Update app.py

app.py

@@ -1,149 +1,197 @@
-from transformers import pipeline, AutoTokenizer, AutoModelForCausalLM
-import torch
 import streamlit as st
-from PIL import Image
-import pytesseract
-import openai
+from transformers import pipeline, AutoTokenizer, AutoModelForCausalLM
 import pandas as pd
-import plotly.express as px
-
-# ✅ 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(
-    emoji_model_id,
-    trust_remote_code=True,
-    torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32
-).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
-model_options = {
-    "Toxic-BERT": "unitary/toxic-bert",
-    "Roberta Offensive": "cardiffnlp/twitter-roberta-base-offensive",
-    "BERT Emotion": "bhadresh-savani/bert-base-go-emotion"
+import torch
+import matplotlib.pyplot as plt
+import numpy as np
+import re
+
+# 页面配置
+st.set_page_config(page_title="🛡️ 智盾内容安全审核平台", layout="wide")
+PAGES = {
+    "🏠 首页": "home",
+    "🏢 金融行业审核": "finance",
+    "🏛 政府行业审核": "government",
+    "🌐 互联网行业审核": "internet",
+    "🧠 产品能力": "capability",
+    "✍️ 文本校正": "text_correction",
+    "📄 文档效正": "doc_alignment",
+    "🎙️ 语音检测": "speech_check",
+    "💼 加入我们": "join_us",
+    "💬 客户反馈": "feedback"
 }
-
-# ✅ 页面配置
-st.set_page_config(page_title="Emoji Offensive Text Detector", page_icon="🚨", layout="wide")
-
-# ✅ 页面布局
-with st.sidebar:
-    st.header("🧠 Navigation")
-    section = st.radio("Select Mode:", ["📍 Text Moderation", "📊 Text Analysis", "🛠️ Agent Build"])
-
-    if section == "📍 Text Moderation":
-        moderation_type = st.selectbox("Select Task Type", ["Normal Text", "Bullet Screen Text"])
-        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)
-
-    elif section == "📊 Text Analysis":
-        st.markdown("You can view the violation distribution chart and editing suggestions.")
-
-    elif section == "🛠️ Agent Build":
-        st.markdown("Upload supporting files for in-context fine-tuning.")
-        uploaded_reference = st.file_uploader("Upload fine-tuning reference file", type=["txt", "csv"])
-
-if "history" not in st.session_state:
-    st.session_state.history = []
-
-
-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)
-    with torch.no_grad():
-        output_ids = emoji_model.generate(**input_ids, max_new_tokens=64, do_sample=False)
-    decoded = emoji_tokenizer.decode(output_ids[0], skip_special_tokens=True)
-    translated_text = decoded.split("输出：")[-1].strip() if "输出：" in decoded else decoded.strip()
-
-    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."
-
-    st.session_state.history.append({"text": text, "translated": translated_text, "label": label, "score": score, "reason": reasoning})
-    return translated_text, label, score, reasoning
-
-# ✅ Section logic
-if section == "📍 Text Moderation":
-    st.title("📍 Offensive Text Classification")
-    st.markdown("### ✍️ Input your sentence:")
-    default_text = "你是🐷"
-    text = st.text_area("Enter sentence with emojis:", value=default_text, height=150)
-
-    if st.button("🚦 Analyze"):
-        with st.spinner("🔍 Processing..."):
-            try:
-                translated, label, score, reason = classify_emoji_text(text)
-                st.markdown("### 🔄 Translated sentence:")
-                st.code(translated, language="text")
-
-                st.markdown(f"### 🎯 Prediction: `{label}`")
-                st.markdown(f"### 📊 Confidence Score: `{score:.2%}`")
-                st.markdown(f"### 🧠 Model Explanation:")
-                st.info(reason)
-
-            except Exception as e:
-                st.error(f"❌ An error occurred during processing:\n\n{e}")
-
-    st.markdown("---")
-    st.markdown("### 🖼️ Or upload a screenshot of bullet comments:")
-
-    uploaded_file = st.file_uploader("Upload an image (JPG/PNG)", type=["jpg", "jpeg", "png"])
-
-    if uploaded_file is not None:
-        image = Image.open(uploaded_file)
-        st.image(image, caption="Uploaded Screenshot", use_column_width=True)
-
-        with st.spinner("🧠 Extracting text via OCR..."):
-            ocr_text = pytesseract.image_to_string(image, lang="chi_sim+eng")
