Missing relation prediction in knowledge graph using local and neighbour aware entity embedding

Predicting relations involves deducing the absent connections between entities within a knowledge graph. Knowledge graph embedding has recently emerged as a technique to represent entities and their relationships in a condensed vector space. While effective for entity prediction tasks, this approach escalates in complexity, particularly for extensive knowledge graphs. This study introduces a new method for relation prediction in knowledge graphs, employing a mechanism termed Local and Neighbor Aware Entity Embedding (LNAEE). LNAEE utilizes skip-gram initialization to capture local entity features at the triple level. It then employs an attention mechanism to update the central entity's features based on its immediate neighbors. Finally, a multilayer neural network predicts the most likely relationship type between the given entity pair. LNAEE is streamlined and less intricate owing to the reduced relation search space. Benchmark dataset experiments demonstrate that LNAEE outperforms the baselines. This makes LNAEE ideal for applications requiring a lightweight and affordable model for knowledge graph enrichment.