|
|
--- |
|
|
license: cc-by-4.0 |
|
|
tags: |
|
|
- sentiment-classification |
|
|
- telugu |
|
|
- multilingual |
|
|
- mbert |
|
|
- baseline |
|
|
language: te |
|
|
datasets: |
|
|
- DSL-13-SRMAP/TeSent_Benchmark-Dataset |
|
|
model_name: mBERT_WOR |
|
|
--- |
|
|
|
|
|
# mBERT_WOR: Telugu Sentiment Classification Model (Without Rationale) |
|
|
|
|
|
## Model Overview |
|
|
|
|
|
**mBERT_WOR** is a Telugu sentiment classification model based on Google's mBERT (BERT-base-multilingual-cased), trained specifically for sentence-level sentiment analysis **without rationale supervision**. The acronym "WOR" stands for "Without Rationale," indicating that this model was trained using only the sentiment labels and not the human-annotated rationales provided in the TeSent_Benchmark-Dataset. |
|
|
|
|
|
--- |
|
|
|
|
|
## Model Details |
|
|
|
|
|
- **Architecture:** mBERT (BERT-base-multilingual-cased, 12 layers, ~100M parameters) |
|
|
- **Pretraining Data:** Wikipedia articles in 104 languages (including Telugu), using Masked Language Modeling (MLM) and Next Sentence Prediction (NSP) objectives. |
|
|
- **Fine-tuning Data:** TeSent_Benchmark-Dataset (Telugu only), using only the sentence-level sentiment labels (positive, negative, neutral); rationale annotations are not used in training. |
|
|
- **Task:** Sentence-level sentiment classification (3-way) |
|
|
- **Rationale Usage:** Not used during training or inference |
|
|
|
|
|
--- |
|
|
|
|
|
## Intended Use |
|
|
|
|
|
- **Primary Use:** Benchmarking Telugu sentiment classification on the TeSent_Benchmark-Dataset, especially as a **baseline** for models trained with and without rationales. |
|
|
- **Research Setting:** Designed for academic research, particularly in low-resource and explainable NLP settings. |
|
|
|
|
|
--- |
|
|
|
|
|
## Performance and Limitations |
|
|
|
|
|
- **Strengths:** |
|
|
- Leverages shared multilingual representations, enabling cross-lingual transfer and reasonable performance for Telugu even with limited labeled data. |
|
|
- Serves as a robust baseline for Telugu sentiment tasks. |
|
|
- **Limitations:** |
|
|
- Not specifically optimized for Telugu morphology or syntax, which may impact its ability to capture fine-grained, language-specific sentiment cues. |
|
|
- May underperform compared to Telugu-specialized models such as IndicBERT or L3Cube-Telugu-BERT, especially for nuanced or idiomatic expressions. |
|
|
- Since rationales are not used, the model cannot provide explicit explanations for its predictions. |
|
|
|
|
|
--- |
|
|
|
|
|
## Training Data |
|
|
|
|
|
- **Dataset:** [TeSent_Benchmark-Dataset](https://github.com/DSL-13-SRMAP/TeSent_Benchmark-Dataset) |
|
|
- **Data Used:** Only the **Content** (Telugu sentence) and **Label** (sentiment label) columns; **rationale** annotations are ignored for mBERT_WOR training. |
|
|
|
|
|
--- |
|
|
|
|
|
## Language Coverage |
|
|
|
|
|
- **Language:** Telugu (the only language in the dataset) |
|
|
- **Note:** While mBERT is a multilingual model, this implementation and evaluation are strictly for 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](LICENSE). |
