Answers are grounded in this paper and cite their sources.
Although the sequence to sequence with attention and copy mechanism model is a mature and excellent natural language generation framework. Automatic natural language description generation for entities in the structured knowledge graph is still an open challenging task. Firstly, the sequence to sequence model takes the knowledge graph as a sequence input, so it cannot efficiently learn the structural features and the multi-relational properties of the knowledge graph. Secondly, the generated natural language descriptions should be able to correctly express the positional relationship between the attributes of the entity. In order to address the above two problems, a novel knowledge graph embedding with Dual Position Attention mechanism based Natural Language Generation model (referred to as DPANLG) is proposed. In it, we first leverage the graph attention neural network that have been proven to be efficient and effective methods for capturing the structural features and the multi-relational properties of the knowledge graph. Then, a dual position attention mechanism is employed to deal with the inter-dependency attributes of an entity. Finally, the generated text undergoes further refinement aided by large language models and elaborately designed prompts. We perform empirical experiments on two benchmark datasets, and the results demonstrate that the proposed DPANLG model achieves promising results and outperforms state-of-the-art approaches.
No takes yet. Share an insight, caveat, or question.
Li et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: