SGPT: A Generative Approach for SPARQL Query Generation From Natural Language Questions

SPARQL query generation from natural language questions is complex because it requires an understanding of both the question and underlying knowledge graph (KG) patterns. Most approaches are template-based, tailored to a specific require pipelines with multiple steps, including entity relation linking. Template-based also difficult adapt for new KGs manual efforts domain experts construct templates. To overcome this hurdle, we propose approach, dubbed SGPT, that combines benefits end-to-end modular systems leverages recent advances in large-scale models. Specifically, devise novel embedding technique can encode linguistic features which enables system learn In addition, training techniques allow implicitly employ graph-specific information (i.e., entities relations) into model’s parameters generate queries accurately. Finally, introduce strategy standard automatic metrics evaluating generation. A comprehensive evaluation demonstrates effectiveness SGPT over state-of-the-art methods across several benchmark datasets.