classify

app.py

@@ -55,6 +55,13 @@ if TRANSFORMERS_AVAILABLE:
 else:
     print("Transformers not available, skipping model load.")
 
+zsc = None
+if TRANSFORMERS_AVAILABLE:
+    try:
+        zsc = pipeline("zero-shot-classification", model="joeddav/xlm-roberta-large-xnli", device=-1)
+    except Exception:
+        zsc = None
+
 # Load Bangla Stopwords
 stopwords_path = os.path.join(BASE_DIR, "bangla_stopwords.txt")
 bangla_stopwords = set()
@@ -267,6 +274,71 @@ def extract_hashtags(text):
 # -------------------------
 # Figures builder
 # -------------------------
+def detect_emotion(text, sentiment):
+    t = normalize_text(text)
+    tl = t.lower()
+    if zsc and len(t) > 3:
+        labels = ["Joy", "Anger", "Sadness", "Fear", "Surprise", "Disgust", "Confusion", "Neutral"]
+        try:
+            r = zsc(t[:512], candidate_labels=labels, multi_label=False)
+            return str(r.get("labels", ["Neutral"])[0])
+        except Exception:
+            pass
+    if _RE_LAUGH.search(tl) or re.search(r"[😂🤣😊😄😁😃🙂]", t):
+        return "Joy"
+    if _RE_ANGER.search(t) or re.search(r"[💢😠😡👿]", t) or re.search(r"\b(রাগ|খারাপ|বিরক্ত|ঘৃণা|নিন্দা)\b", tl):
+        return "Anger"
+    if re.search(r"[😢😭😞☹️😔]", t) or re.search(r"\b(দুঃখ|কষ্ট|খারাপ লাগছে|মন খারাপ)\b", tl):
+        return "Sadness"
+    if "?" in t and re.search(r"\b(কেন|কি|কী|কিভাবে|কোথায়|কবে)\b", tl):
+        return "Confusion"
+    s = str(sentiment or "").lower()
+    if "pos" in s:
+        return "Joy"
+    if "neg" in s or s == "error":
+        return "Sadness"
+    return "Neutral"
+
+def detect_tone(text):
+    t = normalize_text(text)
+    tl = t.lower()
+    if zsc and len(t) > 3:
+        labels = ["Formal", "Casual", "Critical", "Inquisitive", "Neutral", "Sarcastic"]
+        try:
+            r = zsc(t[:512], candidate_labels=labels, multi_label=False)
+            return str(r.get("labels", ["Neutral"])[0])
+        except Exception:
+            pass
+    if _RE_LAUGH.search(tl) or re.search(r"[😂🤣😜😝]", t):
+        return "Casual"
+    if re.search(r"\b(দয়া করে|অনুগ্রহ|স্যার|ম্যাডাম|জনাব|জনাবা)\b", tl):
+        return "Formal"
+    if _RE_ANGER.search(t) or re.search(r"\b(রাগ|নিন্দা|খারাপ|বিরক্ত)\b", tl):
+        return "Critical"
+    if "?" in t:
+        return "Inquisitive"
+    return "Neutral"
+
+def detect_intent(text, sentiment):
+    t = normalize_text(text)
+    tl = t.lower()
+    if zsc and len(t) > 3:
+        labels = ["Question", "Request", "Praise", "Complaint", "Statement", "Suggestion"]
+        try:
+            r = zsc(t[:512], candidate_labels=labels, multi_label=False)
+            return str(r.get("labels", ["Statement"])[0])
+        except Exception:
+            pass
+    if "?" in t or re.search(r"\b(কেন|কি|কী|কিভাবে|কোথায়|কবে)\b", tl):
+        return "Question"
+    if re.search(r"\b(দয়া করে|অনুগ্রহ|please|করুন|শেয়ার|লাইক|subscribe)\b", tl):
+        return "Request"
+    if re.search(r"\b(ধন্যবাদ|শুভেচ্ছা|অভিনন্দন|ভাল|সুন্দর|great|awesome)\b", tl):
+        return "Praise"
+    if re.search(r"\b(সমস্যা|অভিযোগ|খারাপ|রাগ|বিরক্ত)\b", tl) or str(sentiment or "").lower().startswith("neg"):
+        return "Complaint"
+    return "Statement"
+
 def make_figures_from_post(post_record, comments_records):
     """
     Build figures using only raw scraped fields.
@@ -640,9 +712,6 @@ def run_live_analysis(n_clicks, url):
         return html.Div("Missing APIFY_TOKEN in environment.", style={"color": "red"}), "", no_update
     
     status_msg = ""
-    if not nlp:
-        status_msg = " (Sentiment Model not loaded - skipping analysis)"
-
     try:
         post_data = scrape_facebook_post(url, token)
         comments_data = scrape_facebook_comments(url, token)
@@ -666,9 +735,9 @@ def run_live_analysis(n_clicks, url):
                     "likes_count": c.get("likesCount", 0),
                     "date_created": c.get("date", ""),
                     "sentiment": label,
-                    "emotion": "Unknown",
-                    "tone": "Unknown",
-                    "intent": "Unknown"
+                    "emotion": detect_emotion(text, label),
+                    "tone": detect_tone(text),
+                    "intent": detect_intent(text, label)
                 })
         
         final_data = {