Datasets:

DSL-13-SRMAP
/

TeSent_Benchmark-Dataset

TeSent_Benchmark-Dataset is a large-scale, open Telugu sentiment classification benchmark, developed to advance robust and interpretable sentiment analysis research for low-resource languages. While primarily created for sentence-level sentiment classification, this dataset is also a valuable resource for general text classification and related tasks in Telugu NLP.

Overview

Language: Telugu
Size: 22,505 labeled examples
Task: Sentence-level, three-class sentiment classification (positive, negative, neutral)
Explainability: Each example is annotated with human-provided rationales—text snippets that justify the assigned sentiment label.

Features

Labeled Sentences: Each data sample contains a Telugu sentence and its sentiment label (positive, negative, or neutral).
Human-Annotated Rationales: For every sentence, annotators have highlighted the specific text fragments that they found decisive for their sentiment decision, enabling explainable AI research.
Versatility: While designed for sentiment classification, the dataset is suitable for broader text classification, rationale extraction, and other NLP tasks in Telugu.

Format

Content	Rationale	Label
Telugu sentence	Human-annotated text snippet(s)	Sentiment class (positive/negative/neutral)

Applications

Benchmarking Telugu Sentiment Models: Establish a standard for model comparison and progress tracking.
Explainability Research: Train and evaluate models that can provide human-interpretable sentiment explanations.
Text Classification & NLP Tasks: Useful for training and evaluating models for general text classification, rationale extraction, and other Telugu natural language processing tasks.

Citation

If you use this dataset in your research, please cite as follows:

@inproceedings{tesent,
  title={TeSent: A Benchmark Dataset for Explainable Sentiment Classification in Telugu},
  author={Your Name and Collaborators},
  year={2025},
  note={https://github.com/DSL-13-SRMAP/TeSent_Benchmark-Dataset}
}