Comparison of Deep Learning and Rule-based Method for the Sentiment Analysis Task

Sentiment analysis is a field in the Natural Language Processing. The main objective of this field is the determination of opinions expressed in the text in relation to given objects or aspects of these objects. In this paper we explored two approaches in the sentiment analysis: lexicon-based approach and machine learning. We took four Russian sentiment corpora and tested lexicon-based method and neural language model BERT on them. For lexicon-based method the intersections of ten Russian sentiment lexicons were used. Neural language model showed better performance results for all the corpora.