Mingu Kim
53 lines
1.6 KiB
Python
53 lines
1.6 KiB
Python
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() |