Update app.py

app.py

@@ -365,6 +365,7 @@ def split_content_in_half(text: str) -> tuple:
 def analyze_content_halves(model_manager, text: str, overall_result: Dict = None) -> Dict:
     """
     Analyze text by splitting it into two halves after cleaning
+    Uses BOTH models for ensemble predictions on each half for improved accuracy
     
     Args:
         model_manager: The ModelManager instance
@@ -387,11 +388,13 @@ def analyze_content_halves(model_manager, text: str, overall_result: Dict = None
         # Split into halves
         first_half, second_half = split_content_in_half(cleaned_text)
         
-        # Analyze first half
+        # Analyze first half using BOTH models (ensemble prediction)
+        logger.info("🔍 Analyzing first half with both models...")
         first_half_result = model_manager.classify_text(first_half)
         first_half_words = len(first_half.split())
         
-        # Analyze second half
+        # Analyze second half using BOTH models (ensemble prediction)
+        logger.info("🔍 Analyzing second half with both models...")
         second_half_result = model_manager.classify_text(second_half)
         second_half_words = len(second_half.split())
         
@@ -401,6 +404,10 @@ def analyze_content_halves(model_manager, text: str, overall_result: Dict = None
         first_model = first_half_result["predicted_model"]
         second_model = second_half_result["predicted_model"]
         
+        # Get top predictions from both halves for comparison
+        first_top5 = first_half_result.get("top_5_predictions", [])
+        second_top5 = second_half_result.get("top_5_predictions", [])
+        
         # Calculate average AI score from both halves
         avg_halves_ai_score = (first_ai + second_ai) / 2
         
@@ -410,6 +417,16 @@ def analyze_content_halves(model_manager, text: str, overall_result: Dict = None
         # Overall AI probability (use overall_result if provided, otherwise calculate)
         overall_ai_prob = overall_result["ai_percentage"] / 100 if overall_result else avg_halves_ai_score / 100
         
+        # Check model consistency across halves
+        models_agree = first_model == second_model
+        
+        # Calculate confidence boost from using both models
+        models_used = first_half_result.get("models_used", 1)
+        ensemble_confidence_boost = "High" if models_used > 1 else "Low"
+        
+        logger.info(f"✅ First half: {first_ai}% AI ({first_model}) | Second half: {second_ai}% AI ({second_model})")
+        logger.info(f"📊 Models used per half: {models_used} | Agreement: {models_agree}")
+        
         # ===== FINAL DECISION LOGIC =====
         verdict = None
         confidence = None
@@ -420,17 +437,19 @@ def analyze_content_halves(model_manager, text: str, overall_result: Dict = None
             verdict = "HUMAN"
             confidence = "High" if variance_between_halves < 15 else "Medium"
             reasoning = (
-                f"Both halves scored below 50% AI probability (First: {first_ai}%, Second: {second_ai}%). "
+                f"Both halves scored below 50% AI probability (First: {first_ai}%, Second: {second_ai}%) "
+                f"using ensemble prediction from {models_used} model(s). "
                 f"The second half was classified as human-written. "
                 f"Variance between halves is {variance_between_halves:.2f}%, indicating "
-                f"{'consistent human patterns' if variance_between_halves < 15 else 'some variation but still human-like'}."
+                f"{'consistent human patterns' if variance_between_halves < 15 else 'some variation but still human-like'}. "
+                f"Model predictions {'agree' if models_agree else 'differ'} across halves."
             )
         
         # Condition 2: Both halves > 50% AI AND second_half predicted_model is NOT "human"
         elif first_ai > 50 and second_ai > 50 and second_model.lower() != "human":
             verdict = "AI"
             
-            # Determine confidence based on scores
+            # Determine confidence based on scores and model agreement
             if first_ai > 80 and second_ai > 80:
                 confidence = "Very High"
             elif first_ai > 70 and second_ai > 70:
@@ -438,11 +457,21 @@ def analyze_content_halves(model_manager, text: str, overall_result: Dict = None
             else:
                 confidence = "Medium"
             
