Temporal Pattern-Aware QoS Prediction via Biased Non-Negative Latent Factorization of Tensors

Area of Science:

• Computer Science
• Data Science
• Machine Learning

Background:

• Quality-of-Service (QoS) data exhibit temporal dynamics crucial for accurate prediction.
• Existing latent factor (LF) analysis-based QoS predictors often overlook these temporal patterns, treating data as static.
• This limitation hinders the precise prediction of missing QoS values in dynamic environments.

Purpose of the Study:

• To develop a novel model for temporal pattern-aware QoS prediction that addresses the limitations of static analysis.
• To enhance the accuracy of predicting missing QoS data by effectively capturing time-varying patterns.
• To improve computational efficiency in QoS prediction for dynamic datasets.

Main Methods:

• A biased non-negative latent factorization of tensors (BNLFTs) model is proposed.
• The model incorporates linear biases to account for QoS fluctuations and non-negativity constraints.
• It utilizes a multiplicative update scheme and an alternating direction method for efficient training and faster convergence.

Main Results:

• The BNLFT model demonstrated superior performance in representing temporal patterns compared to state-of-the-art QoS predictors.
• Empirical studies on two real-world dynamic QoS datasets confirmed the model's effectiveness.
• The BNLFT model achieved highly accurate predictions for missing QoS data with significant computational efficiency.

Conclusions:

• The BNLFT model offers a robust solution for temporal pattern-aware QoS prediction.
• It outperforms existing methods in accuracy and efficiency for dynamic QoS datasets.
• This approach is vital for applications requiring precise prediction of time-varying service quality.