Update app.py

app.py

@@ -109,8 +109,12 @@ Rules (priority order, MUST follow strictly):
 5. Otherwise:
    - Respond "positive" if the review is praising.
    - Respond "neutral" if it is neither.
+   
+6. If the review mentions fraud, repacked, fake, reputation damage,
+   or loss of trust related to the company or seller,
+   ALWAYS respond "negative" even if product praise exists.
 
-6. Respond with only one word.
+7. Respond with only one word.
 Do NOT explain. Do NOT add anything else.
 
 Review: {review_text}
@@ -207,9 +211,12 @@ def validate_review(description, rating, image_paths=None):
         (data['rating'].astype(str).str.contains('4|5'))
     )
     conditions_accepted_positive_any_rating = data['LLM'].str.contains('positive', case=False)
-    conditions_accepted_neutral_high = (data['LLM'].str.contains('neutral', case=False)) & (data['rating'].astype(str).str.contains('1|2|3|4|5'))
     description_null = data['description'].astype(str).str.strip().str.lower() == "null"
     conditions_accepted_null = description_null & data['rating'].astype(str).str.contains('3|4|5')
+    conditions_accepted_neutral_with_positive_keywords = (
+                        data['LLM'].str.contains('neutral', case=False) &
+                        data['description'].str.contains('|'.join(Positive_Strings), case=False)
+                    )
 
     # --- REJECT CONDITIONS ---
     conditions_rejected_1 = (data['rating'].astype(str).str.contains('1|2')) & (data['description'].astype(str).str.len() == 0)
@@ -218,16 +225,17 @@ def validate_review(description, rating, image_paths=None):
                             (data['LLM'].str.contains('positive|negative|neutral', case=False))
     conditions_rejected_3 = (data['LLM'].str.contains('negative', case=False)) & \
                             (data['description'].astype(str).str.contains('|'.join(Negative_Strings), case=False))
-    conditions_rejected_4 = (data['LLM'].str.contains('neutral', case=False)) & (data['rating'].astype(str).str.contains('3'))
+    conditions_rejected_4 = ( (data['LLM'].str.contains('neutral', case=False)) & (data['rating'].astype(str).str.contains('1|2')) & ~(data['description'].str.contains('|'.join(Positive_Strings), case=False)) )
     conditions_rejected_5 = (data['LLM'].str.contains('negative', case=False)) & ~(data['description'].astype(str).str.contains('|'.join(Negative_Strings), case=False))
     conditions_rejected_null = description_null & data['rating'].astype(str).str.contains('1|2')
+    conditions_rejected_neutral_negative_words = (data['LLM'].str.contains('neutral')) &  (data['description'].str.contains('|'.join(Negative_Strings), case=False))
 
     # Final flags
     conditions_accepted = conditions_accepted_1 | conditions_accepted_2 | conditions_accepted_3 | \
-                          conditions_accepted_4 | conditions_accepted_neutral_high | \
-                          conditions_accepted_positive_any_rating | conditions_accepted_null
+                          conditions_accepted_4 | \
+                          conditions_accepted_positive_any_rating | conditions_accepted_null |conditions_accepted_neutral_with_positive_keywords
     conditions_rejected = conditions_rejected_1 | conditions_rejected_2 | conditions_rejected_3 | \
-                          conditions_rejected_4 | conditions_rejected_5 | conditions_rejected_null
+                          conditions_rejected_4 | conditions_rejected_5 | conditions_rejected_null |conditions_rejected_neutral_negative_words
     conditions_ignored = ((~(conditions_rejected | conditions_accepted)) |
                           (data['description'].astype(str).str.contains('|'.join(ignore_keywords), case=False)))