changed back to original model

app.py

@@ -7,10 +7,10 @@ from typing import List
 # Initialize FastAPI
 app = FastAPI()
 
-# Load Longformer sentiment analysis model
+# Load BERT multilingual sentiment analysis model
 sentiment_pipeline = pipeline(
     "sentiment-analysis", 
-    model="allenai/longformer-base-4096"
+    model="nlptown/bert-base-multilingual-uncased-sentiment"
 )
 
 # Request models
@@ -28,11 +28,13 @@ class SentimentResponse(BaseModel):
     sentiment_label: str
     confidence: float
 
-# Mapping Longformer labels to a floating-point scale
+# Mapping BERT star ratings to sentiment scores
 LABEL_MAP = {
-    "LABEL_0": -1.0,  # Negative
-    "LABEL_1": 0.0,   # Neutral
-    "LABEL_2": 1.0    # Positive
+    "1 star": -1.0,  # Very Negative
+    "2 stars": -0.5,  # Negative
+    "3 stars": 0.0,   # Neutral
+    "4 stars": 0.5,   # Positive
+    "5 stars": 1.0    # Very Positive
 }
 
 @app.post("/analyze", response_model=SentimentResponse)
@@ -40,11 +42,11 @@ def analyze_sentiment(request: SentimentRequest):
     try:
         # Get model prediction
         result = sentiment_pipeline(request.text)[0]
-        label = result["label"]
+        label = result["label"].lower()
         score = result["score"]
         
-        # Convert Longformer labels to floating-point scores
-        sentiment_score = LABEL_MAP[label]
+        # Convert BERT star rating to floating-point sentiment score
+        sentiment_score = LABEL_MAP.get(label, 0.0)  # Default to neutral (0.0) if label is unexpected
         confidence = round(score, 3)
         
         return SentimentResponse(
@@ -62,11 +64,11 @@ def analyze_sentiment_batch(request: BatchSentimentRequest):
         responses = []
         for post in request.posts:
             result = sentiment_pipeline(post.text)[0]
-            label = result["label"]
+            label = result["label"].lower()
             score = result["score"]
             
-            # Convert Longformer labels to floating-point scores
-            sentiment_score = LABEL_MAP[label]
+            # Convert BERT star rating to floating-point sentiment score
+            sentiment_score = LABEL_MAP.get(label, 0.0)
             confidence = round(score, 3)
             
             responses.append(SentimentResponse(
@@ -82,4 +84,4 @@ def analyze_sentiment_batch(request: BatchSentimentRequest):
 # Root endpoint
 @app.get("/")
 def root():
-    return {"message": "Longformer Sentiment Analysis API is running!"}
+    return {"message": "BERT Multilingual Sentiment Analysis API is running!"}