Integrating hashtags in sentiment analysis on Twitter posts in Spanish language using BETO

Abstract

Amongst the latest developments in Natural language processing are pretrained transformers. One field in which they are used is sentiment analysis, which seeks to identify the intent or emotion expressed in a text. Hashtags in social media texts, such as tweets, are often removed as part of the pre-processing. As hashtags can contain important information or necessary context to understand the sentiment of a tweet, removing them can reduce the classification performance. In this study we performed sentiment analysis with a pre-trained BETO model on Spanish tweets that have the Chilean dictator Pinochet as topic. We show that including hashtags into sentiment analysis can increase f1-score. This indicates that changes in pre-processing procedures can increase efficiency of Natural Language Processing approaches.