Update app.py

app.py

@@ -1,113 +1,31 @@
-import logging
-from functools import lru_cache
-
-from fastapi import FastAPI, HTTPException
-from fastapi.middleware.cors import CORSMiddleware
+import gradio as gr
 from transformers import pipeline
 import os
-import re
-from pydantic import BaseModel, Field
-
-app = FastAPI(title="Sentiment Analyzer API", version="1.0.0")
-
-# CORS middleware ekle
-app.add_middleware(
-    CORSMiddleware,
-    allow_origins=["*"],
-    allow_credentials=True,
-    allow_methods=["*"],
-    allow_headers=["*"],
-)
-
-logger = logging.getLogger("ai-service")
-logging.basicConfig(level=os.getenv("LOG_LEVEL", "INFO"))
-
-# Türkçe metin desteği için daha iyi bir model kullan
-MODEL_NAME = os.getenv("MODEL_NAME", "cardiffnlp/twitter-xlm-roberta-base-sentiment")
-
-
-class AnalyzeRequest(BaseModel):
-    message: str = Field(..., min_length=1, description="Analyzed text")
-
-
-class AnalyzeResponse(BaseModel):
-    sentiment: str
-    score: float
-    original_text: str
-    processed_text: str
-
 
-@lru_cache(maxsize=1)
-def get_analyzer():
-    logger.info("Loading sentiment model %s", MODEL_NAME)
-    return pipeline("sentiment-analysis", model=MODEL_NAME)
-
-def preprocess_text(text):
-    """Metni ön işleme tabi tut - emojileri ve özel karakterleri temizle"""
-    if not text or not isinstance(text, str):
-        return ""
-    
-    # Emojileri temizle
-    emoji_pattern = re.compile("["
-        u"\U0001F600-\U0001F64F"  # emoticons
-        u"\U0001F300-\U0001F5FF"  # symbols & pictographs
-        u"\U0001F680-\U0001F6FF"  # transport & map symbols
-        u"\U0001F1E0-\U0001F1FF"  # flags (iOS)
-        u"\U00002702-\U000027B0"
-        u"\U000024C2-\U0001F251"
-        "]+", flags=re.UNICODE)
-    
-    text = emoji_pattern.sub(r'', text)
-    
-    # Fazla boşlukları temizle
-    text = re.sub(r'\s+', ' ', text).strip()
-    
-    return text
-
-@app.get("/health")
-def health():
-    return {"ok": True}
-
-@app.post("/analyze", response_model=AnalyzeResponse)
-async def analyze(payload: AnalyzeRequest):
-    text = payload.message
-
-    # Metni ön işleme tabi tut
-    processed_text = preprocess_text(text)
-
-    if not processed_text:
-        return AnalyzeResponse(
-            sentiment="neutral",
-            score=0.5,
-            original_text=text,
-            processed_text=processed_text,
-        )
-
-    try:
-        analyzer = get_analyzer()
-        result = analyzer(processed_text)[0]
-    except Exception as exc:
-        logger.exception("Sentiment analysis failed")
-        raise HTTPException(status_code=500, detail="analysis failed") from exc
+MODEL_NAME = os.getenv("MODEL_NAME", "distilbert-base-uncased-finetuned-sst-2-english")
+analyzer = pipeline("sentiment-analysis", model=MODEL_NAME)
 
+def analyze(message: str):
+    if not isinstance(message, str) or not message.strip():
+        return {"sentiment": "neutral", "score": 0.0}
+    result = analyzer(message)[0]  # {'label': 'POSITIVE'|'NEGATIVE', 'score': 0.xx}
     label = result["label"].lower()
     score = float(result.get("score", 0.0))
-
-    # Model çıktılarını standartlaştır
-    if "positive" in label or "joy" in label:
-        sentiment = "positive"
-    elif "negative" in label or "sadness" in label or "anger" in label:
-        sentiment = "negative"
-    else:
+    if label not in ("positive", "negative"):
         sentiment = "neutral"
-
-    # Nötr bant genişletilmiş (0.4-0.6 arası)
-    if 0.4 <= score <= 0.6:
+    elif 0.48 <= score <= 0.52:
         sentiment = "neutral"
+    else:
+        sentiment = label
+    return {"sentiment": sentiment, "score": score}
+
+demo = gr.Interface(
+    fn=analyze,
+    inputs=gr.Textbox(label="Mesaj"),
+    outputs=gr.JSON(label="Sonuç"),
+    title="Duygu Analizi",
+    allow_flagging="never"
+)
 
-    return AnalyzeResponse(
-        sentiment=sentiment,
-        score=score,
-        original_text=text,
-        processed_text=processed_text,
-    )
+if __name__ == "__main__":
+    demo.launch()