added logic to train the model for sentiment

app.py

@@ -4,8 +4,16 @@ from transformers import pipeline
 # Load the Hugging Face text classification pipeline
 classifier = pipeline("sentiment-analysis", model="distilbert-base-uncased-finetuned-sst-2-english")
 
-# Custom mapping based on classification score
+# Enhanced classification logic with override for safe routines
 def classify_event(text):
+    # List of keywords that indicate normal activity
+    safe_keywords = ["janitor", "maintenance", "cleaning", "mopping", "scheduled", "authorized personnel"]
+
+    # Check if the input text contains any routine or safe keywords
+    if any(keyword in text.lower() for keyword in safe_keywords):
+        return "Prediction: Normal Activity (manually classified: routine task)"
+
+    # Otherwise, use the classifier
     result = classifier(text)[0]
     label = result['label']
     score = result['score']
@@ -16,7 +24,8 @@ def classify_event(text):
         classification = "Normal Activity"
 
     return f"Prediction: {classification} ({label} - confidence {score:.2f})"
-# Gradio UI
+
+# Gradio Interface
 demo = gr.Interface(
     fn=classify_event,
     inputs=gr.Textbox(lines=4, placeholder="Describe the surveillance event here..."),
@@ -27,6 +36,7 @@ demo = gr.Interface(
         ["A person is standing at the emergency exit for 20 minutes"],
         ["An unknown bag left near the main lobby unattended"],
         ["Two staff members chatting during break in cafeteria"],
+        ["A janitor is cleaning the hallway with a mop and cart during scheduled maintenance hours"],
         ["A car drove into the loading dock after hours without a badge"]
     ]
 )