Team ISCL_WINTER at SemEval-2023 Task 12:AfriSenti-SemEval: Sentiment Analysis for Low-resource African Languages using Twitter Dataset

Proceedings of the The 17th International Workshop on Semantic Evaluation (SemEval-2023), pages 1085–1089 July 13-14, 2023 ©2023 Association for Computational Linguistics Team ISCL_WINTER at SemEval-2023 Task 12:AfriSenti-SemEval: Sentiment Analysis for Low-resource African Languages using Twitter Dataset Alina Hancharova and John Wang and Mayank Kumar Eberhard Karls University of Tübingen Abstract This paper presents a study on the effectiveness of various approaches for addressing the chal- lenge of multilingual sentiment analysis in low- resource African languages. The study focuses on Task 12 of the SemEval-2023 Competition, which aims to promote interest in these lan- guages and develop efficient models for their analysis. The approaches evaluated in the study include Support Vector Machines (SVM), trans- lation, and an ensemble of pre-trained multilin- gual sentimental model methods. The paper provides a detailed analysis of the performance of each approach based on experimental results. In our findings, we suggest that the ensemble method is the most effective with an F1 Score of 0.68 on the final testing. This system ranked 19 out of 33 participants in the competition. 1 Introduction There is a growing interest in using AI for various natural language processing (NLP) tasks such as sentiment analysis, machine translation, and hate- ful content detection in African languages. How- ever, most of these languages do not have curated datasets available for developing such AI applica- tions. Recently, there have been individual and funded initiatives to create datasets for African lan- guages, but research is needed to determine the suitability of current NLP techniques and the devel- opment of techniques to maximize the applications of such datasets. Subtask B, Task 12 of shared task the SemEval-2023 Competition Muhammad et al. (2023b) focus on 12 African languages and is designed to strengthen their further development by including languages such as Hausa, Yoruba, Igbo, Nigerian Pidgin, Amharic, Tigrinya, Oromo, Swahili, Algerian Arabic dialect, Kinyarwanda, Twi, Mozambique Portuguese, and Moroccan Ara- bic/Darija. In this study, the goal is to tackle the problem of sentiment analysis in African languages, which are under-represented in natural language processing (NLP) research. To address this chal- lenge, we employed an ensemble strategy using deep language pre-trained models. These mod- els were individually able to handle the languages present in our dataset, which included 12 African languages. 1.1 Competition Details The competition is divided into 3 sub-tasks. In the first task, a language is chosen and sentiment analysis is performed. These languages are Hausa, Yoruba, Igbo, Nigerian Pidgin, Amharic, Alge- rian Arabic, Moroccan Arabic/Darija, Swahili, Kin- yarwanda, Twi, Mozambican Portuguese, and Xit- songa (Mozambique Dialect), Setswana, TisiZulu, Xitsonga (South African Dialect). Languages are related to Afro-Asiatic, Niger-Congo, and Semitic families. These languages are low resource and diverse. In the second task, the participant works on the combination of the 12 languages database. The third subtask, unlabelled tweets in two African languages (Tigrinya and Oromo), and zero-shot sentiment analysis are performed. For the com- petition participation, our team chose the second task. 1.2 Dataset Description The AfriSenti dataset is a set of tweets, written …

Content truncated to 500 words