Aspect Extraction for Opinion Mining with a Semantic Model

In this article, we present a semantic model for aspect extraction from Spanish text as part of a complete aspect-based sentiment analysis system. The model uses ontology, semantic similarity, and double propagation techniques to detect explicit and implicit aspects. The proposed approach allows the implementation of a scalable system for any language or domain. The experimental tests were carried out using the SemEval-2016 dataset for task 5, corresponding to the aspect-based sentiment analysis sentence level. The implemented system obtained an F1 score of 73.07 for the aspect extraction, achieving the best results among the systems participating in the comparison, and an F1 score of 89.18 for the hotel domain using a ten-iteration cross-validation.