基于加速无约束张量隐因子分解模型的Web服务QoS估计
Accelerated unconstrained latent factorization of tensor model for Web service QoS estimation
林铭炜 1李文强 1许秀琴 2刘健1
作者信息
- 1. 福建师范大学福建省公共服务大数据挖掘与应用工程技术研究中心,福建 福州 350117;福建师范大学计算机与网络空间安全学院,福建 福州 350117
- 2. 福建师范大学数学与统计学院,福建 福州 350117
- 折叠
摘要
针对基于张量非负隐因子分解模型的Web服务QoS估计方法过于依赖非负初始随机数据以及特意设计的非负训练方法,导致模型的兼容性和扩展性不高的问题,提出了加速无约束张量隐因子分解模型.其主要思想包括三部分:将非负性约束从决策参数转移到输出的隐因子,并通过单元素映射函数连接它们;运用结合动量方法的随机梯度下降算法,有效提高模型的收敛速度与估计精度;给出加速无约束张量隐因子分解模型的详细算法和结果分析.在实际工业应用中的 2 个动态QoS数据集上的实证研究表明,与最先进的QoS估计模型相比,所提模型具有较高的计算效率和估计精度.
Abstract
Aiming at the problem that the Web service quality of service(QoS)estimation methods based on the non-negative latent factorization of tensor model(NLFT)depend heavily on non-negative initial random data and spe-cially designed non-negative training schemes,which lead to low compatibility and scalability,an accelerated uncon-strained latent factorization of tensor(AULFT)model was proposed.The proposed model consisted of three main parts.The non-negative constraints from decision parameters were transferred to output latent factors and they were connected through the single-element-dependent mapping function.A momentum-incorporated stochastic gradient descent(MSGD)algorithm was used to effectively improve the convergence rate and estimation accuracy of the proposed AULFT model.The detailed algorithm and result analysis of the proposed AULFT model were presented.The empirical study on two dynamic QoS datasets in real industrial applications demonstrates that the proposed AULFT model has higher computa-tional efficiency and estimation accuracy than the state-of-the-art QoS estimation models.
关键词
服务质量/隐因子分解分析/张量非负隐因子分解模型/无约束非负/动量方法Key words
quality of service/latent factorization analysis/non-negative latent factorization of tensor model/uncon-strained non-negative/momentum method引用本文复制引用
基金项目
国家自然科学基金(62272103)
福建省自然科学基金杰出青年基金(2022J06020)
福建省"雏鹰计划"青年拔尖人才计划(F21E0011202B01)
出版年
2024
国家科技期刊平台
NSTL
万方数据