license: apache-2.0
Hugging Face Armenian News Sentiment Dataset Repository
Introduction
This repository contains a collection of Armenian texts annotated with sentiment information (negative, positive, or negative). The collection contains training, development and test sets, and was created at the Center of Advanced Technologies at Russian-Armenian University.
Training and Development Data
For training and development datasets, we translated the Stanford Sentiment Treebank, MELD, and SLSD datasets into Armenian using Google Translate. The translated data was split between train/dev linearly by hand at 80%/20%.
| Dataset | Total Examples | Negative | Neutral | Positive |
|---|---|---|---|---|
| Training | 15,983 | 4,574 | 5,773 | 5,636 |
| Development | 4,179 | 1,184 | 1,558 | 1,437 |
Test Dataset
The test dataset consists of Armenian news articles collected from various Armenian news pages. Each news article is annotated by a native speaker with a sentiment label indicating whether the sentiment expressed in the article is negative (0), neutral (1), or positive (2). The sentiment labels were assigned based on natural language understanding and the subjective judgment of the annotators. Each example was processed by at least 2 annotators. We have included only those examples in the dataset where all annotators unanimously agreed on the sentiment labels. The test set contains 956 examples in total (295 negative, 476 neutral, 185 positive).
Repository Information
- The repository contains the annotated train, development and test datasets. The authors hope that the resources provided in this repository can serve as a starting point for researchers and developers interested in sentiment analysis for the Armenian language.
- The dataset was created by Elen Petikyan, with the help of a group of annotators (the full list of contributors is provided in CONTRIBUTING.md). The dataset creation process was supervised by Tsolak Ghukasyan.
Contact Information
For any inquiries or feedback related to this dataset repository, please contact Elen Petikyan (ellen.petikyan@gmail.com).