EMOTION CLASSIFICATION IN AMHARIC SOCIAL MEDIA TEXT  COMMENTS USING DEEP LEARNING

SILESHI, BOGALE

dc.contributor.author	SILESHI, BOGALE
dc.date.accessioned	2022-03-18T06:49:06Z
dc.date.available	2022-03-18T06:49:06Z
dc.date.issued	2021-09
dc.identifier.uri	http://ir.bdu.edu.et/handle/123456789/13215
dc.description.abstract	Nowadays, social media has become a widespread way to exchange information, thoughts, feelings, and good memories to express our emotions through text without using a lot of words. Text emotion classification is one of the methods used to organize massively available emotions into a predefined category to maximize the utilization of information in social media. From previous research it is noted that a lot of work has been done about emotion classification for different languages; however, to the best of our knowledge, there is no work done in Amharic languages using text. The other motivational factor to conduct the study is the absence of benchmark corpora in Amharic text to build an emotion classification model. Moreover, it is difficult for comment holders to express the proper emotion they feel, and it is also for the social media authors to create a common understanding of emotion in the written text. This research aims to develop an emotion classification model in Amharic social media text comments using deep learning approaches. The main challenges in emotion classification is identifying better deep learning approaches for classification and the existence of similar text comments expressed in different emotions from the data source. We evaluate the model using 31,716 text comments for both training and testing to compare the models for emotion classification and annotate it with seven different classes. In our model, we use word2Vec in CBOW approach to convert each word to vector values in the corpus and to build the model and TensorFlow as the backend. As a result, the vector representations of words are used as input for the component that builds the deep network. We apply five deep learning approaches namely: long short-term memory, bidirectional long short-term memory, CNN-BILSTM, BIGRU, and convolutional neural network, models with word2Vec to our corpus. After experimentation and evaluation, the results show that the CNN model achieve better performance with 86% test accuracy and 86% f-score. For the future, we recommend to assess the influence of various current contextualized word embedding (e.g., GloVe, ELMO, BERT) on the effectiveness of our proposed deep learning model using different Amharic emotion classification datasets. Keywords: Natural Language Processing, emotion classification, deep learning, social media, word2Vec	en_US
dc.language.iso	en_US	en_US
dc.subject	computer science	en_US
dc.title	EMOTION CLASSIFICATION IN AMHARIC SOCIAL MEDIA TEXT COMMENTS USING DEEP LEARNING	en_US
dc.type	Thesis	en_US