Scholarlink

<p>Information propagation prediction is a critical task for comprehending the propagation of information among users. Previous researches have faced significant challenges in the task of information propagation prediction, including the inadequate capture of higher-order interactions among users, an imbalance between broad contextual and fine-grained local relational information, and the insufficient integration of learned representations. To address these issues, we propose a Representation Fusion Model of Multi-Category Hypergraphs (RFMH) for predicting information propagation. Specifically, dynamic hypergraph, cascade hypergraph, and dependency hypergraph are constructed by using cascade sequences. Then a hypergraph attention network and a hypergraph convolution are used to learn the short-term influence representations of dynamic hypergraph and the long-term influence representations of cascade and dependency hypergraph, respectively. Additionally, two adjacent layers of graph convolutional networks are employed to learn the broad contextual information representation of the social network graph while preserving fine-grained local relational information. Furthermore, we implement a fusion method of multi-category representations based on a multi-head attention mechanism to nonlinearly fuse the learned representations. Experimental results on four real-world datasets demonstrate that the RFMH model improves Hits@k metrics by 0.48% to 7.87% and MAP@k metrics by 1.52% to 3.64%. Paired t-tests show statistically significant results of <InlineEquation ID="IEq1"> <EquationSource Format="TEX">\({{\textbf {P}}_{Hits@k}}\)</EquationSource> </InlineEquation>=0.001&lt;0.05 and <InlineEquation ID="IEq2"> <EquationSource Format="TEX">\({{\textbf {P}}_{MAP@k}}\)</EquationSource> </InlineEquation>=0.036&lt;0.05, respectively. The results of paired t-tests confirm that our model significantly outperforms the optimal baseline, as well as verify its validity and reasonableness. The following is the location of the code for this manuscript: <a href="https://github.com/yangyui/RFMH.git">https://github.com/yangyui/RFMH.git</a></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

A representation fusion model of multi-category hypergraphs for information propagation prediction

  • 期刊论文

  • 发表日期：

    2026年5月16日

查看全文
  • Chao Liu,
  • Jiayu Yang,
  • Haohan Ma

摘要

Information propagation prediction is a critical task for comprehending the propagation of information among users. Previous researches have faced significant challenges in the task of information propagation prediction, including the inadequate capture of higher-order interactions among users, an imbalance between broad contextual and fine-grained local relational information, and the insufficient integration of learned representations. To address these issues, we propose a Representation Fusion Model of Multi-Category Hypergraphs (RFMH) for predicting information propagation. Specifically, dynamic hypergraph, cascade hypergraph, and dependency hypergraph are constructed by using cascade sequences. Then a hypergraph attention network and a hypergraph convolution are used to learn the short-term influence representations of dynamic hypergraph and the long-term influence representations of cascade and dependency hypergraph, respectively. Additionally, two adjacent layers of graph convolutional networks are employed to learn the broad contextual information representation of the social network graph while preserving fine-grained local relational information. Furthermore, we implement a fusion method of multi-category representations based on a multi-head attention mechanism to nonlinearly fuse the learned representations. Experimental results on four real-world datasets demonstrate that the RFMH model improves Hits@k metrics by 0.48% to 7.87% and MAP@k metrics by 1.52% to 3.64%. Paired t-tests show statistically significant results of \({{\textbf {P}}_{Hits@k}}\) =0.001<0.05 and \({{\textbf {P}}_{MAP@k}}\) =0.036<0.05, respectively. The results of paired t-tests confirm that our model significantly outperforms the optimal baseline, as well as verify its validity and reasonableness. The following is the location of the code for this manuscript: https://github.com/yangyui/RFMH.git