowledge graph in a low-dimensional continuous vector space. Previous embedding models pay little attention to the sequential semantic information in triples and as a result, may lead to the semantic drift problem. Towards this end, we propose a novel sequential semantic embedding (SeqSemE) model to address this problem in this paper. Firstly, we utilize a sequential language model to capture sequential information of triples and interactions between entities and relations. Secondly, we propose a method of learning two embeddings for each relation to avoid semantic drift. Extensive experiments on link prediction show that our SeqSemE is efficient and effective. It can obtain better performance than previous state-of-the-art embedding models.