Enhancing Spanish Aspect-Based Sentiment Analysis Through Deep Learning Approach

Aspect-based sentiment analysis is the task of monitoring user sentiment on textual opinions about the characteristics of a given entity. Recognizing the aspects present in the opinion and determining its sentimental orientation (positive or negative) in a similar way as if it were done by a human being continues to be a challenging task, but at the same time necessary. Achieving quality improvement of existing aspect-based sentiment analysis solutions remains a challenge and the vast majority of solutions reported in this task are focused on the English language, so further progress is needed in languages like Spanish. This paper presents an aspect-based sentiment analysis method which uses language models based on Transformers (BERT models) with a linear layer to extract aspects from opinions in Spanish and a similar model to perform Aspect Sentiment Classification, demonstrating that its use allows us to surpass the state of the art for said language. The proposed solution was evaluated using the Semeval2016 Task 5 dataset achieving promising results, respect those reported by other solutions.