Comparison of Effectiveness of Word Representations Methods in Vector Space for the Text Sentiment Analysis
Natalia M. Lychenko, Anastasija V. Sorokovaja · Matematičeskie struktury i modelirovanie · 2019
The word representations in vector space is used for various tasks of automated processing of a natural language. There are many methods of vector representation of words, including neural network methods Word2Vec and GloVe, and the classical method of latent-sematic analysis LSA. This work is devoted to the study of the effectiveness of the application of vector representation of words in the neural network classifier for sentiment analysis of Russian and English texts based on the LSTM network. The features of word representations methods in vector space (LSA, Word2Vec, GloVe) are described, the architecture of a neural network classifier for sentiment analysis of text based on the LSTM network and the considered methods of vector representation of words are presented, the results of computational experiments and their discussion are presented. It is shown that the LSA model is the best model for the vector representation of words from the standpoint of learning speed, less corpus of words for learning, better accuracy and learning speed of the neural network classifier.