Ensemble Language Models for Multilingual Sentiment Analysis

TL;DR

This paper explores multilingual sentiment analysis on social media, demonstrating that ensemble models, especially majority voting, improve performance, with monolingual models excelling and addressing low-resource language challenges.

Contribution

It introduces two ensemble language models for multilingual sentiment analysis and compares their effectiveness across different languages and datasets.

Findings

Monolingual models outperform multilingual ones.

Ensemble models outperform baseline models.

Majority voting ensemble performs best among tested methods.

Abstract

The rapid advancement of social media enables us to analyze user opinions. In recent times, sentiment analysis has shown a prominent research gap in understanding human sentiment based on the content shared on social media. Although sentiment analysis for commonly spoken languages has advanced significantly, low-resource languages like Arabic continue to get little research due to resource limitations. In this study, we explore sentiment analysis on tweet texts from SemEval-17 and the Arabic Sentiment Tweet dataset. Moreover, We investigated four pretrained language models and proposed two ensemble language models. Our findings include monolingual models exhibiting superior performance and ensemble models outperforming the baseline while the majority voting ensemble outperforms the English language.