language:
- ps library_name: transformers pipeline_tag: text-classification tags:
- sentiment-analysis
- pashto
- romanized-pashto
- low-resource
- code-mixed
- xlm-roberta
- nlp datasets:
- Romanized_Pashto_Sentiment base_model:
- xlm-roberta-base
Romanized Pashto Sentiment Analysis
This model performs three-class sentiment classification on Romanized Pashto text written using the Latin/Roman script.
Model Description
This model is based on xlm-roberta-base and was fine-tuned for sentiment classification of Romanized Pashto.
Romanized Pashto refers to Pashto written using the Latin/Roman script rather than the Pashto Arabic-based script. Romanized Pashto can contain substantial spelling variation, informal writing, transliteration differences, and code-mixing with languages such as Urdu and English. These characteristics make sentiment classification challenging.
Intended Use
The model is intended for:
- Research and experimentation in Romanized Pashto NLP
- Sentiment analysis of Romanized Pashto text
- Low-resource language research
- Research on multilingual and code-mixed language processing
- Evaluation of transformer-based models on Romanized Pashto
This model is primarily intended for research purposes and should be evaluated on the target domain before being used in production systems.
Model Details
| Property | Description |
|---|---|
| Base model | xlm-roberta-base |
| Architecture | XLM-RoBERTa |
| Task | Text classification |
| Language | Romanized Pashto |
| Script | Latin/Roman |
| Number of classes | 3 |
| Classes | Positive, Negative, Neutral |
| Dataset | Romanized_Pashto_Sentiment |
Training
The model was fine-tuned from xlm-roberta-base for three-class sentiment classification.
The training data consists of Romanized Pashto text collected for sentiment analysis research. The dataset may contain informal spelling, transliteration variation, and code-mixed Urdu and English text.
Evaluation
The model achieved the following results on the evaluation set:
- Weighted F1: 0.878679
- Weighted F1: 87.87%
- Evaluation loss: 0.357379
- Training loss: 0.450354
The reported weighted F1 score reflects performance on the evaluation dataset used during the experiment.
Limitations
Romanized Pashto does not have a standardized spelling system. The same Pashto word may be written in several different Romanized forms.
Model performance may therefore be affected by:
- Spelling and transliteration variation
- Informal social-media language
- Code-mixing with Urdu and English
- Unseen vocabulary and expressions
- Domain differences between training and real-world text
- Ambiguous or context-dependent sentiment
The reported evaluation results should not be interpreted as representative of all Pashto speakers, domains, or Romanized Pashto writing styles.
The model may perform differently on text from domains that differ substantially from the evaluation data.
Research Context
This model is part of ongoing research on multilingual and low-resource NLP, with a particular focus on Romanized Pashto, code-mixed language, language resources, and model evaluation.
Romanized Pashto is used as a research setting for investigating challenges associated with low-resource and non-standardized language varieties.
Citation
If you use this model in academic research, please cite the associated research work when available.
License
Please check the repository settings for the applicable model and dataset license before redistributing or using the model commercially.
- Downloads last month
- 13