Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
@@ -19,31 +19,55 @@ model_options = {
|
|
19 |
"BERT Emotion": "bhadresh-savani/bert-base-go-emotion"
|
20 |
}
|
21 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
22 |
# Streamlit 侧边栏模型选择
|
23 |
selected_model = st.sidebar.selectbox("Choose classification model", list(model_options.keys()))
|
24 |
selected_model_id = model_options[selected_model]
|
25 |
-
|
26 |
classifier = pipeline("text-classification", model=selected_model_id, device=0 if torch.cuda.is_available() else -1)
|
27 |
|
|
|
|
|
|
|
|
|
|
|
|
|
28 |
def classify_emoji_text(text: str):
|
29 |
-
"""
|
30 |
-
Step 1: 翻译文本中的 emoji
|
31 |
-
Step 2: 使用分类器判断是否冒犯
|
32 |
-
"""
|
33 |
prompt = f"输入:{text}\n输出:"
|
34 |
input_ids = emoji_tokenizer(prompt, return_tensors="pt").to(emoji_model.device)
|
35 |
with torch.no_grad():
|
36 |
output_ids = emoji_model.generate(**input_ids, max_new_tokens=64, do_sample=False)
|
37 |
decoded = emoji_tokenizer.decode(output_ids[0], skip_special_tokens=True)
|
38 |
-
|
39 |
-
# 保留真正输出部分(移除 prompt)
|
40 |
-
if "输出:" in decoded:
|
41 |
-
translated_text = decoded.split("输出:")[-1].strip()
|
42 |
-
else:
|
43 |
-
translated_text = decoded.strip()
|
44 |
|
45 |
result = classifier(translated_text)[0]
|
46 |
label = result["label"]
|
47 |
score = result["score"]
|
48 |
|
49 |
return translated_text, label, score
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
19 |
"BERT Emotion": "bhadresh-savani/bert-base-go-emotion"
|
20 |
}
|
21 |
|
22 |
+
# ✅ 页面配置
|
23 |
+
st.set_page_config(page_title="Emoji Offensive Text Detector", page_icon="🚨", layout="wide")
|
24 |
+
|
25 |
+
# ✅ 页面标题
|
26 |
+
st.title("🧠 Emoji-based Offensive Language Classifier")
|
27 |
+
|
28 |
+
st.markdown("""
|
29 |
+
This application translates emojis in a sentence and classifies whether the final sentence is offensive or not using two AI models.
|
30 |
+
- The **first model** translates emoji or symbolic phrases into standard Chinese text.
|
31 |
+
- The **second model** performs offensive language detection.
|
32 |
+
""")
|
33 |
+
|
34 |
# Streamlit 侧边栏模型选择
|
35 |
selected_model = st.sidebar.selectbox("Choose classification model", list(model_options.keys()))
|
36 |
selected_model_id = model_options[selected_model]
|
|
|
37 |
classifier = pipeline("text-classification", model=selected_model_id, device=0 if torch.cuda.is_available() else -1)
|
38 |
|
39 |
+
# ✅ 输入区域
|
40 |
+
st.markdown("### ✍️ Input your sentence:")
|
41 |
+
default_text = "你是🐷"
|
42 |
+
text = st.text_area("Enter sentence with emojis:", value=default_text, height=150)
|
43 |
+
|
44 |
+
# ✅ 主逻辑封装函数
|
45 |
def classify_emoji_text(text: str):
|
|
|
|
|
|
|
|
|
46 |
prompt = f"输入:{text}\n输出:"
|
47 |
input_ids = emoji_tokenizer(prompt, return_tensors="pt").to(emoji_model.device)
|
48 |
with torch.no_grad():
|
49 |
output_ids = emoji_model.generate(**input_ids, max_new_tokens=64, do_sample=False)
|
50 |
decoded = emoji_tokenizer.decode(output_ids[0], skip_special_tokens=True)
|
51 |
+
translated_text = decoded.split("输出:")[-1].strip() if "输出:" in decoded else decoded.strip()
|
|
|
|
|
|
|
|
|
|
|
52 |
|
53 |
result = classifier(translated_text)[0]
|
54 |
label = result["label"]
|
55 |
score = result["score"]
|
56 |
|
57 |
return translated_text, label, score
|
58 |
+
|
59 |
+
# ✅ 触发按钮
|
60 |
+
if st.button("🚦 Analyze"):
|
61 |
+
with st.spinner("🔍 Processing..."):
|
62 |
+
try:
|
63 |
+
translated, label, score = classify_emoji_text(text)
|
64 |
+
st.markdown("### 🔄 Translated sentence:")
|
65 |
+
st.code(translated, language="text")
|
66 |
+
|
67 |
+
st.markdown(f"### 🎯 Prediction: `{label}`")
|
68 |
+
st.markdown(f"### 📊 Confidence Score: `{score:.2%}`")
|
69 |
+
|
70 |
+
except Exception as e:
|
71 |
+
st.error(f"❌ An error occurred during processing:\n\n{e}")
|
72 |
+
else:
|
73 |
+
st.info("👈 Please input text and click the button to classify.")
|