A Sesotho news headlines dataset for sentiment analysis.

Area of Science:

• Natural Language Processing (NLP)
• Computational Linguistics
• African Language Technologies

Background:

• Sentiment Analysis (SA) research is advanced in high-resource languages but lags in low-resource African languages like Sesotho due to data scarcity.
• Limited annotated datasets hinder the development of language-specific NLP services for Sesotho speakers.

Discussion:

• This study presents the Sesotho News (SN) dataset, an annotated corpus for SA and Aspect-Based Sentiment Analysis (ABSA).
• The SN dataset comprises 4651 headlines for ABSA and 2401 headlines for SA, adhering to Lesotho's Sesotho orthography.
• Annotation quality is validated through Spearman's correlation and Cohen's Kappa Index, indicating high inter-annotator agreement.

Key Insights:

• The creation of the SN dataset addresses the critical need for resources in low-resource language NLP.
• The dataset supports research in Sentiment Analysis and Aspect-Based Sentiment Analysis for Sesotho.
• High inter-annotator agreement suggests the SN dataset is a reliable, gold-standard resource.

Outlook:

• The SN dataset is expected to spur further NLP research and development for the Sesotho language.
• This resource can facilitate the creation of language-specific NLP applications benefiting millions of speakers in Lesotho and South Africa.
• Future work may involve expanding the dataset and exploring advanced SA and ABSA models for Sesotho.