WORD2VEC AND BERT LANGUAGE MODELS USED FOR A SENTIMENT ANALYSIS OF TEXT POSTS IN SOCIAL NETWORKS
Nadezhda Glebovna Yarushkina, Vadim Moshkin, Andrei A. Konstantinov · Автоматизация процессов управления · 2020
The paper proposes an original algorithm for the formation of a training sample for a neural network that provides a sentiment analysis of text posts in social networks. A feature of the algorithm is the use of the extended Russian-language semantic thesaurus WordNetAffect and the expert dictionary of author’s symbols for expressing emotions. In addition, the paper describes the application of a neural network based on the LSTM architecture to determine the emotional coloring of text messages on a social network using two text vectorization algorithms “word2vec” and “BERT”. As a result of the experiments, an indicator of the accuracy of determining the emotional coloring of messages of 87% was achieved using lemmatization as a text preprocessing algorithm and the BERT algorithm when converting it into a vector.