AUTOMATIC MULTI LABEL CLASSIFICATION FOR AFAAN OROMO DOCUMENT USING DEEP LEARNING APPROACH

Abstract

Advisor: Mr. Kasahun Abdisa (Phd candidate) Document classification is a technique which classifies textual information into a predefined set of categories. With the continuously increasing amount of online information, there is a pressing need to structure information. Automatic document classification is an inevitable solution in this regard. However, the present approaches are multi label document classification study aims to achieve news document classification. In this paper we present an automatic document classification using deep learning approach, considering the multi-label document classification problem. Therefore, we are motivated to design, develop, and implement automatic multi-label classification for Afaan Oromo document using the Deep Learning algorithm. The model takes document as input and classifies it to the predefined labels/categories based on the content of the document. Document classification is a technique that classifies textual information into a predefined set of classes. The reason to select „Afaan Oromo" is that even though Afaan Oromo has a large number of speakers it is an under-resourced language. For this study, 7151 documents were collected from Oromia Broadcasting Network (OBN) and Fana Broadcasting corporate (FBC) media and class labeled into twelve classes namely Barnoota, Teeknoloojii, Balaa, Fayyaa, Hawaasaa, Qonnaa, Siyaasa, Seera fi haqa, Dinaagdee, Biizinasii, Baankii, and Ispoorti. The research followed an experimental approach to determine the best deep learning algorithm. The deep learning techniques experimented with in this research are Convolutional Neural Network (CNN), Recurrent Neural Network (RNN), and Long Short-Term Memory (LSTM) techniques. To evaluate the performance of each technique, the researcher used several performance evaluation metrics such as Accuracy, F-score, Precession, and Recall. The feature extraction techniques are used as pre-trained word embedding (Word2vec, and FastText) methods. We used the Word2vec feature extraction and accuracy for performance evaluation because they achieved the highest performance result. The CNN technique achieved on 12 category accuracy of 85.3%, RNN achieved on 12 category accuracy of 84.7% and LSTM achieved on 12 category accuracy of 86.4% by using word2vec feature extraction techniques. According to the classification performance result from the entire techniques applied, the LSTM technique achieved the highest accuracy and we used LSTM to deploy our models. Keywords: Multi-Label document classification, Afaan Oromo, Deep Learning, Word embedding