mssrchapelier/amharic-text-classifier

A news headline classifier for Amharic (2022). # `amharic-text-classifier` This is a Jupyter notebook applying a multinomial naïve Bayes classifier to news headlines in Amharic (for which the Amharic News Text Classification Dataset was used). Completed in 2022 as a personal project (as a master's student of theoretical and applied linguistics at Moscow State University at that moment). ### Output examples ``` --- pipeline: no_tfidf --- text: የኢትዮጵያ መንግሥት "ሰላም ለማምጣት እየሠራሁ ነው" አለ en: Ethiopian government: "We are working to bring peace" predicted class: Ethiopian news class probabilities: Ethiopian news: 98.77% business: 0.00% entertainment: 0.00% politics: 0.15% sports: 0.00% world news: 1.08% -------------------------------------------------- text: የሶማሊያው ሐሰን ሼኽ ሞሃሙድ ቃለ መሃላ ፈፀሙ en: Somalia's Hassan Sheikh Mohamud swears in predicted class: world news class probabilities: Ethiopian news: 0.00% business: 0.02% entertainment: 0.21% politics: 1.85% sports: 0.00% world news: 97.92% -------------------------------------------------- text: ፌስቡክ የጥላቻ ንግግሮችን የያዙ ማስታወቂያዎችን ይለይ እንደሆነ በቀረበለት ፈተ... en: Facebook fails again to detect hate speech in ads predicted class: world news class probabilities: Ethiopian news: 2.80% business: 4.02% entertainment: 1.00% politics: 8.84% sports: 0.00% world news: 83.33% -------------------------------------------------- text: ኢትዮጵያ በአፍሪካ ዋንጫ የማጣርያ ጨዋታ ግብፅን 2 - 0 አሸነፈች en: Ethiopia defeats Egypt 2-0 in Africa Cup of Nations qualifiers predicted class: sports class probabilities: Ethiopian news: 0.00% business: 0.00% entertainment: 0.00% politics: 0.00% sports: 100.00% world news: 0.00% ``` ### Performance on the test set