Scholarlink

<p>Recommendation systems play a critical role in our daily lives. Despite great progress, existing graph-based recommendation methods still suffer from challenges including skewed data distribution, vulnerability to noises, and sparse supervision signal. We attribute the inferior performance to the limited discriminative ability of the learned representations. To remedy this, in this paper, we develop a framework termed Multi-ACG by introducing self-supervised learning, adversarial learning, and multitask learning to learn representations with higher discrimination. Specifically, self-supervised learning and adversarial learning are first employed to synthesize hard samples for training. Meanwhile, multi-task learning is adopted to balance different loss terms for optimization. Experiments are conducted on benchmark datasets and the results have demonstrated the state-of-the-art performance of the proposed method against previous ones. The code is at <a href="https://github.com/xiaoma666123/Multi-ACG.">https://github.com/xiaoma666123/Multi-ACG.</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

Multitask learning of adversarial-contrastive graph for recommendation

  • 期刊论文

  • 发表日期：

    2025年2月13日

查看全文
  • Xingyu Ma,
  • Chuanxu Wang

摘要

Recommendation systems play a critical role in our daily lives. Despite great progress, existing graph-based recommendation methods still suffer from challenges including skewed data distribution, vulnerability to noises, and sparse supervision signal. We attribute the inferior performance to the limited discriminative ability of the learned representations. To remedy this, in this paper, we develop a framework termed Multi-ACG by introducing self-supervised learning, adversarial learning, and multitask learning to learn representations with higher discrimination. Specifically, self-supervised learning and adversarial learning are first employed to synthesize hard samples for training. Meanwhile, multi-task learning is adopted to balance different loss terms for optimization. Experiments are conducted on benchmark datasets and the results have demonstrated the state-of-the-art performance of the proposed method against previous ones. The code is at https://github.com/xiaoma666123/Multi-ACG.