Update app.py

app.py

@@ -124,19 +124,32 @@ class SpeechAnalyzer:
         
         return aggregated_scores
 
-    def analyze_emotional_trajectory(self, text, window_size=5):
-        """Perform emotional trajectory analysis"""
-        segments = self.split_text(text, max_length=256)
+    def analyze_emotional_trajectory(self, text, window_size=5, ngram_size=3):
+        """Enhanced emotional trajectory analysis using n-grams and relative scoring"""
+        # Create overlapping n-grams
+        words = text.split()
+        ngrams = [' '.join(words[i:i+ngram_size]) for i in range(0, len(words)-ngram_size+1)]
         
+        # Get raw sentiment scores
         sentiment_scores = []
-        for segment in segments:
-            result = self.sentiment_pipeline(segment)[0]
-            score = 1 if result['label'] == 'POSITIVE' else -1
-            sentiment_scores.append(score)
+        for ngram in ngrams:
+            result = self.sentiment_pipeline(ngram)[0]
+            # Use confidence score directly without binary transformation
+            raw_score = result['score']
+            sentiment_scores.append(raw_score)
+        
+        # Normalize scores to show relative variations
+        if sentiment_scores:
+            min_score = min(sentiment_scores)
+            max_score = max(sentiment_scores)
+            score_range = max_score - min_score
+            if score_range > 0:
+                sentiment_scores = [(s - min_score) / score_range * 2 - 1 for s in sentiment_scores]
         
+        # Apply smoothing for cleaner visualization
         smoothed_scores = (savgol_filter(sentiment_scores, window_length=window_size, polyorder=2) 
-                           if len(sentiment_scores) > window_size else sentiment_scores)
-
+                          if len(sentiment_scores) > window_size else sentiment_scores)
+        
         return smoothed_scores
 
     def detect_named_entities(self, text):
@@ -287,28 +300,19 @@ def main():
                 
         with tab2:
             st.subheader("Speech Trajectory Analysis")
-            
             col1, col2 = st.columns(2)
             
             with col1:
                 st.write("### Emotional Flow")
-                # Process text in manageable chunks
-                segments = analyzer.split_text(text, max_length=512)
-                sentiment_results = []
-                
-                for segment in segments:
-                    result = analyzer.sentiment_pipeline(segment[:512])[0]
-                    # Convert score to a more granular value instead of binary
-                    sentiment_value = result['score'] if result['label'] == 'POSITIVE' else -result['score']
-                    sentiment_results.append(sentiment_value)
+                # Get enhanced sentiment trajectory with n-grams
+                sentiment_scores = analyzer.analyze_emotional_trajectory(text, window_size=5, ngram_size=3)
                 
-                # Create segment labels
-                num_segments = len(sentiment_results)
-                segment_labels = [f"{i+1}" for i in range(num_segments)]
+                # Create segment labels based on n-gram windows
+                segment_labels = [f"{i+1}" for i in range(len(sentiment_scores))]
                 
                 trajectory_fig = go.Figure(data=go.Scatter(
                     x=segment_labels,
-                    y=sentiment_results,
+                    y=sentiment_scores,
                     mode='lines+markers',
                     line=dict(
                         color='#1f77b4',
@@ -318,15 +322,16 @@ def main():
                         size=8,
                         color='#1f77b4',
                         symbol='circle'
-                    )
+                    ),
+                    hovertemplate='Segment %{x}<br>Score: %{y:.2f}<extra></extra>'
                 ))
                 
                 trajectory_fig.update_layout(
                     title='Emotional Flow Throughout the Speech',
                     xaxis_title='Speech Segments',
-                    yaxis_title='Emotional Tone',
+                    yaxis_title='Relative Emotional Tone',
                     yaxis=dict(
-                        ticktext=['Very Negative', 'Neutral', 'Very Positive'],
+                        ticktext=['Most Negative', 'Neutral', 'Most Positive'],
                         tickvals=[-1, 0, 1],
                         range=[-1.1, 1.1],
                         gridcolor='lightgray'
@@ -345,7 +350,6 @@ def main():
                 moral_trajectories = {foundation: [] for foundation in MORAL_FOUNDATIONS}
                 
                 for segment in segments:
-                    segment = segment[:512]
                     segment_scores = analyzer.analyze_moral_foundations(segment)
                     for foundation, score in segment_scores.items():
                         moral_trajectories[foundation].append(score)