基于时长感知的短视频序列推荐  

Duration-Aware for Short Video Sequential Recommendation

作　　者：王航 尹玲 史志才[1] 黄勃[1] 高志荣[2] WANG Hang;YIN Ling;SHI Zhicai;HUANG Bo;GAO Zhirong(School of Electronic and Electrical Engineering,Shanghai University of Engineering Science,Shanghai 201620,China;School of Computer Science,South-Central Minzu University,Wuhan 430074,China)

机构地区：[1]上海工程技术大学电子电气工程学院,上海201620 [2]中南民族大学计算机科学学院,武汉430074

出　　处：《计算机工程与应用》2024年第22期304-313,共10页Computer Engineering and Applications

基　　金：国家自然科学基金(61802251);中南民族大学2020年校级重点教研项目(JYZD20020);中南民族大学2022年校级研究生课程思政示范课程项目(YJS22039)。

摘　　要：针对短视频序列推荐中存在的点击数据稀疏性、观看时长反馈中的噪声以及偏差问题,提出了一种基于时长感知的短视频序列推荐模型(duration-aware for short video sequential recommendation,DASR)。该模型通过对用户观看时长反馈的深入建模,有效地缓解了数据稀疏性问题。提出了一种无偏差的多语义观看时长反馈标签生成方法。该方法结合了K近邻算法和训练数据的百分位数分析,动态生成适应不同视频时长的标签阈值,有效地消除了视频时长偏差的影响。提出了一种基于强弱注意力网络的噪声提取方法,从观看时长中准确地提取正向和负向兴趣信号,从而解决了观看时长反馈中存在的噪声。在开源的短视频数据集上进行了广泛实验,结果表明该模型在多个评价指标上优于其他主流方法。Addressing the issues of data sparsity in click data,noise in watch duration feedback,and bias in short video sequential recommendation,a duration-aware for short video sequential recommendation model(DASR)is proposed.This model effectively alleviates the data sparsity issue by deeply modeling user watch duration feedback.Additionally,an unbiased multi-semantic watch duration feedback label generation method is proposed.This method combines the K-nearest neighbors algorithm and percentile analysis of training data to dynamically generate label thresholds adapted to different video durations,effectively eliminating the impact of video duration bias.Furthermore,a noise extraction method based on a strong-weak attention network is introduced,accurately extracting positive and negative interest signals from the watch duration,thus addressing the noise issue in watch duration feedback.Extensive experiments on open-source datasets demonstrate that this model outperforms other mainstream methods on multiple evaluation metrics.

关 键 词：数据稀疏性 序列推荐 注意力网络 时长偏差 动态阈值 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]