Analýza sentimentu pro komentáře na sociálních sítích pomocí řešení úlohy klasifikace

The content of this thesis is Twitter sentiment analysis. In descriptive part is explained natural language processing(NLP), application of NLP, the concept of sentiment and language methods for sentiment analysis, such as tokenization, POS tagging, stematization and lemmatization, TF-IDF, word2vec, machine learning methods for text analysis, such as k- nearest neighbors, multinomic naive Bayesian classifier , BERT model, neural network and SVM.The practical part is devoted to the implementation, experiments and comparison of the results of classifiers. Were implemented five different methods, which include a text preparation module, a feature extraction module and a classification module.These methods are SVM, k-nearest neighbors, multinomically Bayes classifier, BERT model, and word2vec in combination with SVM. The dataset for training and testing classifiers was downloaded from the Kaggle server and contains comments with tagged sentiment from Twitter social network.The experiments are mainly focused on the comparison of machine learning methods and the effectiveness of the application of different methods for preparation and feature extraction from the text. Metrics such as accuracy, precision, recall and F1 score were used to evaluate the classification.According to the experiments, the best algorithm of the ones described (studied) is the BERT model.