Dependable service-oriented computing relies on multiple Quality-of-Service (QoS) parameters to assess service performance. However, real-world QoS data are often sparse, noisy, and shaped by hierarchical dependencies arising from QoS interactions, geographical constraints, and network-level factors, making accurate prediction challenging. Existing deep learning-based methods often predict each QoS parameter separately, requiring multiple similar models, which increases computational overhead and limits their suitability for applications requiring joint decisions across multiple QoS parameters. Although recent joint QoS prediction studies employ shared architectures, differences in numerical ranges across QoS parameters can lead to task imbalance and negative transfer, while inadequate representation learning and suboptimal joint optimization can degrade prediction accuracy. We present SHARP-QoS, a joint QoS prediction framework with three components. First, hyperbolic graph and hypergraph convolutions in the Poincaré ball extract hierarchical features from QoS invocation and contextual graphs. Second, sparsely gated sub-network routing selectively shares contextual and cross-QoS features, and a gated fusion module combines them with the structural features. Third, an EMA-based loss-balancing strategy smooths short-term fluctuations in task losses to provide less fluctuating task weights. Experiments on three multi-QoS datasets covering two, three, and four QoS parameters show that SHARP-QoS achieves lower prediction errors than existing joint QoS prediction methods. Because its contextual branch relies only on public region and AS attributes, it supports prediction for new users and services. Ablation, hyperparameter, outlier, and scalability analyses further assess its performance, highlighting the trade-off between prediction accuracy and computational cost.
Figures & tables
Fig. 1 : SHARP-QoS: Overall framework.
Symbol
Description
U,S
Sets of users and services
n,m,N
Numbers of users, services, and nodes ( N=n+m )
P
Number of QoS parameters
Qp,Q^p
Observed and predicted QoS matrices for parameter p
Gp
QoS invocation graph for parameter p
Gup,Gsp
User and service hypergraphs for parameter p
TABLE I : Key notations.
Attribute
WSDREAM-2T [ 28 ]
Small-3T [ 24 ]
gRPC-4T [ 15 ]
Users / services
339 / 5,825
112 / 36
57 / 150
User regions / AS
31 / 137
24 / 80
13 / 16
Service regions / AS
74 / 2,699
16 / 28
1 / 1
QoS records (density)
1,831,253 (92.7%)
2,870 (71.2%)
8,550 (100%)
RT (s)
(min, max)
(0.001, 19.990)
(0.011, 25.806)
–
mean ± std
0.909 ± 1.973
1.267 ± 2.523
–
TABLE II : Dataset statistics.
QoS
Method
MAE (TD, % ) ↓
RMSE (TD, % ) ↓
5
10
15
20
5
10
15
20
WSDREAM-2T
RT
JQSP
0.5079
0.4406
0.4154
0.4072
1.9491
1.4445
1.3496
1.3181
WAMTL
0.4652
0.4097
0.3844
0.3663
1.3607
1.2800
1.2345
1.2017
PLHEN
0.4449
0.3863
0.3630
0.3449
1.3623
1.2691
1.2202
1.1878
DNM
0.4121
0.3621
0.3471
0.3214
1.3866
1.2670
1.2216
1.2070
TABLE III : Comparison with joint QoS prediction methods on WSDREAM-2T, Small-3T, and gRPC-4T under four training densities (TD) on MAE and RMSE error metrics.
Method
Params (M) ↓
FLOPs (G) ↓
Inference ( ×10−8 s) ↓
JQSP [ 21 ]
1.02
3207.91
79.40
WAMTL [ 25 ]
2.96
6863.40
50.20
PLHEN [ 26 ]
16.56
64076.28
508.00
DNM [ 20 ]
0.56
90.49
4.81
MGEN [ 22 ]
1.98
5872.59
66.30
SHARP-QoS
1.31
385.91
7.08
TABLE IV : Computational cost on WSDREAM-2T. Inference time is per prediction.