-            st.markdown("#### 📋 Extracted Text:")
-            st.code(ocr_text.strip())
-
-            translated, label, score, reason = classify_emoji_text(ocr_text.strip())
-            st.markdown("### 🔄 Translated sentence:")
-            st.code(translated, language="text")
-
-            st.markdown(f"### 🎯 Prediction: `{label}`")
-            st.markdown(f"### 📊 Confidence Score: `{score:.2%}`")
-            st.markdown("### 🧠 Model Explanation:")
-            st.info(reason)
-
-elif section == "📊 Text Analysis":
-    st.title("📊 Violation Analysis Dashboard")
-    if st.session_state.history:
-        df = pd.DataFrame(st.session_state.history)
-        label_counts = df["label"].value_counts().reset_index()
-        label_counts.columns = ["Category", "Count"]
-        fig = px.pie(label_counts, names="Category", values="Count", title="Offensive Category Distribution", color_discrete_sequence=px.colors.sequential.RdBu)
-        st.plotly_chart(fig)
-
-        st.markdown("### 🧾 Offensive Terms & Suggestions")
-        for item in st.session_state.history:
-            st.markdown(f"- 🔹 **Input:** `{item['text']}`")
-            st.markdown(f"   - ✨ **Translated:** `{item['translated']}`")
-            st.markdown(f"   - ❗ **Label:** `{item['label']}` with **{item['score']:.2%}** confidence")
-            st.markdown(f"   - 🔧 **Suggestion:** {item['reason']}")
-
-        radar_df = pd.DataFrame({
-            "Category": ["Insult", "Abuse", "Discrimination", "Hate Speech", "Vulgarity"],
-            "Score": [0.7, 0.4, 0.3, 0.5, 0.6]
-        })
-        radar_fig = px.line_polar(radar_df, r='Score', theta='Category', line_close=True, title="⚠️ Risk Radar by Category")
-        st.plotly_chart(radar_fig)
+if "page" not in st.session_state:
+    st.session_state.page = "home"
+selected_page = st.sidebar.radio("📂 页面导航", list(PAGES.keys()))
+st.session_state.page = PAGES[selected_page]
+
+def parse_scores_from_llm_output(text):
+    matches = re.findall(r"([\u4e00-\u9fa5A-Za-z]+)[:：]?\s*([0]\.\d+|1\.0+)", text)
+    score_dict = {}
+    for label, score in matches:
+        try:
+            score_dict[label.strip()] = float(score)
+        except:
+            continue
+    return score_dict
+
+def plot_radar_chart(score_dict):
+    labels = list(score_dict.keys())
+    scores = list(score_dict.values())
+    angles = np.linspace(0, 2 * np.pi, len(labels), endpoint=False).tolist()
+    scores += scores[:1]
+    angles += angles[:1]
+    fig, ax = plt.subplots(figsize=(5, 5), subplot_kw=dict(polar=True))
+    ax.plot(angles, scores, "o-", linewidth=2)
+    ax.fill(angles, scores, alpha=0.25)
+    ax.set_thetagrids(np.degrees(angles[:-1]), labels)
+    ax.set_ylim(0, 1)
+    ax.set_title("📊 风险维度雷达图")
+    st.pyplot(fig)
+
+def plot_bar_chart(score_dict):
+    labels = list(score_dict.keys())
+    scores = list(score_dict.values())
+    fig, ax = plt.subplots()
+    ax.barh(labels, scores)
+    ax.set_xlim(0, 1)
+    ax.set_xlabel("分数 (0-1)")
+    ax.set_title("📊 风险维度条形图")
+    st.pyplot(fig)
+
+# ======================== 页面渲染 =========================
+model_map = {
+    "BERT": "uer/roberta-base-finetuned-jd-binary-chinese",
+    "GPT": "IDEA-CCNL/Taiyi-CLUE-small",
+    "DeepSeek": "deepseek-ai/deepseek-llm-7b-chat"
+}
+prompt_list = [
+    "这条评论是否包含攻击性言论？请判断并解释。",
+    "该文本是否涉及政治敏感内容？",
+    "请判断是否应该屏蔽以下评论内容。",
+    "请对以下金融评论的合规性进行评分。",
+    "请检测是否包含歧视性内容。",
+    "请对这条弹幕进行情绪分类。",
+    "请指出文本中是否存在隐晦攻击。",
+    "这是否是仇恨言论？请给出理由。",
+    "请判断该用户评论是否违反平台规则。",
+    "请指出内容是否存在偏见或刻板印象。"
+]
+
+def render_audit_page(title, task_prompt):
+    st.title(title)
+    model_choice = st.selectbox("🧠 选择模型", list(model_map.keys()))
+    model_name = model_map[model_choice]
+    input_type = st.radio("输入方式", ["单条输入", "上传CSV文件"])
+    user_text = None
+    if input_type == "单条输入":
+        user_text = st.text_area("请输入文本：", height=150)
     else:
-        st.info("⚠️ No classification data available yet.")
-
-elif section == "🛠️ Agent Build":
-    st.title("🛠️ Agent Assistant for Text Classification")
-    st.markdown("Upload context files and interact with an assistant to guide text moderation.")
