重庆理工大学学报2025，Vol.39Issue(7)：182-189,8.DOI:10.3969/j.issn.1674-8425(z).2025.04.023
融合注意力机制和会话推荐的点击率预测模型
A click-through rate prediction model combining attention mechanisms and session recommendations
李唯唯 1孙永冠 1周正楠 1夏萱1
作者信息
- 1. 重庆理工大学计算机科学与工程学院,重庆 400054
- 折叠
摘要
Abstract
How to effectively mine potential user interest from their behavioral sequences is a key issue in click-through rate prediction research.Current research directly models interests with historical behavioral features while ignoring the intrinsic structure of sequence information.To address the problem,this paper proposes an interest extraction session recommendation incorporating the attention(ISRA)for the click-through prediction task.First,the method subdivides the user's click sequence into several sessions according to the time interval and performs interest modeling in terms of sessions to enhance the model's ability to capture changes in the user's short-term interest.Meanwhile,it employs the attention mechanism with multiple positional encoding to weigh the session information for differentiation and feature interaction to achieve a more significant expression of the long-term and short-term interest evolution process.Finally,the effectiveness of the method is verified by extensive experimental analysis on the datasets Criteo and MovieLens-1M.关键词
点击率预测/会话推荐/注意力机制/兴趣提取/行为序列Key words
click-through prediction/session recommendation/attention mechanism/interest extraction/behavioral sequences分类
信息技术与安全科学引用本文复制引用
李唯唯,孙永冠,周正楠,夏萱..融合注意力机制和会话推荐的点击率预测模型[J].重庆理工大学学报,2025,39(7):182-189,8.
