Multi-Language Text Classification 🧠

This project is a simple text classification 📝 using multi-language BERT. 🇬🇧 | 🇧🇷

Project ☁️

Details of the BERT model used:

• BERT multilingual base model (uncased)

This project was inspired by this source:

• Documents Classification

Training and validation accuracy:

Note: We reached 98% accuracy.

Requirements 🛠️

It is necessary:

• Python 3.11
• Poetry
• Git

Note: A machine with a GPU is not required, but it is recommended to accelerate training.

Datasets 📊

It was used the BBC dataset to classify texts into the following labels:

• Business 💼
• Entertainment 🎬
• Sport ⚽
• Tech 💻
• Politics 🏛️

The texts in the dataset were first translated into Brazilian Portuguese, using the Google Translator API. After that, the English 🇬🇧 and Portuguese 🇧🇷 texts were combined to create a multilingual version.

Build and Running 🚀

Clone the project to your computer using Git and go to the project root folder:

git clone [email protected]:NeuroQuestAi/ml-text-classification.git && \
 cd ml-text-classification.git

Use poetry to access the project:

poetry shell

Install all dependencies:

poetry install && poetry update 

Run the model training and evaluation:

./train 

This will generate the torch model in the models folder. Then just test the predictions with the command:

./predictor 

Output example:

Text	Lang	Prediction
Os negócios são o tecido vital da economia...	🇧🇷	BUSINESS
A variedade de formas de entretenimento reflete...	🇧🇷	ENTERTAINMENT
Os valores como fair play, respeito e camaradagem são...	🇧🇷	SPORT
Desde a revolução digital até as últimas descobertas...	🇧🇷	TECH
A política reflete as diferentes visões, valores...	🇧🇷	POLITICS
Businesses are the lifeblood of the economy, where ideas...	🇬🇧	BUSINESS
From thrilling movies to engaging games, and soul-touching...	🇬🇧	ENTERTAINMENT
Sports are a universal passion that brings people...	🇬🇧	SPORT
Artificial intelligence, cloud computing, the Internet...	🇬🇧	TECH
Active citizen participation in political life...	🇬🇧	POLITICS

Note: Model settings are in the config.json file.

Authors 👨‍💻

• Ederson Corbari 👽