Evaluating Impact of Emoticons and Pre-processing on Sentiment Classification of Translated African Tweets

Saurav Keshari Aryal, Gaurav Adhikari

Keywords: Preprocessing, Emoticons, English Translation, Supervised Modeling, Sentiment Classification, AfriSenti-SemEval

TL;DR: Perform translation on African Language Tweeets use translation and remove/retain emoticons with different Roberta models for sentiment classification.

Abstract: This paper examines the impact of emoticons and pre-processing on sentiment classification for English translations of 11 African languages. Using AfriSenti-SemEval datasets, Roberta and Twitter-Roberta models are fine-tuned, and standard classification metrics are used to assess performance. The study concludes no significant performance differences with emoticons and pre-processing and no distinction between standard Roberta and domain-specific Twitter-Roberta.