价格引导双流自注意力序列推荐模型  

Price-guided Dual Self-attention Sequential Recommendation Model

作　　者：孙克雷[1] 吕自强 Sun Kelei;Lyu Ziqiang(Anhui University of Science and Technology,Huainan 232001,China)

机构地区：[1]安徽理工大学,安徽淮南232001

出　　处：《廊坊师范学院学报（自然科学版）》2024年第2期29-35,共7页Journal of Langfang Normal University(Natural Science Edition)

基　　金：国家自然科学基金项目(61703005);安徽省高校科研重点项目(2022AH050821)。

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%。Aiming at the problem that the traditional sequential recommendation algorithm has a weak ability to capture the long-termdependence 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 itemprice information to analyze user price preference and assist in calculating itemsimilarity,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-termdependence 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 minimumand 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 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]