Analysis on Content Multi-Classification of Chinese Tweets

This paper focuses on messages classification in Chinese tweets without additional information. It classifies tweets into five valid categories-A, B, C, D, E and one invalid category-Other. The data only in five valid categories are trained, and the other category applies rejection method if the five valid categories deny it. To improve the efficiency of classification, we pre-process the data on three proposed rules. The precision of the improved-SVM (ISVM) algorithm we proposed is approximately 10% higher than Naive Bayesian classification and 1.3% higher than classical SVM.