Update app.py

app.py

@@ -107,14 +107,6 @@ def sliding_window(kalimat, window=4):
         for i in range(len(kalimat) - window + 1)
     ]
 
-def event_completeness(text):
-    text = text.lower()
-    return sum(any(w in text for w in words) for words in EVENT_UNITS.values())
-
-def is_metaphor(text):
-    text = text.lower()
-    return any(a in text and b in text for a, b in METAPHOR_PAIRS)
-
 def rating_usia(kategori):
     if "sadisme" in kategori:
         return 21
@@ -125,7 +117,7 @@ def rating_usia(kategori):
     return 0
 
 # ======================
-# CORE ANALYSIS + DEBUG (JANGAN DIUBAH)
+# CORE ANALYSIS + DEBUG (ASLI — JANGAN DIUBAH)
 # ======================
 def analyze_text(judul, isi):
     kalimat = split_kalimat(isi)
@@ -166,21 +158,28 @@ def analyze_text(judul, isi):
     return usia, ", ".join(sorted(detected))
 
 # ======================
-# FILTER ABUSIVE (TERPISAH, AMAN)
+# FILTER ABUSIVE (TERPISAH + LOG LENGKAP)
 # ======================
 def filter_abusive(isi):
+    log("\n" + "=" * 80)
+    log("[ABUSIVE] START")
+    log(f"[ABUSIVE] INPUT:\n{isi}")
+
     paragraphs = re.split(r'\n+', isi)
     output = []
 
-    for para in paragraphs:
+    for p_idx, para in enumerate(paragraphs):
         para = para.strip()
         if not para:
             continue
 
+        log(f"\n[ABUSIVE] PARAGRAPH {p_idx}: {para}")
         sentences = re.split(r'(?<=[.!?])\s+', para)
         notes = []
 
-        for s in sentences:
+        for s_idx, s in enumerate(sentences):
+            log(f"[ABUSIVE]   SENTENCE {s_idx}: {s}")
+
             inputs = abusive_tokenizer(
                 s,
                 return_tensors="pt",
@@ -195,20 +194,30 @@ def filter_abusive(isi):
                 probs = F.softmax(out.logits, dim=-1)
                 pred = torch.argmax(probs, dim=-1).item()
 
-            if id2label[pred] == "Abusif":
-                notes.append(
-                    f'Kalimat "{s}" mengandung kalimat abusif, tidak baik diucapkan'
-                )
+            label = id2label[pred]
+            log(f"[ABUSIVE]     PRED={label}")
+
+            if label == "Abusif":
+                note = f'Kalimat "{s}" mengandung kalimat abusif, tidak baik diucapkan'
+                notes.append(note)
+                log(f"[ABUSIVE]     ⚠️ {note}")
 
         output.append(para)
         output.extend(notes)
 
-    return "\n".join(output)
+    result = "\n".join(output)
+    log("\n[ABUSIVE] OUTPUT:")
+    log(result)
+    log("=" * 80)
+
+    return result
 
 # ======================
-# ROUTER /ANALYZE (MODE-BASED)
+# ROUTER /ANALYZE (MODE BASED — TETAP)
 # ======================
 def analyze_router(judul, isi, mode):
+    log(f"\n[ROUTER] MODE = {mode}")
+
     if mode == "rating":
         usia, kategori = analyze_text(judul, isi)
         return {
@@ -217,8 +226,9 @@ def analyze_router(judul, isi, mode):
         }
 
     if mode == "abusive":
+        filtered = filter_abusive(isi)
         return {
-            "filtered_text": filter_abusive(isi)
+            "filtered_text": filtered
         }
 
     return {
@@ -226,7 +236,7 @@ def analyze_router(judul, isi, mode):
     }
 
 # ======================
-# GRADIO API (ROUTE TETAP /analyze)
+# GRADIO API — ROUTE TETAP /analyze
 # ======================
 demo = gr.Interface(
     fn=analyze_router,
@@ -248,4 +258,4 @@ if __name__ == "__main__":
         server_port=7860,
         ssr_mode=False,
         show_error=True
-    )
+    )