허깅페이스 실습 및 스페이스에 프로젝트 등록

HUGGINGFACE/sentiment.py

@@ -0,0 +1,53 @@
+import gradio as gr
+from transformers import pipeline
+import re
+
+english_classifier = pipeline("sentiment-analysis", top_k = None)
+korean_classifier = pipeline("sentiment-analysis", model="WhitePeak/bert-base-cased-Korean-sentiment", top_k = None)
+
+def is_korean(text):
+    korean = re.search(r"[가-힣]", text)
+    return korean is not None
+    
+
+def predict_sentiment(text):
+    
+    # 한국말인지 확인하기
+    if is_korean(text):
+        language = "한국어 모델"
+        results = korean_classifier(text)[0]
+    else:
+        language = "영어 모델"
+        results = english_classifier(text)[0]
+        
+    # {'label' : 'POSITIVE', 'score' : 0.9192341028490124}
+    # {'label' : 'LABEL_1', 'score' : 0.9192341028490124}
+    
+    # label = results[0]["label"]
+    label_map = {"LABEL_0" : "부정 😡", "LABEL_1" : "긍정 😄", "NEGATIVE" : "부정 😡", "POSITIVE" : "긍정 😄"}
+    
+    # label = label_map.get(label, label)
+    # score1 = result[0]["score"]
+    # score2 = result[1]["score"]
+    
+    # return f"사용모델 : {language}\n 감정 : {label}\n 확률 : ({score1:.4f})\n OTHER : ({score2:.4f})"
+    
+    scores = {}
+    
+    for item in results:
+        label = item["label"]
+        scores[label_map.get(label, label)] = item["score"]
+        
+    return scores
+    
+    
+
+demo = gr.Interface(
+    fn = predict_sentiment,
+    inputs=[gr.Text(lines=3, placeholder="문장을 입력하세요")],
+    outputs=[gr.Label(num_top_classes=2)],
+    title = "AI 감정분석 웹",
+    description="HuggingFace Transformer 기반 감정 분석 프로그램",
+)
+
+demo.launch()