Abstract Aspect-based sentiment analysis (ABSA) aims to identify the sentiment associated with specific aspects or entities in a text. In order to facilitate the development and evaluation of ABSA systems, it is crucial to have annotated datasets that contain information about the aspects, entities, and the sentiments expressed towards them. However, the amount of information in existing datasets (for example those used in the SemEval shared tasks) is very limited. We innovate on existing corpora by introducing a multi-layered annotation schema that includes not only entities and aspects, but also lexical items and, crucially, functional discourse units (FDUs). These FDUs are text segments (typically sentences or clauses) that play a specific role or function within the overall text, such as “description”, “evaluation”, or “advice”, a type of information which we believe can be of great help in ABSA. Our corpus focuses on user reviews of tourist attractions (specifically monuments) in the region of Andalusia (Spain), but the same schema can be used to annotate reviews of other domains simply by adapting the aspects layer, which is domain-dependent. The annotation schema is described, and the validation process is carried out on a sample of 400 reviews from this domain. Results show a substantial level of agreement among the annotators, indicating that the schema is reliable and consistent. We go on to illustrate and discuss some difficult cases where annotation showed discrepancy among annotators. The annotation of FDUs in the corpus is a significant advancement for aspect-based sentiment analysis.