价格引导双流自注意力序列推荐模型
Price-guided Dual Self-attention Sequential Recommendation Model
孙克雷 1吕自强1
作者信息
摘要
针对传统序列推荐算法捕获交互序列中的长期依赖性能力较弱,以及由于数据稀疏性导致推荐结果缺乏个性化的问题,提出了一种价格引导双流自注意力序列推荐模型.通过融合项目价格信息分析用户价格偏好并辅助计算项目相似度,提高推荐结果的个性化程度;将两种信息输入到两个独立的自注意力机制,学习不同位置的重要性、提取其特征,并将输出进行拼接后输入到门控单元学习时间依赖性,提高模型的长期依赖性建模能力.在三个公开数据集上验证了模型的有效性,命中率和归一化折损累积增益最低提升1.11％,最高提升5.34％.
Abstract
Aiming at the problem that the traditional sequential recommendation algorithm has a weak ability to capture the long-term dependence in the interaction sequence,and the lack of personalization of the recommendation results due to data sparsity,this paper proposes a Price-guided Dual Self-attention Sequential Recommendation(PG-DSASR).By inte-grating item price information to analyze user price preference and assist in calculating item similarity,the personalized de-gree of recommendation results is improved.The two kinds of information are input into two independent self-attention mechanisms to learn the importance of different positions and extract their features,and the output is spliced and input into the gated unit to learn the time dependence,so as to improve the long-term dependence modeling ability of the model.The effectiveness of the model is verified on three public datasets.The hit rate and the cumulative gain of normalized loss are in-creased by 1.11％at the minimum and 5.34％at the maximum.
关键词
推荐算法/序列推荐/项目价格/自注意力机制/长期依赖性Key words
recommendation algorithm/sequential recommendation/item price/self-attention mechanism/long-term de-pendence引用本文复制引用
出版年
2024
NETL
NSTL
万方数据