Update sentiment_analyzer.py

sentiment_analyzer.py

@@ -67,8 +67,6 @@ Reason: [Brief explanation of your analysis]"""
             
             inputs = self.tokenizer(prompt, return_tensors="pt", truncation=True, max_length=1024)
             inputs = inputs.to(self.device)
-            
-            # --- MODIFIED: Get prompt length to slice output correctly ---
             prompt_length = inputs['input_ids'].shape[1]
 
             with torch.no_grad():
@@ -80,7 +78,6 @@ Reason: [Brief explanation of your analysis]"""
                     pad_token_id=self.tokenizer.eos_token_id
                 )
             
-            # --- MODIFIED: Decode *only* the new tokens, not the prompt ---
             new_tokens = outputs[0][prompt_length:]
             response = self.tokenizer.decode(new_tokens, skip_special_tokens=True)
             
@@ -91,36 +88,41 @@ Reason: [Brief explanation of your analysis]"""
             return self._fallback_sentiment(text)
 
     def _parse_llm_analysis(self, response):
-        """แยก sentiment, score, theme, impact และ explanation จาก LLM response"""
+        """
+        แยก sentiment, score, theme, impact และ explanation จาก LLM response
+        (เวอร์ชันแก้ไขให้ทนทานต่อ Markdown และข้อผิดพลาด)
+        """
         sentiment = "Neutral"
         score = 0.5
         theme = "Other"
         impact = "Neutral"
-        explanation = "Unable to parse" # Default explanation if parse fails
+        explanation = "Unable to parse"
 
         try:
-            sentiment_line = re.search(r'Sentiment:\s*(\w+)', response, re.IGNORECASE)
+            # --- MODIFIED: Made Regex more robust ---
+            # (Handles optional markdown "**" and optional colon ":")
+            
+            sentiment_line = re.search(r'\**Sentiment:?\**\s*(\w+)', response, re.IGNORECASE)
             if sentiment_line:
                 sentiment = sentiment_line.group(1).capitalize()
             
-            score_line = re.search(r'Score:\s*([\d.]+)', response)
+            score_line = re.search(r'\**Score:?\**\s*([\d.]+)', response)
             if score_line:
                 score = float(score_line.group(1))
                 score = max(0.0, min(1.0, score))
 
-            theme_line = re.search(r'Theme:\s*([\w\/ -]+)', response, re.IGNORECASE)
+            theme_line = re.search(r'\**Theme:?\**\s*([\w\/ -]+)', response, re.IGNORECASE)
             if theme_line:
                 theme = theme_line.group(1).strip()
 
-            impact_line = re.search(r'Impact:\s*(\w+)', response, re.IGNORECASE)
+            impact_line = re.search(r'\**Impact:?\**\s*(\w+)', response, re.IGNORECASE)
             if impact_line:
                 impact = impact_line.group(1).capitalize().strip()
 
-            # --- MODIFIED: More robust regex for Reason (captures multi-line) ---
-            reason_match = re.search(r'Reason:\s*(.*)', response, re.DOTALL | re.IGNORECASE)
+            reason_match = re.search(r'\**Reason:?\**\s*(.*)', response, re.DOTALL | re.IGNORECASE)
             if reason_match:
                 explanation = reason_match.group(1).strip()
-            # If parsing fails, explanation will remain "Unable to parse" or the last good value
+            # --- End of MODIFIED block ---
 
             if sentiment not in ["Positive", "Negative", "Neutral"]:
                 sentiment = "Neutral"