基于Transformer的融合用户负反馈的重排序推荐方法  

Re-ranking recommendation method based on Transformer with user negative feedback

作　　者：胡德敏[1] 光萍 HU Demin;GUANG Ping(School of Optical-Electronical and Computer Engineering,University of Shanghai for Science and Technology,Shanghai 200093,China)

机构地区：[1]上海理工大学光电信息与计算机工程学院,上海200093

出　　处：《智能计算机与应用》2022年第12期196-201,共6页Intelligent Computer and Applications

基　　金：国家自然科学基金(61170277,61472256)。

摘　　要：针对现有的Top-n推荐模型只利用用户正反馈进行推荐的问题,本文提出一种基于Transformer的融合用户负反馈的重排序推荐方法。具体来说,本文使用GRU提取用户负反馈行为特征;基于门控单元思想构建一个融合单元,融合用户正负反馈行为特征,获得完整的用户行为特征;使用Transformer编码器整合用户特征和物品的位置信息,经过softmax计算物品得分获得重排序列表。实验结果表明,与几种先进的重排序推荐模型相比,本文的模型在AUC和Precision@k指标上有显著提升。Aiming at the problem that the existing Top-n recommendation model only applies users′ positive feedback for recommendation, this paper proposes a re-ranking recommendation method based on Transformer with users’ negative feedback. Specifically, this paper uses GRU to extract the characteristics of users′ negative feedback behavior;based on the idea of gating unit, a fusion unit is constructed to fuse the positive and negative feedback behavior characteristics of users to obtain complete user behavior characteristics. After that, Transformer encoder is used to integrate the user characteristics and the position information of items, and the reordering list is obtained by calculating the item score through softmax function. The experimental results show that compared with several advanced re-ranking recommendation models, AUC and Precision@k are significantly improved.

关 键 词：重排序 负反馈 TRANSFORMER 位置信息 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]