基于偏差的邻域模型：云与IoT服务上下文感知QoS预测
49 浏览量
更新于2024-08-31
收藏 866KB PDF 举报
"Deviation-based neighborhood model for context-aware QoS prediction of cloud and IoT services"
随着互联网上云服务（Cloud services）和物联网服务（IoT services）的爆发式增长，如何为用户提供个性化、高质量的服务选择成为一个日益重要的问题。服务质量（Quality of Service, QoS）预测是解决这一问题的关键，它借鉴了推荐系统的思想，即协同QoS预测。在这个框架下，本文提出了一种新颖的基于偏差的邻域模型，旨在利用大众智慧进行QoS预测。
现有的工作主要关注于传统的邻域模型，而我们的方法引入了一个双层结构，这使得模型能够更有效地适应上下文感知（Context-aware）的QoS预测。上下文感知是指根据用户的环境、时间、位置等多维度信息来提供定制化的服务质量预测。在我们的方法中，"偏差"是关键概念，用于度量用户的历史行为与服务实际表现之间的差异。通过分析这些偏差，模型可以更好地理解用户对服务质量的需求和预期，从而提供更为精确的预测。
我们的模型首先通过收集并分析大量用户和服务的交互数据，构建一个全局的服务质量数据库。然后，利用用户的历史行为数据，找到与当前用户具有相似QoS偏好和上下文条件的邻居用户。接下来，通过对这些邻居用户的服务评价进行加权平均，我们得到一个基础预测值。然而，由于每个用户的偏好可能有微小但重要的差异，我们引入偏差调整机制，对基础预测值进行精细化调整，从而提升预测的准确性。
实验结果表明，提出的偏差基邻域模型在预测精度和全局优化效率方面优于传统方法。尤其在处理具有细粒度调整需求的场景下，我们的模型表现出更优的性能。此外，由于模型考虑了上下文因素，它能更好地满足不同环境和条件下用户对服务质量的个性化需求。
这篇研究论文为云服务和物联网服务的QoS预测提供了新的视角和解决方案，通过结合用户的行为偏差和邻域学习，实现了更准确、更智能的预测，有望为服务提供商和用户在服务选择过程中提供有力的支持。未来的研究可以进一步探索如何动态更新邻域模型，以适应快速变化的用户需求和技术环境。
552 H. Wu et al. / Future Generation Computer Systems 76 (2017) 550–560
2.2. QoS prediction based on matrix factorization
Different from neighborhood-based approaches, matrix factor-
ization approaches have attractive accuracy and scalability thus
recently become popular in recommender systems [16]. A typical
model associates each user u with an user-factors vector A
u
∈ R
f
,
and each item i with an item-factors vector B
i
∈ R
f
. The prediction
is done by taking an inner product, i.e.,
ˆ
q
ui
= A
T
u
B
i
. To exploit matrix
factorization for QoS prediction, Zheng et al. [8,12] using proba-
bilistic matrix factorization (PMF [33]) approach to decompose the
QoS matrix. For identifying latent factors, A and B, a least-square’s
problem like (2) are built and solved using gradient descent [16].
min
A,B
(u,i)∈E
(q
ui
− A
T
u
B
i
)
2
+ λ
u
∥A
u
∥
2
+ λ
i
∥B
i
∥
2
. (2)
Matrix factorization can partially alleviate sparsity-sensitive
problem of collaborative filtering thus increase the accuracy of
QoS prediction. For the last two years, numerous efforts have been
made on improving MF-based models. These works concentrate
on utilizing additional information, such as spatial and temporal
information associated with users or services. Zhang et al. [22]
use collective matrix factorization that simultaneously factor
the user–service quality matrix, service category and location
context matrices. Zhang et al. [23] factorize user–service–time
matrix of QoS using non-negative tensor factorization with time
information. Yin et al. [25] develop a location-based regularization
framework for PMF prediction model. Lo et al. [24] exploit PMF
prediction model with a location-based pre-filtering stage on QoS
matrix. He et al. [34] develop location-based hierarchical matrix
factorization. Yu et al. [26] experience trace-norm regularized
matrix factorization. Following neighborhood-integrated matrix
factorization(NIMF) [8], Qi et al. [30] propose a MF-based method,
integrating both user network neighborhood information and
service neighborhood information, to predict personalized QoS
values.
Matrix factorization can provide accurate predictions while sac-
rificing explainability, as the learned latent factors are unexplain-
able. Lack of explainability weakens the ability to persuade users
and help users make better decisions in practical systems. Also,
data sparsity has a negative impact on these methods, as data be-
comes extremely sparse, the prediction performance will be not
optimistic.
In addition, neural network based deep learning technology
has been proposed for collaborative filtering, e.g., Restricted
Boltzmann Machines [35]. Similar to matrix factorization, deep
learning methods allow us to discover the latent features
underlying data. It may indicate a new trend in the QoS prediction,
however, deep learning methods always incur high computational
overheads, and suffer from the similar problems in matrix
factorization techniques.
