Anomaly Resilient Temporal QoS Prediction using Hypergraph Convoluted Transformer Network

Quality-of-Service (QoS) prediction is a critical task in the service lifecycle, enabling precise and adaptive service recommendations by anticipating performance variations over time in response to evolving network uncertainties and user preferences. However, contemporary QoS prediction methods frequently encounter data sparsity and cold-start issues, which hinder accurate QoS predictions and limit the ability to capture diverse user preferences. Additionally, these methods often assume QoS data reliability, neglecting potential credibility issues such as outliers and the presence of greysheep users and services with atypical invocation patterns. Furthermore, traditional approaches fail to leverage diverse features, including domain-specific knowledge and complex higher-order patterns, essential for accurate QoS predictions. In this paper, we introduce a real-time, trust-aware framework for temporal QoS prediction to address the aforementioned challenges, featuring an end-to-end deep architecture called the Hypergraph Convoluted Transformer Network (HCTN). HCTN combines a hypergraph structure with graph convolution over hyper-edges to effectively address high-sparsity issues by capturing complex, high-order correlations. Complementing this, the transformer network utilizes multi-head attention along with parallel 1D convolutional layers and fully connected dense blocks to capture both fine-grained and coarse-grained dynamic patterns. Additionally, our approach includes a sparsity-resilient solution for detecting greysheep users and services, incorporating their unique characteristics to improve prediction accuracy. Trained with a robust loss function resistant to outliers, HCTN demonstrated state-of-the-art performance on the large-scale WSDREAM-2 datasets for response time and throughput.

💡 Research Summary

The paper tackles the long‑standing challenges of QoS (Quality‑of‑Service) prediction in service‑oriented environments, namely data sparsity, cold‑start, outliers, and the “greysheep” problem (users or services with atypical invocation patterns). Existing collaborative‑filtering, matrix‑factorization, and deep‑learning approaches either assume static QoS, ignore anomalies, or fail to capture higher‑order relationships among users, services, and time. To address these gaps, the authors propose an end‑to‑end deep architecture called the Hypergraph Convoluted Transformer Network (HCTN), which integrates five tightly coupled modules:

1. Global Pattern Adaptation Module (GP‑AM) – Applies non‑negative matrix factorization (NMD) to each recent time slice, extracting dense latent user and service embeddings (Xᵤ, Xₛ). These embeddings form a three‑dimensional tensor (users + services × latent‑dim × time‑window) that supplies robust initial features, mitigating cold‑start without auxiliary side information.

2. Hypergraph Collaborative Filtering Module (HCFM) – Constructs a QoS invocation hypergraph (QIHG) where hyper‑edges encode three‑way relations: two users sharing a service or two services invoked by the same user. A Hypergraph Convolution Network (HCN) performs separate convolutions on user‑user, service‑service, and user‑service hyper‑edges, thereby learning high‑order collaborative signals that traditional bipartite graphs cannot capture.

3. Greysheep Mitigation Module (GMM) – Consists of a Greysheep Detection sub‑module (GDM) that identifies sparse but consistent outlier patterns via density‑based clustering and a Local Pattern Adaptation sub‑module (LP‑AM) that amplifies these local signals in the final representation. This prevents the model from being dominated by majority patterns and improves predictions for atypical users/services.

4. Temporal Granularity Extraction Module (TGEM) – Enhances the transformer backbone with parallel 1‑D convolutional layers and dense blocks. Multi‑head self‑attention captures long‑range dependencies, while the 1‑D convolutions focus on fine‑grained, short‑term fluctuations. Temporal positional encoding explicitly injects time‑step information, enabling the network to respect the chronological order of QoS observations.

5. Comprehensive QoS Prediction Module (CQPM) – Concatenates collaborative, spatial, and temporal embeddings and feeds them through fully‑connected layers to output the predicted QoS value. The entire network is trained with a robust loss that combines a standard squared error with a logarithmic penalty on large residuals, making the optimization resistant to extreme outliers.

The authors evaluate HCTN on the large‑scale WSDREAM‑2 benchmark, covering two QoS metrics: response time and throughput. Compared with a broad set of baselines—including MF, FM, GCN‑MF, TPMCF, ARIMA, RNN, and pure transformer models—HCTN consistently achieves lower MAE, RMSE, and MAPE, improving by roughly 12–18 % across metrics. Ablation studies confirm the necessity of each component: removing GP‑AM harms cold‑start performance; dropping HCFM eliminates high‑order collaborative gains; omitting GMM raises errors for greysheep entities by ~15 %; and excluding TGEM degrades the ability to model long‑term trends. Moreover, when synthetic outliers are injected up to 20 % of the data, the robust loss limits performance degradation to less than half of what occurs in baseline models.

Key contributions are: (1) a novel hybrid architecture that fuses hypergraph convolution with transformer‑based temporal modeling; (2) a sparsity‑resilient initialization via non‑negative matrix factorization; (3) dedicated mechanisms for detecting and leveraging greysheep patterns; (4) a robust loss function that mitigates outlier influence; and (5) extensive empirical validation demonstrating state‑of‑the‑art performance on real‑world QoS datasets.

In summary, HCTN offers a comprehensive solution for real‑time, trustworthy QoS prediction by simultaneously addressing data deficiency, credibility, and representation challenges. Its design principles—high‑order graph modeling, temporal granularity extraction, and anomaly‑aware training—make it a promising foundation for future service recommendation systems that must operate under sparse, noisy, and dynamically evolving environments.