Detection of Emotions in Afan Oromo Social Media Texts Using Deep Learning Method

Abstract

Emotion analysis in foreign languages is common because of its numerous useful applications in commercial activities and decision-making. However, there was a lack of emotion-detection work for Afan Oromo language. Manually identifying and aggregating millions of social media users' emotions into a swift and effective decision-making process is challenging task. Thus, the main objective of this study was to detect emotion in the Afan Oromo social media texts. To achieve this, state-of-the-art deep learning models, namely convolutional neural networks (CNN), long short-term memory (LSTM), bidirectional long short-term memory (BiLSTM), and hybrid of them (CNN-LSTM and CNN-BiLSTM), were designed and investigated to select the best-suited model for emotion detection in Afan Oromo. Data was collected from official Facebook pages, then manually annotated, and preprocessed by using normalization, tokenization and stop-word removal. Word embedding was used for feature encoding, and Keras Python libraries were employed for implementation. The study’s results revealed that the proposed models performed well with accuracies of 92, 87, 88, 88 and 90 % for CNN, LSTM, BiLSTM, CNN-LSTM, and CNN-BiLSTM, respectively. Thus, the CNN model outperformed all the other models. It was also found out that the CNN model suited and gave a better result when it was worked on a small dataset and a short sequence of texts. The accuracy of the comparatively less performing models, particularly the performance of the hybrid models, can be increased through the construction of sufficient data, because they leverage the benefits of each of them.