Sentiment Analysis with DistilBERT

Using Hugging Face transformers pipeline for sentiment analysis with DistilBERT model.

This notebook demonstrates how to perform sentiment analysis using Hugging Face’s pipeline API with a pre-trained DistilBERT model.

Importing Pipeline

Import the pipeline function from transformers to create a ready-to-use sentiment analysis model.

from transformers import pipeline

Creating Sentiment Analysis Pipeline

Create a sentiment analysis pipeline using DistilBERT, a smaller and faster version of BERT that’s been fine-tuned on the Stanford Sentiment Treebank dataset.

sentiment_analyzer = pipeline('sentiment-analysis', model='distilbert-base-uncased-finetuned-sst-2-english')

Device set to use cuda:0

texts = [
    'I love to play and watch cricket',
    'I hate when virat kohli misses a century'
]

results = sentiment_analyzer(texts)
for res in results:
    print(res)

{'label': 'POSITIVE', 'score': 0.9997660517692566}
{'label': 'NEGATIVE', 'score': 0.9990317821502686}

Understanding the Results

The model returns: - LABEL: Either ‘POSITIVE’ or ‘NEGATIVE’ - Score: Confidence level (0-1) for the prediction

Higher scores indicate more confidence in the classification.

Best Practices

• Test with various text types to understand model behavior
• Consider domain-specific fine-tuning for specialized use cases
• Be aware that models may have biases based on training data

Summary

This notebook demonstrated: - Using Hugging Face pipeline for sentiment analysis - Working with the DistilBERT model - Analyzing multiple texts at once - Interpreting confidence scores

The pipeline API makes sentiment analysis accessible with just a few lines of code!

Testing with Sample Texts

Test the sentiment analyzer with cricket-related texts to see how it classifies positive and negative sentiments.