Study on Tibetan Word Vector based on Word2vec

Abstract

This paper uses Word2vec to study Tibetan word vector. Word2vec is optimized by two methods: Hierarchical Softmax and Negative Sampling in CBOW and Skip-gram models. Through the training of neural network, the words in Tibetan sentences are converted into vector form. Word2vec transforms the Tibetan text content processing into a simple vector space operation, calculates the similarity in the vector space, and then obtains the semantic similarity of the text, providing an accurate word vector for the training of the language model.