Amharic News Text Classification Corpus

A collection of Amharic news articles presented in the Fidel orthography, structured into a CSV file containing both article text and corresponding news categories. The dataset also includes a standalone list of Amharic stop words, which has been curated, edited, and expanded by a native Amharic speaker using various academic sources. The foundation of this dataset draws on research published in 2021 concerning Amharic news text classification, with subsequent additions made to expand the corpus size. This resource is primarily designed for tasks related to text classification and deep linguistic analysis of Amharic.

The primary data file is named Amharic_corpus_merged_2023-04-16.csv and is approximately 260.01 MB in size. The dataset includes 61.9 thousand valid records. The file format is standard CSV. In addition to the corpus, a stop words list containing 714 unique tokens is included. Note that this stop word list is general for the Amharic language and is not tailored specifically to this particular news text corpus. The dataset is expected to receive updates on an annual basis.

This dataset is suitable for developing and testing machine learning models for natural language processing (NLP), particularly news text classification, including advanced techniques like LSTM. It is an ideal resource for research into low-resource language processing and for building robust Amharic text analysis systems.

The data covers Amharic news texts classified across seven distinct topics. The categories and their associated text counts from the original sources include: 20,674 Local News (ሀገር አቀፍ ዜና), 10,411 Sport (ስፖርት), 9,325 Politics (ፖለቲካ), 6,543 World News (ዓለም አቀፍ ዜና), 3,894 Business (ቢዝነስ), and 635 Entertainment (መዝናኛ). Our own additions include 5,276 Business texts and 5,156 Politics texts. The latest update to the collection was noted in April 2023.

CC0: Public Domain

Intended users include data scientists focused on classification models, machine learning researchers studying under-resourced languages, academic linguists interested in Amharic morphology and syntax, and developers creating region-specific NLP applications.

Original Data Source: Amharic News Text Classification Corpus