Research on energy-efficient text classification

People rely on data mining techniques like text categorization more and more to explore valuable information, due to the ever-increasing electronic documents produced. Although the energy consumed by text categorization increases with the data, people usually pay attention to its effectiveness and there is little research about its energy cost. In this paper, we evaluate the energy cost of different classifiers and reduce energy cost by parallelization, trying to find a classifier that performs best on both aspects - effectiveness and efficiency. Several classifiers are obtained by using existing libraries or implementing classification algorithms. Comprehensive experiments on three real datasets show that an improved version of Naive Bayes can have competitive precision compared to SVM while has low energy costs. Parallelization can further reduce its energy cost by a factor of 10 for RCV1 dataset.