Update app.py

app.py

@@ -16,8 +16,8 @@ os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2'
 from tensorflow.keras.models import load_model
 from tensorflow.keras.preprocessing.sequence import pad_sequences
 
-nltk.download('punkt')
-nltk.download('stopwords')
+nltk.download('punkt', quiet=True)
+nltk.download('stopwords', quiet=True)
 
 app = FastAPI()
 
@@ -75,11 +75,12 @@ def preprocessing(statement):
     tokens = [token for token in tokens if token.isalpha() and len(token) > 2]
     return ' '.join(tokens)
 
-def analyze_text(text: str, mode: str, model_type: str):
+# FUNGSI BARU: Menganalisis banyak teks sekaligus (Batch Processing)
+def analyze_bulk(texts: list, mode: str, model_type: str):
     if mode not in ["fake_review", "spam"]: return {"error": "Mode invalid."}
-    
-    clean_text = preprocessing(text)
-    score = 0.0
+    if not texts: return []
+
+    clean_texts = [preprocessing(t) for t in texts]
     
     if model_type == "lstm":
         tokenizer = models_dict[mode]["tokenizer"]
@@ -88,10 +89,11 @@ def analyze_text(text: str, mode: str, model_type: str):
         if not tokenizer or not lstm_model:
             return {"error": f"File model LSTM untuk {mode} tidak ditemukan."}
             
-        seq = tokenizer.texts_to_sequences([clean_text])
-        padded = pad_sequences(seq, maxlen=100, padding='post', truncating='post')
-        prediction = lstm_model.predict(padded, verbose=0)[0][0]
-        score = float(prediction) * 100
+        seqs = tokenizer.texts_to_sequences(clean_texts)
+        padded = pad_sequences(seqs, maxlen=100, padding='post', truncating='post')
+        # Prediksi sekaligus dalam satu tarikan
+        predictions = lstm_model.predict(padded, batch_size=32, verbose=0).flatten()
+        scores = [float(p) * 100 for p in predictions]
         
     elif model_type in ["lr", "rf", "xgb"]:
         tfidf = models_dict[mode]["tfidf"]
@@ -100,30 +102,34 @@ def analyze_text(text: str, mode: str, model_type: str):
         if not tfidf or not clf_model:
             return {"error": f"File model {model_type} untuk {mode} tidak ditemukan."}
             
-        vec_text = tfidf.transform([clean_text])
-        proba = clf_model.predict_proba(vec_text)[0]
-        score = float(proba[1]) * 100
-    
+        vec_texts = tfidf.transform(clean_texts)
+        # Prediksi sekaligus dalam satu tarikan
+        probas = clf_model.predict_proba(vec_texts)
+        scores = [float(p[1]) * 100 for p in probas]
     else:
         return {"error": "Model type tidak dikenali."}
 
-    is_flagged = score > 50
-    confidence = score if is_flagged else 100 - score
-    
-    if mode == "spam":
-        label = "Spam" if is_flagged else "Safe (Ham)"
-        reason = "The text contains word order patterns similar to those found in promotional messages or scams." if is_flagged else "The sentence appears similar to ordinary conversational message."
-    else:
-        label = "Fake" if is_flagged else "Genuine"
-        reason = "The writing style bears a strong resemblance to paid review templates or no clear context." if is_flagged else "The review has a natural variety of vocabulary and recounts the experience in detail."
+    results = []
+    for text, score in zip(texts, scores):
+        is_flagged = score > 50
+        confidence = score if is_flagged else 100 - score
         
-    return {
-        "text": text,
-        "label": label,
-        "confidence": confidence,
-        "is_flagged": is_flagged,
-        "reason": reason
-    }
+        if mode == "spam":
+            label = "Spam" if is_flagged else "Safe (Ham)"
+            reason = "The text contains word order patterns similar to those found in promotional messages or scams." if is_flagged else "The sentence appears similar to ordinary conversational message."
+        else:
+            label = "Fake" if is_flagged else "Genuine"
+            reason = "The writing style bears a strong resemblance to paid review templates or no clear context." if is_flagged else "The review has a natural variety of vocabulary and recounts the experience in detail."
+            
+        results.append({
+            "text": text,
+            "label": label,
+            "confidence": confidence,
+            "is_flagged": is_flagged,
+            "reason": reason
+        })
+        
+    return results
 
 class TextRequest(BaseModel):
     text: str
@@ -136,10 +142,11 @@ class TextRequest(BaseModel):
 
 @app.post("/api/analyze/text")
 def analyze_single(req: TextRequest):
-    result = analyze_text(req.text, req.mode, req.model_type)
-    if "error" in result:
+    # Memanfaatkan fungsi bulk untuk 1 teks
+    result = analyze_bulk([req.text], req.mode, req.model_type)
+    if isinstance(result, dict) and "error" in result:
         raise HTTPException(status_code=500, detail=result["error"])
-    return {"status": "success", "result": result}
+    return {"status": "success", "result": result[0]}
 
 @app.post("/api/analyze/file")
 async def analyze_file(mode: str = Form(...), model_type: str = Form(...), file: UploadFile = File(...)):
@@ -148,20 +155,19 @@ async def analyze_file(mode: str = Form(...), model_type: str = Form(...), file:
     
     content = await file.read()
     decoded = content.decode('utf-8')
-    results = []
+    texts = []
     
     if file.filename.endswith('.txt'):
-        lines = [line.strip() for line in decoded.split('\n') if line.strip()]
-        for line in lines:
-            res = analyze_text(line, mode, model_type)
-            if "error" not in res: results.append(res)
+        texts = [line.strip() for line in decoded.split('\n') if line.strip()]
             
     elif file.filename.endswith('.csv'):
         df = pd.read_csv(StringIO(decoded))
         text_col = next((col for col in ['text', 'review', 'v2', 'text_'] if col in df.columns), df.columns[0])
+        texts = df[text_col].dropna().astype(str).tolist()
             
-        for text in df[text_col].dropna().astype(str):
-            res = analyze_text(text, mode, model_type)
-            if "error" not in res: results.append(res)
-            
+    # Seluruh teks dikirim langsung ke fungsi bulk
+    results = analyze_bulk(texts, mode, model_type)
+    if isinstance(results, dict) and "error" in results:
+        raise HTTPException(status_code=500, detail=results["error"])
+        
     return {"status": "success", "total": len(results), "results": results}