3. Deviation-based neighborhood model for QoS prediction
To cope with existing drawbacks of CF-based prediction
methods, we suggest using machine learning techniques to build
neighborhood model for QoS prediction. New models allow
an efficient global optimization scheme and exploit different
baseline estimate components to improve prediction accuracy. To
distinguish users from services, we take different indexing letters:
for users u, v, and for services i, j. The notation q
ui
indicates
a known quality-score observed by user u on service i and the
notation
ˆ
q
ui
represents the predicted value of q
ui
. The (u, i) pairs
for which q
ui
is observed are stored in the set E = {(u, i)|q
ui
=
null}. Usually the vast majority of QoS scores are missing. To
combat overfitting in learning prediction model on the sparse data,
models are regularized so estimates are shrunk towards baseline
estimates [16].
3.1. The framework
User-oriented methods estimate unknown quality based on
recorded QoS of like minded users. Analogously, in service-
oriented methods, a QoS value is estimated using known QoS made
by the same user on comparable services. In cloud/IoT computing
environment, the context with users is more complex and dynamic
than that of services. Prediction leveraged by similar users other
than services is more practical. Therefore, our focus is on user-
oriented approaches, but parallel techniques can be developed in
a service-oriented fashion, by switching the roles of users and
services.
We develop the QoS prediction model on the basis of
neighborhood-based collaborative filtering [16], which allows an
efficient global optimization scheme and offers improved accuracy.
To facilitate global optimization, we would want to abandon
user-specific similarity, s(u, v) in (1), in favor of global weights
independent of a specific user. The weight from user v to user u
is denoted by x
uv
are able to be learned from the data through
optimization. By this, we can overcome the weaknesses with
existing neighborhood-based models. An initial sketch of the
model describes each quality score q
ui
by:
ˆ
q
ui
= b
ui
+
v∈N
u
x
uv
(q
vi
− b
vi
), (3)
where N
u
is the neighbor set of u, b
ui
is the baseline estimate that
we will gradually construct considering different factors.
As for the interpretation of weights, usually they represent
interpolation coefficients relating unknown quality score to the
existing ones in a traditional neighborhood model [16] (recall
s(u, v) in (1)). Here, we adopt them in a different viewpoint that
weights represent offsets to baseline estimates and residual, q
vi
−
b
vi
, is viewed as the coefficients multiplying those offsets. For two
similar users u and v, x
uv
is always expected to get high, and vice
versa. So, our estimate will not deviate much from the baseline
estimate by a user v that accessed i just as expected (q
vi
− b
vi
is
around zero), or by a user v that is not known to be predictive on
u (x
uv
is close to zero).
Generally, we can take all users in N
u
other than u, however, this
would increase the number of weights to be estimated. In order to
reduce complexity of the model, we suggest pruning parameters
corresponding to unlikely user–user relations. Let N
k
u
be the set of
k users who are most similar to u, as determined by the similarity
measure s(u, v). Further, we let N
k
(i;u)
, N
i
∩N
k
u
, where N
i
is the set
of users have used the service i. Now, when predicting q
ui
according
to formula (3), it is expected that the most influential weights will
be associated with users similar to u. Hence, we replace (3) with:
ˆ
q
ui
= b
ui
+ |N
k
(i;u)
|
−
1
2
v∈N
k
(i;u)
x
uv
(q
vi
− b
vi
). (4)
When k = ∞, rule (4) coincides with (3). When k = 0,
ˆ
q
ui
= b
ui
.
However, for other values of k, it offers the potential to significantly
reduce the number of variables involved. This final prediction rule
permits fast online prediction, since more computational works,
such as similarity calculation and parameter estimation, have
been made in the pre-processing stage. Recall that unlike matrix-
factorization, the neighborhood models allow a direct explanation
of their predictions, and do not require re-training the model for
handling new services.
3.2. Components for baseline estimation
Typical QoS data exhibit large user and service effects-
i.e., systematic tendencies for some users to achieve better QoS
than others, and for some services to receive better QoS than
剩余10页未读，继续阅读
2013-09-01 上传
2022-01-10 上传
2021-02-25 上传
2021-02-13 上传
2021-03-12 上传
2016-11-09 上传
2015-03-29 上传
2021-03-08 上传
2022-03-02 上传
weixin_38536841
- 粉丝: 3
- 资源: 946
我的内容管理
展开
最新资源
- StarModAPI: StarMade 模组开发的Java API工具包
- PHP疫情上报管理系统开发与数据库实现详解
- 中秋节特献：明月祝福Flash动画素材
- Java GUI界面RPi-kee_Pilot：RPi-kee专用控制工具
- 电脑端APK信息提取工具APK Messenger功能介绍
- 探索矩阵连乘算法在C++中的应用
- Airflow教程：入门到工作流程创建
- MIP在Matlab中实现黑白图像处理的开源解决方案
- 图像切割感知分组框架：Matlab中的PG-framework实现
- 计算机科学中的经典算法与应用场景解析
- MiniZinc 编译器：高效解决离散优化问题
- MATLAB工具用于测量静态接触角的开源代码解析
- Python网络服务器项目合作指南
- 使用Matlab实现基础水族馆鱼类跟踪的代码解析
- vagga：基于Rust的用户空间容器化开发工具
- PPAP: 多语言支持的PHP邮政地址解析器项目