Backbone
TD = 10%
TD = 20%
RT
TP
RT
TP
MAE
RMSE
MAE
RMSE
MAE
RMSE
MAE
RMSE
GCN
0.3291
1.2628
11.2514
41.2364
0.2941
1.1807
10.0092
37.5613
HGCN
0.3278
1.2570
11.2415
41.1169
0.2936
1.1800
10.0202
37.2023
HyGCN
0.3254
1.2511
11.1710
40.9829
0.2909
1.1749
9.9972
36.9759
HHGCN (ours)
0.3234
1.2496
11.0828
40.7926
0.2894
1.1694
9.8934
36.5829
TABLE V : Graph convolution backbones for the QoS graphs on WSDREAM-2T.
Variant
TD = 10%
TD = 20%
RT
TP
RT
TP
MAE
RMSE
MAE
RMSE
MAE
RMSE
MAE
RMSE
Module removal
w/o C
0.3278
1.2585
11.1198
41.0943
0.2938
1.2040
10.0173
37.1704
w/o S
0.3241
1.2531
11.1678
41.2077
0.2924
1.1790
10.0208
37.7488
w/o H
0.3267
1.2570
11.4262
41.4448
0.2955
1.1796
10.0867
37.2375
TABLE VI : Module ablation and fusion variants on WSDREAM-2T. “w/o” removes HHGCN (H), contextual SNR (S), or Cross-SNR (C); fusion variants replace gated fusion.
Graphs
TD = 10%
TD = 20%
RT
TP
RT
TP
MAE
RMSE
MAE
RMSE
MAE
RMSE
MAE
RMSE
QoS only
0.3317
1.2734
11.1948
41.3736
0.2949
1.1871
10.1523
37.2919
QoS + region
0.3285
1.2596
11.1704
41.3532
0.2915
1.1752
9.9939
37.2060
QoS + AS
0.3298
1.2676
11.1856
41.1025
0.2907
1.1721
9.9293
37.2759
Merged
0.3299
1.2732
11.5039
42.0504
0.2961
1.1994
10.0540
37.4746
TABLE VII : Graph configurations on WSDREAM-2T. “Merged” combines QoS, region, and AS edges in one graph; SHARP-QoS (Ours) learns from each graph separately.
Fig. 2 : Single-QoS methods versus SHARP-QoS on WSDREAM-2T at 10% and 20% training densities. Shading highlights SHARP-QoS; lower MAE is better.
Method
Density 10%
Density 20%
RT
TP
RT
TP
Classical MTL (NMF features)
MMoE [ 18 ]
0.3701
13.1410
0.3395
12.7715
Cross-Stitch [ 17 ]
0.3634
12.4842
0.3233
11.2576
PLE [ 19 ]
0.3554
13.7637
0.3234
12.7865
SHARP-QoS variants
TABLE VIII : Classical MTL architectures and SHARP-QoS variants on WSDREAM-2T (MAE).
Method
Density 10%
Density 20%
RT
TP
RT
TP
Gradient-based
NashMTL [ 40 ]
0.3935
36.8322
0.3637
36.5003
CAGrad [ 41 ]
0.3666
23.0353
0.3172
13.2636
PCGrad [ 42 ]
0.3461
11.6027
0.3334
10.7505
MGDA [ 43 ]
0.3441
24.0792
0.3060
21.2061
TABLE IX : Multi-task optimization strategies within SHARP-QoS on WSDREAM-2T (MAE).
Fig. 3 : Test error of SHARP-QoS on WSDREAM-2T after removing the 2% – 10% most anomalous test records identified by Isolation Forest [ 48 ] , for each training density (TD).
Method
CSP = 10%
CSP = 20%
RT
TP
RT
TP
MAE
RMSE
MAE
RMSE
MAE
RMSE
MAE
RMSE
Cold records
WAMTL [ 25 ]
0.8828
2.1813
37.7567
98.5129
0.8709
2.0572
39.0988
101.1953
JQSP [ 21 ]
0.8619
2.0832
35.7774
94.1139
0.8901
1.9557
36.1012
99.2150
PLHEN [ 26 ]
0.8352
2.1440
29.4874
85.6311
0.9549
2.2848
30.9929
89.0028
TABLE X : Cold-start results on WSDREAM-2T (TD = 10%), averaged over three runs. CSP is the percentage of users and services withheld; cold records involve a withheld entity.
