Bacround: Link prediction is one of the most important research areas in the
analysis of various networks, including social networks, biological networks, and
knowledge graphs. This field seeks to identify missing relationships or predict
future relationships in graphs. Knowledge graph completion, as a sub-problem in
this field, deals with identifying and filling in the nodes and relationships that are
missing in the knowledge graph. This topic is of particular importance due to its
many applications in intelligent question-and-answer systems, intelligent search,
recommenders, and data management.
Aim: The objective of this research is to present an effective method for solving the
knowledge graph completion problem using deep learning methods. In this regard,
an attempt has been made to improve the performance of link prediction by utilizing
advanced deep learning architectures, such as encoder-decoder models, while also
helping to reduce computational complexity. The proposed model is capable of
embedding entities and relationships in low dimensions, thereby achieving higher
accuracy in detecting missing relationships in the knowledge graph.
Methodology: A combination of deep learning methods is used to solve the
knowledge graph completion problem. The proposed architecture includes a
converter-based encoder-decoder model that uses the attention mechanism to
understand the long-term dependencies between entities and relationships. In
addition, convolutional networks are used to extract local features from the entity
and relationship embedding vectors. This combination allows the model to exploit
both the advantages of the attention mechanism and the advantages of convolutional
networks. The datasets used in this study include FB15k-237, NELL-995, and
WN18RR, which are known as standard benchmarks in the field of knowledge
graph completion. To evaluate the proposed model, common metrics in this field
such as Hits@k, and Mean Reciprocal Rank (MRR) are used. These me