+            # Boost confidence if models agree
+            if models_agree and confidence != "Very High":
+                confidence_levels = ["Low", "Medium", "High", "Very High"]
+                current_idx = confidence_levels.index(confidence)
+                if current_idx < len(confidence_levels) - 1:
+                    confidence = f"{confidence} (boosted by model agreement)"
+            
             reasoning = (
-                f"Both halves scored above 50% AI probability (First: {first_ai}%, Second: {second_ai}%). "
+                f"Both halves scored above 50% AI probability (First: {first_ai}%, Second: {second_ai}%) "
+                f"using ensemble prediction from {models_used} model(s). "
                 f"The pattern matches {second_model} outputs. "
+                f"First half suggests {first_model} while second half suggests {second_model}. "
                 f"Variance between halves is {variance_between_halves:.2f}%, "
-                f"{'showing consistent AI patterns throughout' if variance_between_halves < 20 else 'with some variation in AI generation style'}."
+                f"{'showing consistent AI patterns throughout' if variance_between_halves < 20 else 'with some variation in AI generation style'}. "
+                f"{'Both halves agree on the AI model type, strengthening confidence' if models_agree else 'Different AI models detected in each half'}."
             )
         
         # Condition 3: Mixed results - one half AI, one half human
@@ -450,7 +479,8 @@ def analyze_content_halves(model_manager, text: str, overall_result: Dict = None
             verdict = "MIXED"
             confidence = "Low"
             reasoning = (
-                f"Mixed signals detected. First half: {first_ai}% AI, Second half: {second_ai}% AI. "
+                f"Mixed signals detected using {models_used} model(s) for ensemble prediction. "
+                f"First half: {first_ai}% AI ({first_model}), Second half: {second_ai}% AI ({second_model}). "
                 f"One portion appears AI-generated while the other seems human-written. "
                 f"This could indicate: partial AI assistance, human editing of AI content, "
                 f"or AI completion of human-started text. High variance of {variance_between_halves:.2f}% supports mixed authorship."
@@ -463,7 +493,8 @@ def analyze_content_halves(model_manager, text: str, overall_result: Dict = None
                 verdict = "LIKELY_HUMAN"
                 confidence = "Low"
                 reasoning = (
-                    f"Borderline case with scores near 50% threshold (First: {first_ai}%, Second: {second_ai}%). "
+                    f"Borderline case with scores near 50% threshold (First: {first_ai}%, Second: {second_ai}%) "
+                    f"analyzed using {models_used} model(s). "
                     f"Second half classified as human-written. The text shows characteristics of both "
                     f"human and AI writing. Variance: {variance_between_halves:.2f}%."
                 )
@@ -471,12 +502,13 @@ def analyze_content_halves(model_manager, text: str, overall_result: Dict = None
                 verdict = "LIKELY_AI"
                 confidence = "Low"
                 reasoning = (
-                    f"Borderline case with scores near 50% threshold (First: {first_ai}%, Second: {second_ai}%). "
+                    f"Borderline case with scores near 50% threshold (First: {first_ai}%, Second: {second_ai}%) "
+                    f"analyzed using {models_used} model(s). "
                     f"Pattern suggests {second_model} but confidence is low. "
                     f"Variance: {variance_between_halves:.2f}%."
                 )
         
-        # Prepare final decision structure
+        # Prepare final decision structure with enhanced model information
         final_decision = {
             "verdict": verdict,
             "confidence": confidence,
@@ -489,7 +521,9 @@ def analyze_content_halves(model_manager, text: str, overall_result: Dict = None
                 "variance_between_halves": round(variance_between_halves, 2),
                 "first_half_model": first_model,
                 "second_half_model": second_model,
-                "models_agree": first_model == second_model
+                "models_agree": models_agree,
+                "ensemble_models_used": models_used,
+                "ensemble_confidence": ensemble_confidence_boost
             }
         }
         
@@ -505,12 +539,21 @@ def analyze_content_halves(model_manager, text: str, overall_result: Dict = None
                 "human_percentage": first_half_result["human_percentage"],
                 "predicted_model": first_model,
                 "word_count": first_half_words,
-                "preview": first_half[:200] + "..." if len(first_half) > 200 else first_half
+                "preview": first_half[:200] + "..." if len(first_half) > 200 else first_half,
+                "top_5_predictions": first_top5,
+                "models_used": models_used
             },
             "second_half": {
                 "ai_percentage": second_ai,
                 "human_percentage": second_half_result["human_percentage"],
                 "predicted_model": second_model,
+                "word_count": second_half_words,
+                "preview": second_half[:200] + "..." if len(second_half) > 200 else second_half,
+                "top_5_predictions": second_top5,
+                "models_used": models_used
+            },
+            "final_decision": final_decision
+        }
                 "word_count": second_half_words,
                 "preview": second_half[:200] + "..." if len(second_half) > 200 else second_half
             },