A text categorization system with soft real-time guarantee

Abstract

In order to provide predictable runtime performance for text categorization (TC) systems, an innovative system design method is proposed for soft real-time TC systems. An analyzable mathematical model is established to approximately describe the nonlinear and time-varying TC systems. According to this mathematical model, the feedback control theory is adopted to prove the system's stableness and zero steady state error. The experiments result shows that the error of deadline satisfied ratio in the system is kept within 4% of the desired value. And the number of classifiers can be dynamically adjusted by the system itself to save the computation resources. The proposed methodology enables the theoretical analysis and evaluation to the TC systems, leading to a high-quality and low-cost implementation approach.