Abstract: A classification method based on word contribution was proposed to classify spam short messages. The concept of word contribution was introduced for representing weight difference of a word in different categories, the word contributionclassification matrix was constructed, then the mean square deviation of each row in the matrix was computed to reduce dimensionalities. To determine the classification a short message belongs to, short messagecategory membership degrees were calculated based on word contribution. Furthermore if category candidates were more than one, the classification conflict problem could be resolved by comparing the densities of short messagecategory membership degree. The experimental results showed that the proposed method was superior to other classification methods in the classification result and realtime.

Key words: spam short message, text classification, word contribution, variance, dimensionality reduction