Analýza sentimentů v textech

Abstract

This thesis deals with the sentiment analysis in text documents. The aim of the thesis was to describe selected methods, to implement them and to test them on appropriately selected datasets. The theoretical part deals with the issue of natural language processing and the description of methods for sentiment recognition. Levels of analysis as well as text preprocessing, possible complications and division of methods for solving this task are presented in this part. There is a also brief description of the methods, that currently achieve the best results in the sentiment classification and detailed description of selected algorithms used in the practical part of the work. All selected algorithms are in the field of machine learning and were selected according to current trends. The practical part is focused on the details of the implementation of selected methods and neural network designs, followed by the verification and comparison of all methods using various experiments. These experiments test the effect of text preprocessing, hyperparameter optimization and the effect of changes in neural network structures. To perform said experiments, three datasets were selected to validate the methods both with insufficient and excessive amount of data given.