feel it italian sentiment

Introduction

FEEL-IT is a sentiment and emotion classification model for Italian text, specifically designed to analyze emotions such as anger, fear, joy, and sadness in Twitter posts. The model provides a nuanced understanding of sentiment beyond a simple positive or negative classification. It is part of a larger effort to advance sentiment analysis in Italian, offering both a model and a Python package for easy integration into projects.

Architecture

The FEEL-IT model is based on the UmBERTo architecture, which is a transformer-based model fine-tuned for Italian sentiment analysis. The fine-tuning was conducted on a newly created dataset called FEEL-IT, which includes a wide range of annotated tweets.

Training

The model was trained by annotating 2,037 tweets with emotion labels. Performance evaluation was conducted using the SENTIPOLC16 Evalita dataset, where emotions were mapped to binary sentiment classes: joy to positive, and anger, fear, and sadness to negative. The FEEL-IT dataset achieved a Macro-F1 score of 0.81 and an accuracy of 0.84, outperforming other configurations.

Guide: Running Locally

1. Install Hugging Face Transformers: Ensure you have the transformers library installed via pip.

   pip install transformers
   

2. Load the Model: Use the pipeline from the transformers library to load and utilize the model.

   from transformers import pipeline
   classifier = pipeline("text-classification", model='MilaNLProc/feel-it-italian-sentiment', top_k=2)
   prediction = classifier("Oggi sono proprio contento!")
   print(prediction)
   

3. Cloud GPUs: For faster inference, consider using cloud GPUs from providers like AWS, Google Cloud, or Azure.

License

The model and related resources are subject to the terms outlined in Twitter's commercial terms, which can be accessed at Twitter Developer Terms.