Update app.py

app.py

@@ -13,14 +13,15 @@ client = OpenAI(
 
 # ---------------- SENTIMENT FUNCTION ---------------- #
 def get_sentiment(review_text: str) -> str:
-    prompt = f"""
+    prompt =f"""
 Classify sentiment of this review in English or Roman Urdu.
 Return EXACTLY one word: positive, negative, or neutral.
 
 Rules (priority order, MUST follow strictly):
 
 1. If the review contains any abusive, offensive, vulgar, or insulting words 
-   in English or Roman Urdu, ALWAYS respond with "negative".
+   in English or Roman Urdu (e.g., bad words, slang, or personal attacks),
+   ALWAYS respond with "negative".
    This overrides ALL other rules — even if the rest of the review is positive.
 
 2. If the review mentions problems with Priceoye's website, service, delivery,
@@ -28,15 +29,21 @@ Rules (priority order, MUST follow strictly):
    ALWAYS respond with "negative".
    This overrides everything else except abusive language.
 
-3. If both positive and negative opinions are present but not directly about
+3. If the review only criticizes or complains about a product 
+   (e.g., "yeh phone bekaar hai", "battery weak hai") 
+   but does NOT contain abusive or insulting words,
+   DO NOT treat it as negative — respond based on the overall tone:
+   - If it seems disappointed but polite → "neutral"
+   - If it praises other things → "positive"
+
+4. If both positive and negative opinions are present but not directly about
    Priceoye's website, service, or delivery → respond "neutral".
 
-4. Otherwise:
+5. Otherwise:
    - Respond "positive" if the review is praising.
-   - Respond "negative" if the review is complaining.
    - Respond "neutral" if it is neither.
 
-5. Respond with only one word.
+6. Respond with only one word.
 Do NOT explain. Do NOT add anything else.
 
 Review: {review_text}