The growth of online content generated by users on the Web and social media has turned Sentiment Analysis into one of the most active research areas of Natural Language Processing aimed to computationally identify the underlying opinions, attitudes or feelings of a given text. In particular, a variety of methods have emerged that focus on Twitter due to its challenging traits, being a central issue how to provide a proper feature-based representation for its short and incomplete messages, i.e. tweets. The contribution of this paper is twofold: on the one hand, we present an approach to perform sentiment tagging of tweets based on text and emoji polarity; on the other hand, a set of manifold dimensionality reductions are carried out that allow a convenient 3D visualization and a rapid prototyping of sentiment patterns. We then compare the performance obtained with such reduced feature spaces when applied to the classification of sentiments of a collection of tweets from a real-life case study of people experiencing the celebration of a massive festivity.