The purpose of this thesis is to investigate several neural network approaches on the task of Aspect-Based Sentiment Analysis. The focus is customer reviews in the English language, concerning the domain of restaurants. We have implemented systems for the tasks of Aspect Category Detection and Sentiment Polarity, as defined in the competition of SemEval 2016. We provide a step by step description of the process followed in the creation of the models and results for the various augmentation techniques,that we incorporated. Our systems are evaluated against the baselines provided by the competition and the best performing systems from previous contestants. At the end we achieve notable results in both tasks, and we make the argument that neural networks can achieve good results even with a limited dataset, if they are combined with the appropriate augmentation techniques.
Admir Demirai is a Machine Learning enthusiast and recently he acquired his Master’s degree in Computer Science.Throughout his studies he has attended more than a few lectures oriented towards that direction and acquired a comprehensive knowledge of the field. The past 2 years he has taken on a few information retrieval tasks involving twitter and sentiment analysis tasks on restaurant reviews. He has been coding almost daily for 2 years and feels comfortable using both Python and Java. He is a good team player, highly motivated and always eagers to learn and improve.