Tibetan text classification using distributed representations of words

Tibetan text classification is one of the most important research topics in Tibetan information processing. In the existing Tibetan text classification method, the representation of documents is based on traditional vector space model which has the high dimension data and lack semantic information. In this paper, a Tibetan text classification based on distributed representations of words method is proposed. With this method one can first tags the POS of the document by using maximum entropy model, and then selects only nouns and verbs as key features. At last document are represented by the weight of the word classes, which are trained by word2vec tool. The experimental results show that our model outperforms competitive traditional Tibetan text classification method, and the F-measure has improved by 9%.