A Non-Negative Snowflake Factorization of Tensors Model for Dynamic QoS Prediction

Dynamic QoS measurements reveal intricate temporal trends in how users interact with services, offering vital insights into both user behavior and service status. However, as the scale of users and services grows, vast portions of QoS observations remain unseen, undermining service selection decisions. To fill in these gaps, we introduce a Nonnegative Snowflake Tensor Factorization framework. At its heart lies a specially structured "snowflake" core tensor that enriches the model’s representational power, while a Single Latent Factor–based Nonnegative Multiplicative Update on the Tensor (SLF-NMUT) scheme drives efficient parameter estimation. Through comprehensive experiments, we show that our approach more faithfully captures evolving user–service dynamics and delivers more accurate forecasts for missing QoS entries.