Sentiment Analysis Tool for Spanish Tweets in the Ecuadorian Context

Abstract

The huge amount of textual information that exists on social networks added by users through comments, has aroused a great interest in companies and research groups, which seek to use this information to identify trends and acceptance levels of brands, products, and services. A technique to know the level of acceptance or rejection of a particular topic, in an automated way, is the Sentiment Analysis. Some informatics tools incorporate this technique, however, there are few contributions to texts in Spanish. It is because of the difficulty of identifying different contexts, dialects, complex grammatical structures, and semantic language variances in each region. This article presents a web tool for the analysis of sentiments in texts written in Spanish that include Ecuadorian dialect or idioms. The tool was developed in R-Shiny with an approach lexicon-based. The tool allows the customization of the lexicons based on context and facilitates the automatic download of tweets according to search criteria such as the place, dates, and topic. To evaluate the effectiveness of the application, their result was compared with two commercial tools (Azure Text Analytics and IBM Watson NLU) and a manual score carried out by a group of people. The tests include the analysis of three corpora created from tweets. The results show the effectiveness of the tool to identify the sentiment polarity, especially in texts that include dialects, colloquial words, and negative expressions.