Scholarlink

<p>Recommendation systems are a branch of machine learning with diverse applications across e-commerce, entertainment, and education. In recent years, sequential/session-based recommendation systems have emerged as a powerful method for capturing user preferences, whether through behavioural history over time or interactions within a single session. This paper explores sequential recommender systems that utilize user behavior sequences within a session. Our research emphasizes session-based product recommendations using Graph Neural Networks (GNNs). Additionally, we propose an approach that integrates a transformer with a target-attentive GNN, incorporates dropout within the GNN, and employs a session graph with Lazy Linear to enhance the performance of the model. We evaluate the proposed method using Yoochoose and Diginetica datasets, measuring performance through two key metrics: precision at 20 (P@20) and mean reciprocal rank at 20 (MRR@20). On the Yoochoose dataset, our approach achieves a P@20 of 72.16 and an MRR@20 of 31.89. On the Diginetica dataset, the results show a P@20 of 53.49 and an MRR@20 of 18.75. Our experimental results demonstrate that the proposed method is competitive with existing approaches on real-world benchmark datasets, showing that incorporating a Transformer and Dropout into GNN for sequential recommendation can improve recommendation results compared to the original GNN.</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

Graph Neural Networks with Transformer for Sequential Recommendation

  • 期刊论文

  • 发表日期：

    2025年9月1日

查看全文
  • Mai Thi Cam-Nhung,
  • Nguyen Thai-Nghe

摘要

Recommendation systems are a branch of machine learning with diverse applications across e-commerce, entertainment, and education. In recent years, sequential/session-based recommendation systems have emerged as a powerful method for capturing user preferences, whether through behavioural history over time or interactions within a single session. This paper explores sequential recommender systems that utilize user behavior sequences within a session. Our research emphasizes session-based product recommendations using Graph Neural Networks (GNNs). Additionally, we propose an approach that integrates a transformer with a target-attentive GNN, incorporates dropout within the GNN, and employs a session graph with Lazy Linear to enhance the performance of the model. We evaluate the proposed method using Yoochoose and Diginetica datasets, measuring performance through two key metrics: precision at 20 (P@20) and mean reciprocal rank at 20 (MRR@20). On the Yoochoose dataset, our approach achieves a P@20 of 72.16 and an MRR@20 of 31.89. On the Diginetica dataset, the results show a P@20 of 53.49 and an MRR@20 of 18.75. Our experimental results demonstrate that the proposed method is competitive with existing approaches on real-world benchmark datasets, showing that incorporating a Transformer and Dropout into GNN for sequential recommendation can improve recommendation results compared to the original GNN.