Scholarlink

<p>Graph Neural Network (GNN)-based recommendation methods exhibit significant advantages in modeling the complex relationships among users and items, users and users, as well as items and items. However, these approaches heavily rely on historical interaction data and can only recommend known items, making it difficult to effectively recommend unknown items that users have not previously interacted with. This limitation results in the filter bubble effect. Therefore, a key challenge is how to recommend unknown items to users, given the limited historical interaction data. To address this issue, this paper proposes a method that models the user decision-making process for unknown-item recommendation in recommendation systems. To simulate real-time user decision-making, we design a dual-dimensional connection network that integrates a cross-attention mechanism and Student’s t-distribution to capture users’ interest intentions and decision-making behaviors. Furthermore, we introduce a few-shot learning strategy to infer the latent structural representations of unknown items within the GNN framework. Based on the predicted comprehensive scores of unknown items, those with the highest scores are selected and recommended. Experimental results on two public datasets, Amazon G&amp;GF and Yelpsmall, show that LUMR outperforms state-of-the-art GNN-based models, achieving improvements of 62.5% in P@10 and 35.9% in MRR@10.</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

LUMR: leveraging user decision processes to model unknowns in recommendation system

  • 期刊论文

  • 发表日期：

    2026年5月18日

查看全文
  • Wenbo Yang,
  • Shiwei Gao,
  • Mengyi Chen,
  • Pengxue Yun,
  • Jingjing Xie,
  • Tianqi Chen

摘要

Graph Neural Network (GNN)-based recommendation methods exhibit significant advantages in modeling the complex relationships among users and items, users and users, as well as items and items. However, these approaches heavily rely on historical interaction data and can only recommend known items, making it difficult to effectively recommend unknown items that users have not previously interacted with. This limitation results in the filter bubble effect. Therefore, a key challenge is how to recommend unknown items to users, given the limited historical interaction data. To address this issue, this paper proposes a method that models the user decision-making process for unknown-item recommendation in recommendation systems. To simulate real-time user decision-making, we design a dual-dimensional connection network that integrates a cross-attention mechanism and Student’s t-distribution to capture users’ interest intentions and decision-making behaviors. Furthermore, we introduce a few-shot learning strategy to infer the latent structural representations of unknown items within the GNN framework. Based on the predicted comprehensive scores of unknown items, those with the highest scores are selected and recommended. Experimental results on two public datasets, Amazon G&GF and Yelpsmall, show that LUMR outperforms state-of-the-art GNN-based models, achieving improvements of 62.5% in P@10 and 35.9% in MRR@10.