Twitter sentiment analysis using deep learning

Current innovations are rapidly changing the world, and internet usage has become a basic requirement in people's lives. Nowadays, people can check what other customers think, about a product, also they can exchange views and experience about any product before they buy on the Internet. Collecting and analyzing people's opinions about a product is essential for business applications when it is properly extracted and analyzed. But, it is impossible to analyze opinions manually because the content is unstructured. For this reason, we applied sentiment analysis that extracts and analyzes the unstructured data automatically. With this thesis, people's opinions and sentiments can be analyzed that help corporation to enhance the quality of the products or services. This study focused on the classification of sentiments for three product reviews of fast-food restaurants, which are McDonald's, Kentucky fried chicken, and Burger king. Twitter was chosen as the data source for analysis. Tweets were collected automatically by using Tweepy, and three different data sets consisting of 50k, 100k, and 200k experimented. First, the raw data was pre-processed. Each tweet was processed with a pre-trained Word2Vec model, and the words were converted into vectors. During the experimental study, these data were trained by giving them to the classifiers. Then, the test data were given to the model and classified as positive or negative. In this study, three deep learning techniques implemented, which are CNN, CNN-Bi-LSTM, and Bi-LSTM. All datasets are split into training, validation, and testing. The performance of each technique was measured and compared in terms of accuracy, precision, recall, and F1-score. Also, the confusion matrix calculated for the best model is given. Finally, Bi-LSTM achieved the highest performance in 200K Twitter dataset in all metrics compared to the other two models and achieved the highest accuracy of 95.35%.