Update app.py

app.py

@@ -6,7 +6,7 @@ import pytesseract
 import pandas as pd
 import plotly.express as px
 
-# Step 1: Emoji 翻译模型（你自己训练的模型）
+# ✅ Step 1: Emoji 翻译模型（你自己训练的模型）
 emoji_model_id = "JenniferHJF/qwen1.5-emoji-finetuned"
 emoji_tokenizer = AutoTokenizer.from_pretrained(emoji_model_id, trust_remote_code=True)
 emoji_model = AutoModelForCausalLM.from_pretrained(
@@ -16,21 +16,29 @@ emoji_model = AutoModelForCausalLM.from_pretrained(
 ).to("cuda" if torch.cuda.is_available() else "cpu")
 emoji_model.eval()
 
-# Step 2: 可选择的冒犯性文本识别模型
+# ✅ Step 2: 可选择的冒犯性文本识别模型
 model_options = {
     "Toxic-BERT": "unitary/toxic-bert",
     "Roberta Offensive": "cardiffnlp/twitter-roberta-base-offensive",
     "BERT Emotion": "bhadresh-savani/bert-base-go-emotion"
 }
 
-# 页面配置
+# ✅ 页面配置
 st.set_page_config(page_title="Emoji Offensive Text Detector", page_icon="🚨", layout="wide")
 
-# 初始化历史记录
+# ✅ 侧边栏：模型选择
+with st.sidebar:
+    st.header("🧠 Settings")
+    selected_model = st.selectbox("Choose classification model", list(model_options.keys()))
+    selected_model_id = model_options[selected_model]
+    classifier = pipeline("text-classification", model=selected_model_id,
+                          device=0 if torch.cuda.is_available() else -1)
+
+# 初始化会话历史
 if "history" not in st.session_state:
     st.session_state.history = []
 
-# Emoji 文本翻译与分类函数
+
 def classify_emoji_text(text: str):
     prompt = f"输入：{text}\n输出："
     input_ids = emoji_tokenizer(prompt, return_tensors="pt").to(emoji_model.device)
@@ -42,82 +50,85 @@ def classify_emoji_text(text: str):
     result = classifier(translated_text)[0]
     label = result["label"]
     score = result["score"]
-    reasoning = f"The sentence was flagged as '{label}' due to potentially offensive phrases. Consider replacing emotionally charged, ambiguous, or abusive terms."
+    reasoning = (
+        f"The sentence was flagged as '{label}' due to potentially offensive phrases. "
+        "Consider replacing emotionally charged, ambiguous, or abusive terms."
+    )
 
-    st.session_state.history.append({"text": text, "translated": translated_text, "label": label, "score": score, "reason": reasoning})
+    st.session_state.history.append({
+        "text": text,
+        "translated": translated_text,
+        "label": label,
+        "score": score,
+        "reason": reasoning
+    })
     return translated_text, label, score, reasoning
 
-# 页面布局
-st.sidebar.header("🧠 Settings")
-selected_model = st.sidebar.selectbox("Choose classification model", list(model_options.keys()))
-selected_model_id = model_options[selected_model]
-classifier = pipeline("text-classification", model=selected_model_id, device=0 if torch.cuda.is_available() else -1)
-
-# 主页面：集成 Text Moderation 和 Text Analysis
-st.title("🚨 Emoji Offensive Text Detector & Violation Analysis")
-
-# 输入与分类
-st.markdown("## ✍️ 输入或上传文本进行分类")
-col1, col2 = st.columns([2,1])
-with col1:
-    text = st.text_area("Enter sentence with emojis:", value="你是🐷", height=150)
-    if st.button("🚦 Analyze Text"):
-        with st.spinner("🔍 Processing..."):
-            try:
-                translated, label, score, reason = classify_emoji_text(text)
-                st.markdown("### 🔄 Translated sentence:")
-                st.code(translated, language="text")
+# 主页面布局
+st.title("🚨 Emoji Offensive Text Detector & Analysis")
+st.markdown("---")
 
