Logo Lanfrica

mathyasg/AmharicSentimentAnalysis

Domaine:

natural language processing

Type de record:

project
Créateur:
mat
Hôte:
A low-resource language project using data augmentation and prompt tuning # Amharic Sentiment Analysis with Prompt Tuning Low-resource binary sentiment classification for Amharic tweets using data augmentation and PEFT (Prompt Tuning). ## Current Status (Initial Version) - Macro F1: ~0.70–0.75 (after basic augmentation) - Trainable parameters: ~0.03% - Techniques: random word swap augmentation + prompt tuning ## Planned Improvements - Class weighting - Back-translation augmentation (NLLB) - Switch to LoRA - Better hyperparameter tuning - Gradio demo ## How to run 1. Open in Colab: 2. Change runtime to T4 GPU 3. Run all cells