Bert Multilingual Go Emtions

🚀 Multilingual (English and Chinese) GoEmotions Classification Model

This repository presents a fine - tuned BERT model for cross - language emotion classification on the GoEmotions dataset. It's trained on a multilingual dataset with English and Chinese texts, and can classify text into 28 different emotion categories.

✨ Features

• Multilingual Support: Capable of handling both English and Chinese texts.
• Rich Emotion Classification: Can classify text into 28 emotion categories, such as 'admiration', 'amusement', 'anger', etc., as per the GoEmotions taxonomy.

📦 Installation

No specific installation steps are provided in the original document.

💻 Usage Examples

Basic Usage

from transformers import AutoTokenizer, AutoModelForSequenceClassification
from transformers import pipeline

# Load the model and tokenizer
tokenizer = AutoTokenizer.from_pretrained("SchuylerH/bert-multilingual-go-emtions")
model = AutoModelForSequenceClassification.from_pretrained("SchuylerH/bert-multilingual-go-emtions")

text = "I love you."
nlp = pipeline("sentiment-analysis", model = model, tokenizer = tokenizer)

result = nlp(text)

print(result)

📚 Documentation

Model Performance

The model shows high performance on the validation set, with the following scores:

• Accuracy: 85.95%
• Precision: 91.99%
• Recall: 89.56%
• F1 Score: 90.17% These results indicate the model's high accuracy and precision in predicting the correct emotion category for a given input text, regardless of the language (English or Chinese).

Training data

The dataset used for training the model is a combined dataset of the original English GoEmotions dataset and a machine - translated Chinese version of the GoEmotions dataset. The dataset is split into two parts:

• Labeled data: Used for initial training. It includes both English and machine - translated Chinese samples. This labeled data is further split into a training set (80%) and a validation set (20%).
• Unlabeled data: Used for making predictions and adding confidently predicted samples to the training data. It includes both English and machine - translated Chinese samples.

Training

The model is trained in two stages:

1. Initial training on the labeled data.
2. Predictions are made on the unlabeled data, and the most confidently predicted samples are added to the training data. The model is then retrained on this updated labeled data. The model is trained for a total of 20 epochs (10 epochs for each stage). Precision, recall, and F1 score are logged during training.

📄 License

This model is released under the apache - 2.0 license.

📋 Information Table