Scholarlink

<p>With growing mobile internet usage, powerful recommendation algorithms are emerging but struggle with real-world challenges such as sparsity, new users, and tiny datasets. Conventional algorithms require retraining for new users or contexts, leading to computational intensity. Therefore, we propose a novel deep learning-based model, task-adaptive intent attention graph neural networks(Cybertron), which is composed of three modules. Firstly, an intent cross attention module is designed to enable Cybertron to obtain inductive and transferable capabilities by a meta-learning training technique. Secondly, a subgraph generation module is designed to generate and divide subgraphs automatically. Thirdly, to better capture user-item interactions, a dynamic subgraph neural network module is proposed to learn user-item interactions at various scales. Experiments on the Steam, MovieLens, and Wanda datasets show that Cybertron performs effectively under inductive and transferable contexts across datasets of varying scales.</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

Cybertron: task-adaptive intent attention graph neural networks for few-shot recommendation

  • 期刊论文

  • 发表日期：

    2025年11月7日

查看全文
  • Kang Yang,
  • Ruiyun Yu

摘要

With growing mobile internet usage, powerful recommendation algorithms are emerging but struggle with real-world challenges such as sparsity, new users, and tiny datasets. Conventional algorithms require retraining for new users or contexts, leading to computational intensity. Therefore, we propose a novel deep learning-based model, task-adaptive intent attention graph neural networks(Cybertron), which is composed of three modules. Firstly, an intent cross attention module is designed to enable Cybertron to obtain inductive and transferable capabilities by a meta-learning training technique. Secondly, a subgraph generation module is designed to generate and divide subgraphs automatically. Thirdly, to better capture user-item interactions, a dynamic subgraph neural network module is proposed to learn user-item interactions at various scales. Experiments on the Steam, MovieLens, and Wanda datasets show that Cybertron performs effectively under inductive and transferable contexts across datasets of varying scales.