Unsupervised model for aspect-based sentiment analysis in spanish

Henríquez Miranda, Carlos | 2019-08-12

This paper presents an unsupervised model for Aspect-Based Sentiment Analysis in Spanish language, which automatically extracts the aspects of opinion and determines its associated polarity. The model uses ontologies for the detection of explicit and implicit aspects, and machine learning without supervision to determine the polarity of a grammatical structure in Spanish. The unsupervised approach used, allows implementing a system quickly scalable to any language or domain. The experimental work was carried out using the dataset used in Semeval 2016 for Task 5 corresponding to Sentence-level ABSA. The implemented system obtained a 73.07 F1 value in the extraction of aspects and 84.8% accuracy in the sentiment classification. The system obtained the best results of all systems participating in the competition in the three issues mentioned above.