| Περίληψη: | The field of Textual Sentiment Analysis has been met with increased interest by the research community in recent years due to the rise of social media and the Internet. The vast amount of opinion-heavy user-generated content that is available to us, whether that is a product/service review or an opinion on an event, shows that effective Sentiment Analysis is needed. However, automatic knowledge extraction about the opinion and emotional state of people can be a very challenging task. This thesis studies the fields of Machine Learning and Deep Learning in-depth, in order to perform Sentiment Analysis and by extension Emotion Recognition classification tasks. A novel Hidden Markov Model-based approach is proposed where a single model is trained for each class label with the help of clustering and a lexicon. Overall, the main goal is to present a variety of Machine Learning models, ranging from basic all the way to state-of-the-art approaches, and implement them in real-world datasets. Initially, the theory behind the aforementioned fields and the related literature is introduced. Then, we present the mathematical background of the proposed approaches as well as expand on their usage for text classification and its challenges. The task at hand is supervised text classification. Additionally, a survey of the available datasets in the Sentiment Analysis domain is performed and the field of Ensemble Learning is explored. Finally, we implement and evaluate the proposed models on benchmark datasets using k-fold cross-validation and come to conclusions regarding each algorithm’s ability to recognize peoples' opinions and emotions. From the experimental results it is observed that the proposed Hidden Markov Models and Deep Neural Networks with word embeddings achieve very high performance, proving that they are potent and suitable tools for Sentiment Analysis and classification tasks in general.
|
|---|
