Tshephisho Joseph Sefara
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Tshephisho Joseph Sefara has a total of 29 co-authors. Here are the top co-authors by number of shared publications:
- Vukosi Marivate
University of Pretoria
6 shared publications - Mapitsi Roseline Rangata
Council for Scientific and Industrial Research
3 shared publications Abiodun Modupe
University of Pretoria South Africa
2 shared publicationsKeamogetswe Makhaya
University of Cape Town South Africa
2 shared publicationsMadimetja Jonas Manamela
Department of Computer Science, University of Limpopo, Polokwane, South Africa
2 shared publicationsRethabile Mokoena
North-West University South Africa
2 shared publications
Latest Publications
- 2 January 2026
- Published by Springer Nature
- p. 177-186
Investigating Gender Bias Using Artificial Intelligence Classification Models on RAVDESS Dataset
- 2 January 2026
- Published by Springer Nature
- p. 177-186
Investigating Gender Bias Using Artificial Intelligence Classification Models on RAVDESS Dataset
- 24 April 2024
- Published by Springer Nature
- p. 283-296
Text Summarisation for Low-Resourced Languages, A Review
- 24 April 2024
- Published by Springer Nature
- p. 283-296
Text Summarisation for Low-Resourced Languages, A Review
- 12 February 2024
- Published by Springer Nature
- p. 248-260
Classification of Exaggerated News Headlines
- 20 December 2023
- Published by Springer Nature
- p. 468-482
Domain-Specific Sentiment Analysis of Tweets Using Machine Learning Methods
- 20 December 2023
- Published by Springer Nature
- p. 468-482
Domain-Specific Sentiment Analysis of Tweets Using Machine Learning Methods
- 4 October 2023
- Published by Springer Nature
- p. 951-962
A Natural Language Processing Technique to Identify Exaggerated News Titles
- 4 October 2023
- Published by Springer Nature
- p. 951-962
A Natural Language Processing Technique to Identify Exaggerated News Titles
- 1 March 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 127-132
Transformer-based Machine Translation for Low-resourced Languages embedded with Language Identification
Recent research on the development of machine translation (MT) models has resulted in state-of-the-art performance for many resourced European languages. However, there has been a little focus on applying these MT services to low-resourced languages. This paper presents the development of neural machine translation (NMT) for low-resourced languages of South Africa. Two MT models, JoeyNMT and transformer NMT with self-attention are trained and evaluated using BLEU score. The transformer NMT with self-attention obtained state-of-the-art performance on isiNdebele, SiSwati, Setswana, Tshivenda, isiXhosa, and Sepedi while JoeyNMT performed well on isiZulu. The MT models are embedded with language identification (LID) model that presets the language for translation models. The LID models are trained using logistic regression and multinomial naive Bayes (MNB). MNB classifier obtained an accuracy of 99% outperforming logistic regression which obtained the lowest accuracy of 97%.
- 1 March 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 127-132
Transformer-based Machine Translation for Low-resourced Languages embedded with Language Identification
Recent research on the development of machine translation (MT) models has resulted in state-of-the-art performance for many resourced European languages. However, there has been a little focus on applying these MT services to low-resourced languages. This paper presents the development of neural machine translation (NMT) for low-resourced languages of South Africa. Two MT models, JoeyNMT and transformer NMT with self-attention are trained and evaluated using BLEU score. The transformer NMT with self-attention obtained state-of-the-art performance on isiNdebele, SiSwati, Setswana, Tshivenda, isiXhosa, and Sepedi while JoeyNMT performed well on isiZulu. The MT models are embedded with language identification (LID) model that presets the language for translation models. The LID models are trained using logistic regression and multinomial naive Bayes (MNB). MNB classifier obtained an accuracy of 99% outperforming logistic regression which obtained the lowest accuracy of 97%.
- Open Access
- 1 January 2021
- Published by Association for Computational Linguistics (ACL)
- p. 20-25
Practical Approach on Implementation of WordNets for South African Languages
Tshephisho Joseph Sefara, Tumisho Billson Mokgonyane, Vukosi Marivate. Proceedings of the 11th Global Wordnet Conference. 2021.
- 1 January 2021
- Published by Association for Computational Linguistics (ACL)
- p. 20-25
Open AccessPractical Approach on Implementation of WordNets for South African Languages
Tshephisho Joseph Sefara, Tumisho Billson Mokgonyane, Vukosi Marivate. Proceedings of the 11th Global Wordnet Conference. 2021.
- Open Access
Improving short text classification through global augmentation methods
We study the effect of different approaches to text augmentation. To do this we use 3 datasets that include social media and formal text in the form of news articles. Our goal is to provide insights for practitioners and researchers on making choices for augmentation for classification use cases. We observe that Word2vec-based augmentation is a viable option when one does not have access to a formal synonym model (like WordNet-based augmentation). The use of \emph{mixup} further improves performance of all text based augmentations and reduces the effects of overfitting on a tested deep learning model. Round-trip translation with a translation service proves to be harder to use due to cost and as such is less accessible for both normal and low resource use-cases.
Open AccessImproving short text classification through global augmentation methods
We study the effect of different approaches to text augmentation. To do this we use 3 datasets that include social media and formal text in the form of news articles. Our goal is to provide insights for practitioners and researchers on making choices for augmentation for classification use cases. We observe that Word2vec-based augmentation is a viable option when one does not have access to a formal synonym model (like WordNet-based augmentation). The use of \emph{mixup} further improves performance of all text based augmentations and reduces the effects of overfitting on a tested deep learning model. Round-trip translation with a translation service proves to be harder to use due to cost and as such is less accessible for both normal and low resource use-cases.
- 18 August 2020
- Published by Springer Nature
- p. 385-399
Improving Short Text Classification Through Global Augmentation Methods
- 18 August 2020
- Published by Springer Nature
- p. 385-399
Improving Short Text Classification Through Global Augmentation Methods
- Open Access
Low resource language dataset creation, curation and classification: Setswana and Sepedi -- Extended Abstract
The recent advances in Natural Language Processing have only been a boon for well represented languages, negating research in lesser known global languages. This is in part due to the availability of curated data and research resources. One of the current challenges concerning low-resourced languages are clear guidelines on the collection, curation and preparation of datasets for different use-cases. In this work, we take on the task of creating two datasets that are focused on news headlines (i.e short text) for Setswana and Sepedi and the creation of a news topic classification task from these datasets. In this study, we document our work, propose baselines for classification, and investigate an approach on data augmentation better suited to low-resourced languages in order to improve the performance of the classifiers.
Open AccessLow resource language dataset creation, curation and classification: Setswana and Sepedi -- Extended Abstract
The recent advances in Natural Language Processing have only been a boon for well represented languages, negating research in lesser known global languages. This is in part due to the availability of curated data and research resources. One of the current challenges concerning low-resourced languages are clear guidelines on the collection, curation and preparation of datasets for different use-cases. In this work, we take on the task of creating two datasets that are focused on news headlines (i.e short text) for Setswana and Sepedi and the creation of a news topic classification task from these datasets. In this study, we document our work, propose baselines for classification, and investigate an approach on data augmentation better suited to low-resourced languages in order to improve the performance of the classifiers.