基于傅里叶变换与近端采样的序列推荐算法
Sequence Recommendation Algorithm Based on Fourier Transform and Recency-Based Sampling
杨兴耀 1李晨瑜 1于炯 1李梓杨1
作者信息
- 1. 新疆大学软件学院,新疆 乌鲁木齐 830008
- 折叠
摘要
传统推荐算法比较注重于模型本身对于推荐效果的提升,实际上数据质量对于算法的影响更为重要,但目前在推荐算法领域对于数据的科学处理方法比较零散,没有形成一个系统的框架.针对以上问题,基于傅里叶变换与近端序列采样的数据预处理,结合SASRec提出可以普遍应用的序列推荐框架FTRRec.首先通过傅里叶变换将序列数据在时域和频域中相互转换,并根据序列数据的特点,过滤序列中的无用信息,其次使用近端序列采样替换传统的滑动窗口采样法,加速样本采样的同时,提升模型对于序列的特征捕获能力.通过在 5 个公开数据集上的实验,将框架应用于三个不同的主流推荐算法时,每种模型均有 3％-5％的提升.
Abstract
Traditional recommendation algorithms pay more attention to the improvement of the recommendation effect of the model itself.In fact,data quality is more important to the algorithm.But at present,scientific methods for data processing in the field of recommendation algorithms are scattered,and there is no systematic framework.To solve the above problems,this paper proposes a sequence processing framework that can be widely used based on Fourier transform and near end sequence sampling.First,the sequence data is converted into each other in the time domain and frequency domain through Fourier transform,and the useless information in the sequence is filtered according to the characteristics of the sequence data.Secondly,the traditional sliding window sampling method is replaced by the near end sequence sampling method,which accelerates the sample sampling and improves the model's ability to cap-ture the characteristics of the sequence.Through the experiments on five public datasets,when the framework is ap-plied to three different mainstream recommendation algorithms,each model has a 3％-5％ improvement.
关键词
序列化推荐/数据处理/傅里叶变换/序列采样Key words
Sequence recommendation/Data processing/Fourier transform/Sequential sampling引用本文复制引用
基金项目
国家自然科学基金项目(61862060)
国家自然科学基金项目(61966035)
国家自然科学基金项目(61562086)
新疆维吾尔自治区教育厅项目(XJEDU2016S035)
新疆大学博士科研启动基金项目(BS150257)
新疆维吾尔自治区自然科学基金面上项目(2022D01C56)
出版年
2024
NETL
NSTL
万方数据