SENTIMENT ANALYSIS OF TEXT IN SERBIAN LANGUAGE USING DEEP LEARNING

Abstract

Sentiment analysis is a scientific field that deals with the analysis of opinions, attitudes and emotions of people who have written a certain text. Such analysis can be used for purposes like brand monitoring, customer service improvement, product analysis, market research, creating recommendation systems etc. In this paper, several machine learning models are trained and compared for the task of sentiment analysis. Reviews used as a dataset were collected from one of the food delivery websites in Serbia. Reviews contain a set of ratings that will be used as an indicator of emotional polarity. Three models of machine learning with different types of vectorization were trained and compared with results of transfer learning approach. Transfer learning is a method of deep learning where a model trained to solve one problem is used as a starting point in solving another problem. Deep learning model based on transformers is used for transfer learning.