Update model.py

model.py

@@ -26,6 +26,11 @@ emotion_pipeline = pipeline(
     model="j-hartmann/emotion-english-distilroberta-base",
     device=device
 )
+topic_pipeline = pipeline(
+    "zero-shot-classification",
+    model="facebook/bart-large-mnli",
+    device=device
+)
 
 image_pipeline = pipeline(
     "image-classification",
@@ -62,6 +67,7 @@ image_pipeline.model.eval()
 audio_pipeline.model.eval()
 caption_pipeline.model.eval()
 emotion_pipeline.model.eval()
+topic_pipeline.model.eval()
 
 
 def multimodal_analyze(text, image, audio):
@@ -70,6 +76,16 @@ def multimodal_analyze(text, image, audio):
 
     text_label = None
     text_conf = 0
+    topics = [
+        "technology",
+        "business",
+        "education",
+        "politics",
+        "entertainment",
+        "health",
+        "science",
+        "sports"
+    ]
     image_label = None
     image_conf = 0
     audio_label = None
@@ -85,6 +101,9 @@ def multimodal_analyze(text, image, audio):
         try:
             res = text_pipeline(text)[0]
             emotion_res = emotion_pipeline(text)[0]
+            topic_res = topic_pipeline(text, topics)
+            topic_label = topic_res["labels"][0]
+            topic_conf = round(topic_res["scores"][0] * 100, 2)
             emotion_label = emotion_res["label"]
             emotion_conf = round(emotion_res["score"] * 100, 2)
 
@@ -99,11 +118,14 @@ def multimodal_analyze(text, image, audio):
             text_result_display = f"""
 ## 📝 Text Intelligence
 
-Sentiment: **{text_label}**  
+Sentiment: **{text_label}**
 Confidence: **{text_conf}%**
 
-Emotion: **{emotion_label}**  
+Emotion: **{emotion_label}**
 Emotion Confidence: **{emotion_conf}%**
+
+Topic: **{topic_label}**
+Topic Confidence: **{topic_conf}%**
 """
         except Exception as e:
             text_result_display = f"Text error: {str(e)}"
@@ -172,10 +194,10 @@ Confidence: **{audio_conf}%**
     reasoning_lines = []
 
     if text_label:
-        reasoning_lines.append(
-            f"The textual input expresses a {text_label.lower()} sentiment "
-            f"({text_conf}% confidence) with detected emotion '{emotion_label}'."
-        )
+    reasoning_lines.append(
+        f"The text expresses a {text_label.lower()} sentiment with emotion "
+        f"'{emotion_label}'. The topic appears related to {topic_label}."
+    )
     if image_label:
         reasoning_lines.append(
             f"The uploaded image appears to contain '{image_label}' "