TechSSN at SemEval-2023 Task 12: Monolingual Sentiment Classification in Hausa Tweets

Proceedings of the The 17th International Workshop on Semantic Evaluation (SemEval-2023), pages 1190–1194 July 13-14, 2023 ©2023 Association for Computational Linguistics TechSSN at SemEval-2023 Task 12: Monolingual Sentiment Classification in Hausa Tweets Nishaanth Ramanathan, Rajalakshmi Sivanaiah, Angel Deborah Suseelan and Mirnalinee Thanka Nadar Thanagathai Department of Computer Science and Engineering, Sri Sivasubramaniya Nadar College of Engineering, Chennai, Tamil Nadu, India nishaanth2010302@ssn.edu.in, rajalakshmis@ssn.edu.in, angeldeborahs@ssn.edu.in, mirnalineett@ssn.edu.in Abstract This paper elaborates on our work in designing a system for SemEval 2023 Task 12: AfriSenti- SemEval, which involves sentiment analysis for low-resource African languages using the Twitter dataset. We utilised a pre-trained model to perform sentiment classification in Hausa- language tweets. We used a multilingual ver- sion of the roBERTa model, which is pre- trained on 100 languages, to classify sentiments in Hausa. To tokenize the text, we used the AfriBERTa model, which is specifically pre- trained on African languages. 1 Introduction Sentiment analysis is a subset of Natural Language Processing (NLP), in which the intent is to classify the sentiment present in textual data (Kincl et al., 2016). It is mostly done in languages that have high resources but not in low-resource languages like Hausa, Yoruba, Igbo, and other African languages, despite the fact that they are being used by nearly one-third of human beings. It is essential for us to respect every language and perform NLP research in all languages. Performing NLP research in all languages is crucial for both political and business reasons (Polpinij, 2014). The rapid growth of the Internet and usage of social platforms like Facebook, Twitter, etc. have expanded the area of NLP research (Martin et al., 2021). The task we participated in involved performing sentiment classification in tweets of an African language, and we chose to focus on the Hausa language (Muhammad et al., 2023b). In this task, tweets must be classified as positive, negative, or neutral. Performing sentiment analysis in low-resource languages like Hausa is crucial not only because it is used by many people but also because of the need to interpret the sentiment of every single person. Our strategy involved tokenizing textual data, fine-tuning a pre-trained model with the tokenized data, and measuring the model’s performance on the development dataset. We fine-tuned a model (Muhammad et al., 2022) that was pre-trained on the NaijaSenti corpus (Muhammad et al., 2022), which was provided in the task overview. Our strategy involved tokenizing textual data, fine-tuning a pre-trained model with the tokenized data, and measuring the model’s performance on the development dataset. Our system ranked 13th among the 35 sub- missions in the track we participated in, with a weighted F1 score of 80.32%. The top-ranked team achieved a score of 82.62%. Although our model performed excellently on training and development data, it did not perform well on the test datasets, i.e., it could not make accurate predictions on new data. 2 Background Task A: monolingual sentiment classification was the sub-task in which we participated. The task required us to determine the polarity (positive, neg- ative, or …

Content truncated to 500 words