sentiment.py

# sentiment.py
from transformers import pipeline

# Create (singleton) sentiment pipeline
def get_sentiment_pipeline():
    # uses distilbert fine-tuned on SST-2 by default
    return pipeline("sentiment-analysis", return_all_scores=False)

def detect_sentiment(text, nlp=None):
    """
    Returns a simplified sentiment: 'positive', 'negative', or 'neutral' (if uncertain).
    Also returns the raw pipeline output.
    """
    if nlp is None:
        nlp = get_sentiment_pipeline()
    result = nlp(text)[0]  # {'label': 'POSITIVE', 'score': 0.99}
    label = result.get("label", "").lower()
    score = float(result.get("score", 0.0))

    # Map labels -> simplified categories
    if label.startswith("pos"):
        sentiment = "positive"
    elif label.startswith("neg"):
        sentiment = "negative"
    else:
        sentiment = "neutral"

    # If the model is unsure, label as neutral (adjust threshold if needed)
    if score < 0.6:
        sentiment = "neutral"

    return {"sentiment": sentiment, "label": result.get("label"), "score": score}