In this Chapter, we explore unsupervised and supervised emotion classification methods for the detection of emotions. We start by examining how unsupervised learning identifies complex emotional patterns without needing predefined datasets and labelled data. This approach is contrasted with supervised learning methods, which rely on annotated datasets to train models, enabling emotion detection based on specific criteria. We also introduce word embeddings, a crucial development in NLP for context disambiguation, and discuss how they capture semantic meanings through word co-occurrences, allowing machines to grasp more subtle language nuances. We then discuss various techniques for detecting emotion categories and dimensions. In unsupervised learning, we examine methods such as the Vector Space Model and Latent Semantic Analysis, which identify patterns in text data independently of predefined categories. For supervised learning, we focus on training models on labelled datasets to accurately recognize specific emotions. Finally, we address the challenges and limitations of both unsupervised and supervised methods for emotion detection, considering the trade-offs between the flexibility of the unsupervised approaches and the precision of the supervised ones.