Emotion Detection for Afaan Oromo Using Deep Learning

Abstract

Emotion detection in text has become more popular due to its various useful applications in a different area, such as tracking product perception, public opinion detection about political tendencies, stock market monitoring, text summarization, information extraction, recommendation system, question answering and etc. However, manually identifying emotion of a million of people and aggregating them towards a rapid and efficient decision is quite a challenging task due to the rapid growth of social media user. This study aimed to develop Afaan Oromo emotion detection model in order to tackle this challenge. This study adopts artificial neural network approach. We used python tools with Keras library. We conduct our experiments on five emotion class (anger(arii), love(jaalala), joy(gamachu), disgust(jibba), and sadness(gadda)) by collecting a total of 1005 emotional sentence of Afaan Oromo language that have been manually annotated. The sentence has been scraped from different official Facebook page such as Oromia Broadcasting Network (OBN) pages, Fana Broadcasting Corporation (FBC) Afaan Oromo page, and British Broadcasting Corporation (BBC) Afaan Oromo pages using Facepager tools by creating Facebook API id. After collecting these data all preprocessing steps like tokenization, stop word removal and normalization have been undertaken. We used word embedding’s for feature extraction of preprocessed data. Subsequently, we have applied three artificial neural network algorithms such as Feed forward neural network, long short-term Memory and Bidirectional long short-term memory for classification purpose of the vectorized sentence into their emotion class. We compared the three artificial neural network algorithms and found out that Bidirectional long short-term memory achieved the best performance. We have achieved an average accuracy of 66%, 78%, 83% using Feed Forward Neural Network, Long Short-Term Memory and Bidirectional Long Short-Term Memory respectively. Based on experimental result, the researcher concluded that increasing amount of dataset, tuning hyper parameters properly and trying by different algorithms can, in some case, improve the performance of the model. Keywords : Emotion Identification, Afaan Oromo, Artificial Neural Network, Social Media DOI: 10.7176/NMMC/92-01 Publication date: August 31 st 2020