Create README.md
Browse files
README.md
ADDED
|
@@ -0,0 +1,22 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# sentiment_analysis_bert_multilingual
|
| 2 |
+
|
| 3 |
+
## Overview
|
| 4 |
+
This model is a fine-tuned version of the Multilingual BERT (mBERT) base model. It is designed to classify the sentiment of text across 100+ languages into three categories: Negative, Neutral, and Positive.
|
| 5 |
+
|
| 6 |
+
## Model Architecture
|
| 7 |
+
The model utilizes the standard BERT-base architecture:
|
| 8 |
+
- **Layers**: 12 Transformer blocks
|
| 9 |
+
- **Hidden Size**: 768
|
| 10 |
+
- **Attention Heads**: 12
|
| 11 |
+
- **Parameters**: ~177M
|
| 12 |
+
It includes a sequence classification head on top of the hidden state of the `[CLS]` token.
|
| 13 |
+
|
| 14 |
+
## Intended Use
|
| 15 |
+
- Social media monitoring for global brands.
|
| 16 |
+
- Customer feedback analysis in multilingual support tickets.
|
| 17 |
+
- Market research across different geographical regions.
|
| 18 |
+
|
| 19 |
+
## Limitations
|
| 20 |
+
- **Context Window**: Limited to 512 tokens; longer texts will be truncated.
|
| 21 |
+
- **Sarcasm**: May struggle with highly idiomatic or sarcastic expressions in low-resource languages.
|
| 22 |
+
- **Bias**: Subject to biases present in the Wikipedia and BookCorpus datasets used for pre-training.
|