Answers are grounded in this paper and cite their sources.
In response to issues such as weak generalization of extraction models and missing information in abstraction models in automatic text summarization, takes the patent specification and its abstract as the research object, a patent text abstract automatic generation model based on "extraction+abstraction", PatentSNU, is constructed. Combining the extractive summary model based on improved SummaRuNNer model with the abstractive summary model based on NEZHA-UniLM model, the output of the extraction model is taken as the input of the abstraction model, and the results of the extraction model are optimized by the abstraction model, and then get an accurate and concise patent specification abstract.The experimental results show that the content of the patent specification abstract generated by thePatentSNU model is comprehensive, the language is fluent, and it is closer to the reference abstract.At the same time, using the ROUGE score as the evaluation index, the PatentSNU model has a significant improvement in the ROUGE score compared with other benchmark models.
No takes yet. Share an insight, caveat, or question.
A 2023 study studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: