Quantificando a importância de emojis e emoticons para a identificação de polaridade

Abstract

Virtual environments, such as online stores (e.g. Amazon, Google Play, Booking), pro- mote a collaborative strategy for reviewing products and services. The users’ opinions represent their degree of satisfaction regarding the reviewed item. The set of reviews of an item serves as a reputation index. Hence, the automatic identification of the user satisfaction, regarding an item, based on his/her textual review, is a tool of great economic and strategic potential for enterprises. In this context, the growing adoption of emojis and emoticons, boosted by the mobile devices and their Apps, the users incre- asingly adopt such a vocabulary to express their opinion and sentiments. In this work, we present a quantitative assessment of the richness of emojis/emoticons to predict the users’ opinion in product reviews in collaborative systems. Our proposal uses the Bag of Words with Support Vector Machine to predict the users’ opinion in a online review, taking into account the use of: (1) only words; (2) words and emojis/emoticons; and (3) only emojis/emoticons. For certain scenarios, considering the F1 metric, the use of words and emojis results in an efficacy of 0.92 using words combined with emojis, compared to 0.81 when only words are used (traditional approach).