app.py

from fastapi import FastAPI, HTTPException
from pydantic import BaseModel
from transformers import pipeline
from typing import List

# Initialize FastAPI
app = FastAPI()

# Load your fine-tuned Longformer model
sentiment_pipeline = pipeline(
    "text-classification", 
    model="spacesedan/reddit-sentiment-analysis-longformer"
)

# Request models
class SentimentRequest(BaseModel):
    content_id: str
    text: str

class BatchSentimentRequest(BaseModel):
    posts: List[SentimentRequest]

# Response model
class SentimentResponse(BaseModel):
    content_id: str
    sentiment_score: float
    sentiment_label: str
    confidence: float

# Updated label-to-score mapping
LABEL_MAP = {
    "very negative": -1.0,
    "negative": -0.7,
    "neutral": 0.0,
    "positive": 0.7,
    "very positive": 1.0
}

def normalize_prediction(label: str, confidence: float) -> (float, str):
    label = label.lower()
    score = LABEL_MAP.get(label, 0.0)

    # Confidence-based fallback to neutral
    if confidence < 0.6 and -0.7 < score < 0.7:
        return 0.0, "neutral"
    return score, label

@app.post("/analyze", response_model=SentimentResponse)
def analyze_sentiment(request: SentimentRequest):
    try:
        result = sentiment_pipeline(request.text)[0]
        confidence = round(result["score"], 3)
        sentiment_score, sentiment_label = normalize_prediction(result["label"], confidence)

        return SentimentResponse(
            content_id=request.content_id,
            sentiment_score=sentiment_score,
            sentiment_label=sentiment_label,
            confidence=confidence
        )
    except Exception as e:
        raise HTTPException(status_code=500, detail=str(e))

@app.post("/analyze_batch", response_model=List[SentimentResponse])
def analyze_sentiment_batch(request: BatchSentimentRequest):
    try:
        responses = []
        for post in request.posts:
            result = sentiment_pipeline(post.text)[0]
            confidence = round(result["score"], 3)
            sentiment_score, sentiment_label = normalize_prediction(result["label"], confidence)

            responses.append(SentimentResponse(
                content_id=post.content_id,
                sentiment_score=sentiment_score,
                sentiment_label=sentiment_label,
                confidence=confidence
            ))
        return responses
    except Exception as e:
        raise HTTPException(status_code=500, detail=str(e))

@app.get("/")
def root():
    return {"message": "Reddit Sentiment Analysis API (Longformer 5-point) is running!"}