Eureka! Institutional Repository

Data Mining and Sentiment Analysis in texts are two important fields in Computer Science and Artificial Intelligence. They are a valuable tool for understanding attitudes and opinions expressed on social networks, such as Twitter. The use of Machine Learning methods and modern Natural Language Processing models allows for the automatic analysis of text content and the extraction of important information, while also offering, accuracy and convenience in drawing conclusions. In this paper, we utilize the Twitter API for data collection from Twitter, in combination with Natural Language Processing (NLP) methods. Specifically, we use Machine Learning models from the Scikit-learn library, as well as more modern models, such as BERT, RoBERTa, DistilBERT, and GPT-2, with the aim of identifying sentiment in text from the Twitter social network, as well as in reviews of stores contained in a specific dataset from the Skroutz.gr online service. According to our experiments, the models that show the best performance in terms of accuracy for predicting on new data, are BERT and SVM combined with the TF-IDF encoding.