metadata

license: cc-by-4.0
tags:
  - sentiment-classification
  - telugu
  - xlm-r
  - multilingual
  - baseline
language: te
datasets:
  - DSL-13-SRMAP/TeSent_Benchmark-Dataset
model_name: XLM-R_WOR

XLM-R_WOR: XLM-RoBERTa Telugu Sentiment Classification Model (Without Rationale)

Model Overview

XLM-R_WOR is a Telugu sentiment classification model based on XLM-RoBERTa (XLM-R), a general-purpose multilingual transformer developed by Facebook AI.
The "WOR" in the model name stands for "Without Rationale", indicating that this model is trained only with sentiment labels from the TeSent_Benchmark-Dataset and does not use human-annotated rationales.

Model Details

Architecture: XLM-RoBERTa (transformer-based, multilingual)
Pretraining Data: 2.5TB of filtered Common Crawl data across 100+ languages, including Telugu
Pretraining Objective: Masked Language Modeling (MLM), no Next Sentence Prediction (NSP)
Fine-tuning Data: TeSent_Benchmark-Dataset, using only sentence-level sentiment labels (positive, negative, neutral); rationale annotations are disregarded
Task: Sentence-level sentiment classification (3-way)
Rationale Usage: Not used during training or inference ("WOR" = Without Rationale)

Intended Use

Primary Use: Benchmarking Telugu sentiment classification on the TeSent_Benchmark-Dataset, especially as a baseline for models trained without rationales
Research Setting: Suitable for cross-lingual and multilingual NLP research, as well as explainable AI in low-resource settings

Why XLM-R?

XLM-R is designed for cross-lingual understanding and contextual modeling, providing strong transfer learning capabilities and improved downstream performance compared to mBERT. When fine-tuned with local Telugu data, XLM-R delivers solid results for sentiment analysis.
However, Telugu-specific models like MuRIL or L3Cube-Telugu-BERT may offer better cultural and linguistic alignment for purely Telugu tasks.

Performance and Limitations

Strengths:

Strong transfer learning and contextual modeling for multilingual NLP
Good performance for Telugu sentiment analysis when fine-tuned with local data
Useful as a cross-lingual and multilingual baseline

Limitations:

May be outperformed by Telugu-specific models for culturally nuanced tasks
Requires sufficient labeled Telugu data for best performance
Since rationales are not used, the model cannot provide explicit explanations for its predictions

Training Data

Dataset: TeSent_Benchmark-Dataset
Data Used: Only the Content (Telugu sentence) and Label (sentiment label) columns; rationale annotations are ignored for XLM-R_WOR training

Language Coverage

Language: Telugu (te)
Model Scope: This implementation and evaluation focus strictly on Telugu sentiment classification

Citation and More Details

For detailed experimental setup, evaluation metrics, and comparisons with rationale-based models, please refer to our paper.

License

Released under CC BY 4.0.