UBC-DLNLP at SemEval-2023 Task 12: Impact of Transfer Learning on African Sentiment Analysis

Proceedings of the The 17th International Workshop on Semantic Evaluation (SemEval-2023), pages 246–255 July 13-14, 2023 ©2023 Association for Computational Linguistics UBC-DLNLP at SemEval-2023 Task 12: Impact of Transfer Learning on African Sentiment Analysis Gagan Bhatia1,⋆ Ife Adebara1,⋆ AbdelRahim Elmadany1 Muhammad Abdul-Mageed1,2 1Deep Learning & Natural Language Processing Group, The University of British Columbia 2Department of Natural Language Processing & Department of Machine Learning, MBZUAI {gagan30@student.,ife.adebara@,a.elmadany@,muhammad.mageed@}ubc.ca Abstract We describe our contribution to the SemEVAl 2023 AfriSenti-SemEval shared task, where we tackle the task of sentiment analysis in 14 different African languages. We develop both monolingual and multilingual models under a full supervised setting (subtasks A and B). We also develop models for the zero-shot setting (subtask C). Our approach involves experiment- ing with transfer learning using six language models, including further pretraining of some of these models as well as a final finetuning stage. Our best performing models achieve an F1-score of 70.36 on development data and an F1-score of 66.13 on test data. Unsurprisingly, our results demonstrate the effectiveness of transfer learning and finetuning techniques for sentiment analysis across multiple languages. Our approach can be applied to other sentiment analysis tasks in different languages and do- mains. 1 Introduction Sentiment Analysis, also referred to as opinion mining, is a Natural Language Processing (NLP) technique that aims to identify, extract, and evalu- ate opinions, attitudes, perceptions, and sentiments towards topics, products, services, and individu- als from textual data (Birjali et al., 2021). With the increased accessibility of the internet, people are increasingly sharing their opinions on various platforms, such as forums, blogs, wikis, websites, and social media pages. Consequently, there is a need for the automatic extraction of sentiments to gain valuable insights into user perception, popu- lar opinion, and trends (Georgiadou et al., 2020; Ramírez-Tinoco et al., 2018). Despite the increasing popularity of sentiment analysis, its application in low-resource African languages is still under-explored (Shode et al., 2022; Diallo et al., 2021). This is because many African languages have limited digital resources, such as annotated data and lexical resources, which Figure 1: A map showing the countries where each language in the shared task is spoken in Africa. can hinder the development and evaluation of sen- timent analysis models. So far, only a handful of African languages have few datasets for senti- ment analysis (Imam Abubakar et al., 2021; Ogbuju and Onyesolu, 2019; Muhammad et al., 2023c,a; Oyewusi et al., 2020; Muhammad et al., 2022). Fur- thermore, African languages often exhibit complex morphology, syntax, semantics, stylistics, prag- matic and orthographic conventions including the use of diacritics, and code-mixing that can make it difficult to accurately identify and extract sentiment from text data (Muhammad et al., 2023c,a; Ori- maye et al., 2012). For instance, for some African languages, a single change in tone assignment can change the sentiment of a text (Adebara and Abdul- Mageed, 2022). In this task, we conduct sentiment analysis on 14 African languages including Algerian Arabic, Amharic, Hausa, Igbo, Kinyarwanda, Moroccan Arabic, Mozambican Portuguese, Nigerian Pidgin, …

Content truncated to 500 words