<p>Keyphrase extraction is a significant challenge in natural language processing with diverse applications. However, existing state-of-the-art methods model documents using convolutional neural networks, focusing solely on local feature subsequences to capture local relationships between adjacent elements. This disregards the importance of the overall structure of the document, directly impacting the quality of generated keyphrases. To address this challenge, we propose an end-to-end framework for keyphrase extraction that explores the integration of dependency-awareness, self-attention, and graph convolutional networks for latent representation learning. Additionally, we construct a label selection filter to optimize keyphrase prediction results based on global representations by integrating selection vectors, thereby enhancing the accuracy of keyphrases. Competitive experimental results on public datasets validate the effectiveness of the proposed method over state-of-the-art methods. The experimental results show an average improvement of about <b>1</b>.<b>9</b>% and <b>7</b>.<b>8</b>% in <InlineEquation ID="IEq1"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="11227_2025_7303_Article_IEq1.gif" Format="GIF" Height="16" Rendition="HTML" Resolution="72" Type="Linedraw" Width="18" /> </InlineMediaObject> <EquationSource Format="TEX">\(F_1\)</EquationSource> <EquationSource Format="MATHML"><math> <msub> <mi>F</mi> <mn>1</mn> </msub> </math></EquationSource> </InlineEquation>@5 and <InlineEquation ID="IEq2"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="11227_2025_7303_Article_IEq2.gif" Format="GIF" Height="16" Rendition="HTML" Resolution="72" Type="Linedraw" Width="18" /> </InlineMediaObject> <EquationSource Format="TEX">\({F_1}\)</EquationSource> <EquationSource Format="MATHML"><math> <msub> <mi>F</mi> <mn>1</mn> </msub> </math></EquationSource> </InlineEquation>@10 scores, respectively, on the four major public datasets. </p>

错误:搜索内容不能为空,请输入英文关键词
错误:关键词超出字数限制,请精简
高级检索

Optimizing keyphrase extraction with dependency relation-aware attention graph convolutional networks

  • Yuyu Dong,
  • Fang’ai Liu,
  • Xuqiang Zhuang,
  • Ran Bai,
  • Xuejian Gao

摘要

Keyphrase extraction is a significant challenge in natural language processing with diverse applications. However, existing state-of-the-art methods model documents using convolutional neural networks, focusing solely on local feature subsequences to capture local relationships between adjacent elements. This disregards the importance of the overall structure of the document, directly impacting the quality of generated keyphrases. To address this challenge, we propose an end-to-end framework for keyphrase extraction that explores the integration of dependency-awareness, self-attention, and graph convolutional networks for latent representation learning. Additionally, we construct a label selection filter to optimize keyphrase prediction results based on global representations by integrating selection vectors, thereby enhancing the accuracy of keyphrases. Competitive experimental results on public datasets validate the effectiveness of the proposed method over state-of-the-art methods. The experimental results show an average improvement of about 1.9% and 7.8% in \(F_1\) F 1 @5 and \({F_1}\) F 1 @10 scores, respectively, on the four major public datasets.