Adaptive core-enhanced latent factor model for highly accurate QoS prediction

Area of Science:

• Distributed Systems and Cloud Computing
• Machine Learning and Artificial Intelligence
• Data Mining and Service Computing

Background:

• Accurate Quality of Service (QoS) prediction is vital for service recommendation and selection in distributed systems.
• Traditional Latent Factor (LF) models, while scalable, often fail to capture complex interactions and rely on manual regularization, limiting prediction accuracy.

Purpose of the Study:

• To propose a novel Adaptive Core-Enhanced Latent Factor (ACELF) model for superior QoS prediction.
• To enhance the expressiveness and robustness of LF models in capturing intricate user-service interactions.

Main Methods:

• Developed a learnable core interaction matrix to model latent user and service factor interactions beyond standard bilinear assumptions.
• Integrated an incremental Proportional-Integral-Derivative (PID)-driven adaptive regularization strategy to dynamically adjust coefficients during training.
• Implemented a dynamic optimization process to balance model expressiveness and prevent overfitting.

Main Results:

• The ACELF model demonstrated consistent performance improvements over state-of-the-art methods on real-world QoS datasets.
• The adaptive regularization strategy effectively managed the trade-off between model complexity and generalization.
• The learnable core interaction matrix captured richer latent representations, enhancing prediction accuracy.

Conclusions:

• The proposed ACELF model offers a significant advancement in QoS prediction accuracy for service recommendation.
• Adaptive regularization and learnable interaction matrices are effective strategies for overcoming limitations of traditional LF models.
• ACELF provides a more robust and accurate solution for large-scale distributed environments.