-                st.markdown(f"### 🎯 Prediction: {label}")
-                st.markdown(f"### 📊 Confidence Score: {score:.2%}")
-                st.markdown("### 🧠 Model Explanation:")
-                st.info(reason)
-            except Exception as e:
-                st.error(f"❌ Error during processing: {e}")
+# 输入与分析
+st.header("✍️ Input & Moderation")
+def text_moderation_section():
+    st.markdown("Enter text with emojis or upload an image with text.")
+    text = st.text_area("Sentence (or OCR text will appear here):", height=120)
 
-with col2:
-    st.markdown("### 🖼️ Or upload a screenshot:")
-    uploaded_file = st.file_uploader("Image (JPG/PNG)", type=["jpg","png","jpeg"])
+    uploaded_file = st.file_uploader("Or upload an image for OCR:", type=["jpg", "jpeg", "png"])
     if uploaded_file:
         image = Image.open(uploaded_file)
         st.image(image, caption="Uploaded Image", use_column_width=True)
-        with st.spinner("🧠 Running OCR..."):
+        with st.spinner("Extracting text via OCR..."):
             ocr_text = pytesseract.image_to_string(image, lang="chi_sim+eng").strip()
-            st.markdown("#### 📋 OCR Extracted Text:")
-            st.code(ocr_text)
-            translated, label, score, reason = classify_emoji_text(ocr_text)
-            st.markdown("#### 🔄 Translated:")
-            st.code(translated)
-            st.markdown(f"#### 🎯 Prediction: {label}")
-            st.markdown(f"#### 📊 Confidence: {score:.2%}")
-            st.markdown("#### 🧠 Explanation:")
-            st.info(reason)
+        st.text_area("Extracted Text:", value=ocr_text, height=120)
+        text = ocr_text
+
+    if st.button("🚦 Analyze Text") and text:
+        with st.spinner("Processing..."):
+            try:
+                translated, label, score, reason = classify_emoji_text(text)
+                st.subheader("🔄 Translated Text")
+                st.code(translated)
+                st.subheader(f"🎯 Prediction: {label}")
+                st.write(f"Confidence: {score:.2%}")
+                st.subheader("🧠 Explanation")
+                st.info(reason)
+            except Exception as e:
+                st.error(f"Error during processing: {e}")
 
+# 分析仪表板
 st.markdown("---")
+st.header("📊 Violation Analysis")
+def analysis_dashboard():
+    if not st.session_state.history:
+        st.info("No data to display. Please analyze some text first.")
+        return
 
-# 违规分析仪表盘
-st.markdown("## 📊 Violation Analysis Dashboard")
-if st.session_state.history:
     df = pd.DataFrame(st.session_state.history)
-    st.markdown("### 🧾 历史记录详情")
+
+    # 建议列表
+    st.subheader("📝 Offensive Terms & Suggestions")
     for item in st.session_state.history:
-        st.markdown(f"- 🔹 **input:** {item['text']} | **Label:** {item['label']} | **Confidence:** {item['score']:.2%}")
-        st.markdown(f"  - **Translated:** {item['translated']}")
-        st.markdown(f"  - **Suggestion:** {item['reason']}")
+        st.markdown(f"- **Input:** {item['text']}")
+        st.markdown(f"  - Translated: {item['translated']}")
+        st.markdown(f"  - Label: {item['label']} ({item['score']:.2%})")
+        st.markdown(f"  - Suggestion: {item['reason']}")
 
+    # 雷达图
     radar_df = pd.DataFrame({
-        "Category": ["Insult","Abuse","Discrimination","Hate Speech","Vulgarity"],
-        "Score": [0.7,0.4,0.3,0.5,0.6]
+        "Category": ["Insult", "Abuse", "Discrimination", "Hate Speech", "Vulgarity"],
+        "Score": [0.7, 0.4, 0.3, 0.5, 0.6]
     })
-    # 优化雷达图，设置线条为黑色
     radar_fig = px.line_polar(
         radar_df,
         r='Score',
         theta='Category',
         line_close=True,
-        title="⚠️ Risk Radar by Category",
-        color_discrete_sequence=['black']
+        title="⚠️ Risk Radar by Category"
     )
+    radar_fig.update_traces(line_color='black')
     st.plotly_chart(radar_fig)
-else:
-    st.info("⚠️ No data available. Please analyze some text first.")
+
+# 渲染各部分
+text_moderation_section()
+analysis_dashboard()