Setting
MAE
STD
90% CI
95% CI
99% CI
RT, TD = 10%
0.3234
0.141
[0.2907, 0.3561]
[0.2844, 0.3624]
[0.2721, 0.3747]
RT, TD = 20%
0.2894
0.122
[0.2611, 0.3177]
[0.2557, 0.3231]
[0.2451, 0.3337]
TP, TD = 10%
11.0828
1.976
[10.6231, 11.5425]
[10.5351, 11.6305]
[10.3630, 11.8026]
TP, TD = 20%
9.8934
1.722
[9.4930, 10.2938]
[9.4163, 10.3705]
[9.2664, 10.5204]
TABLE XI : Mean test MAE, standard deviation (STD) of group MAEs, and confidence intervals for SHARP-QoS on WSDREAM-2T ( G=50 groups).
With the rapid development of cloud computing and Web services, Quality of Service (QoS) has become a key criterion for service selection and recommendation. Tensor latent feature analysis provides an effective way to model multidimensional QoS data, and most existing QoS prediction methods are mainly based on Canonical Polyadic (CP) decomposition or Tucker decomposition. However, constrained by their inherent structural properties, these methods cannot accurately capture the complex and dynamic dependencies in user-service interactions, which limits their prediction performance. To address this issue, this paper proposes a dynamic QoS prediction framework based on the Biased Nonnegative Block Term Tensor Decomposition Model, termed BNBT. Specifically, the proposed framework is developed from three aspects: (1) block term tensor decomposition is employed to enhance the representation capability of latent feature learning; (2) linear bias terms are incorporated to further improve prediction accuracy; and (3) a tensor-oriented single-element-dependent nonnegative multiplicative update algorithm, called SLF-NMUT, is designed for efficient parameter estimation. Extensive experiments on real-world QoS datasets demonstrate that the proposed BNBT framework consistently outperforms several state-of-the-art QoS prediction methods in terms of prediction accuracy.
Wenjing Liu, Yujia Lei, Qu Wang
College of Computer and Information Science Southwest University Chongqing, China
Predicting temporal Quality of Service (QoS) data is critical for optimizing network services and rationalizing resource allocation in cloud computing and service-oriented systems. Existing mainstream methods have achieved promising predictive performance. However, their purely data-driven manner limits their ability to capture non-stationary temporal patterns, thereby leading to accuracy degradation when temporal QoS data exhibits fluctuations. To tackle this limitation, we propose a novel Extended Kalman Filter-Enhanced Latent Feature Analysis (EKL) model to perform efficient and accurate temporal QoS prediction from the perspective of bidirectional model-data-driven learning. Its main idea is three-fold: a) designing a model-driven feature producer to obtain the temporal latent features to capture the intricate temporal pattern following the principle of an Extended Kalman Filter; b) building a data-driven feature producer based on the alternating least squares algorithm to identify time-invariant latent features describing intrinsic user-service characteristics; c) exploiting a density-oriented parallel strategy that achieves workload balancing by sorting users in accordance with their service invocation density, which effectively elevates computational efficiency. In addition, we provide a rigorous theoretical analysis to formally prove the convergence of the proposed EKL. Experimental evaluations conducted on real-world temporal QoS datasets reveal that our proposed EKL surpasses existing state-of-the-art models with respect to both computational efficiency and prediction accuracy for missing temporal QoS data.
Ye Yuan, Song Wang, Hongxun Zhou +2
College of Computer and Information Science, Southwest University, Chongqing 400715, China
Teleoperation serves as the fallback solution to autonomous driving but reliable functions of the teleoperation require a certain amount of mobile network resources, which cannot be guaranteed at all times. Therefore, predictive quality of service (pQoS) is introduced as a concept to increase the resilience of the teleoperation. In this paper, based on a data measurement campaign, we propose a prediction framework to prediction two important network KPIs of teleoperation: uplink data-rate and round-trip latency. Furthermore, we introduce a method to alleviate the performance degradation of machine-learning-based prediction models on previously unseen data due to concept drift by incorporating historic data into the prediction pipeline. Additionally, we introduce the metric of critical scenario detection to evaluate the prediction performance specifically for teleoperation.
Xiyan Su, Jianning Gao, Mahmoud Ashri +1
Institute of Automotive Technology, Technical University of Munich (TUM), Boltzmannstr.15, DE-85748, Garching bei München, Germany