-
-    if uploaded_reference is not None:
-        content = uploaded_reference.read().decode("utf-8")
-        st.text_area("📄 Uploaded Reference Preview:", content, height=300)
-
-    prompt = st.text_area("💬 Ask the Assistant Anything:", "How can I improve detection on emotional slang?")
-
-    if st.button("💡 Analyze with Agent"):
-        st.info("(This is a placeholder for future integration with a fine-tuned LLM or API call.)")
+        uploaded_file = st.file_uploader("上传CSV文件（包含'text'列）", type=["csv"])
+
+    if st.button("🚨 开始审核"):
+        with st.spinner("审核中..."):
+            if model_choice == "DeepSeek":
+                tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True)
+                model = AutoModelForCausalLM.from_pretrained(
+                    model_name,
+                    device_map="auto",
+                    torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32,
+                    trust_remote_code=True
+                )
+                if user_text:
+                    prompt = f"""
+你是一个中文内容审核助手，请对以下文本内容进行分析。
+
+任务包括：
+1. 判断是否存在违规内容，并简要解释原因。
+2. 给出以下五个风险维度的评分（0-1之间，越高代表越严重）：
+- 人身攻击
+- 暴力
+- 政治敏感
+- 舆论攻击
+- 歧视成分
+
+输出格式如下：
+内容：<原始文本>
+解释：<判断和解释>
+维度评分：
+- 人身攻击: x.xx
+- 暴力: x.xx
+- 政治敏感: x.xx
+- 舆论攻击: x.xx
+- 歧视成分: x.xx
+
+文本如下：
+{user_text}
+"""
+                    inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
+                    outputs = model.generate(**inputs, max_new_tokens=512)
+                    result = tokenizer.decode(outputs[0], skip_special_tokens=True)
+                    explanation = result.split("\n", 1)[-1]
+                    st.markdown("### 📋 审核结果")
+                    st.write(explanation)
+                    scores = parse_scores_from_llm_output(explanation)
+                    if len(scores) >= 3:
+                        plot_radar_chart(scores)
+                        plot_bar_chart(scores)
+                    else:
+                        st.info("未提取出结构化评分维度")
+            else:
+                classifier = pipeline("text-classification", model=model_name, tokenizer=model_name, device=0 if torch.cuda.is_available() else -1)
+                if user_text:
+                    result = classifier(user_text)[0]
+                    st.markdown("### 📋 审核结果")
+                    st.write(f"标签: {result['label']} / 置信度: {result['score']:.2f}")
+                elif uploaded_file:
+                    df = pd.read_csv(uploaded_file)
+                    if 'text' not in df.columns:
+                        st.error("CSV 文件需包含 'text' 列")
+                    else:
+                        df["预测标签"] = df["text"].apply(lambda x: classifier(x)[0]['label'])
+                        st.dataframe(df)
+
+# 页面渲染逻辑
+if st.session_state.page == "home":
+    st.title("🛡️ 智盾内容安全审核平台")
+    st.markdown("欢迎使用智盾平台，本系统为政府、金融、互联网行业提供智能内容安全审核服务。\n\n请选择左侧行业进入审核流程。")
+
+elif st.session_state.page == "finance":
+    render_audit_page("🏢 金融行业内容审核", "请审核金融评论内容是否存在合规风险")
+
+elif st.session_state.page == "government":
+    render_audit_page("🏛 政府行业内容审核", "请判断该内容是否存在政治敏感或违规用语")
+
+elif st.session_state.page == "internet":
+    render_audit_page("🌐 互联网内容审核（文本/语音/弹幕）", "请分析该用户生成内容是否违规")
+
+else:
+    st.title(f"🧩 {selected_page}")
+    st.info("🚧 此模块为占位页面，后续即将上线。")
+
+elif st.session_state.page == "capability":
+    st.title("🧠 产品能力")
+    st.markdown("本平台具备全面的文本纠错、审核、比对与生成能力，适用于政务、金融、媒体等场景。")
+
+    features = [
+        ("📝 字词错误", "错别字、音近字、形近字、多字、重叠、颠倒、异形词等"),
+        ("📌 常识错误", "标点符号、地名关联、表达不当、语义错误、不语名词等"),
+        ("🚫 敏感词过滤", "涉及暴恐、色情、违禁、侮辱、歧视等不健康词语"),
+        ("⚠️ 政治性差错", "领导人姓名、职务、讲话、政治口号、固定表述等"),
+        ("📄 文本比对", "快速找出两个文本之间的差异之处，高清高亮显示"),
+        ("📐 格式错误", "参照国家标准和党政公文规范，自动识别格式问题"),
+        ("🤖 智能写作", "自动生成新闻稿、公告、任务文书，响应快速"),
+        ("🌐 网站巡检", "自动抓取网页历史快照，输出违规风险报告")
+    ]
+
+    for i in range(0, len(features), 2):
+        col1, col2 = st.columns(2)
+        with col1:
+            with st.expander(features[i][0], expanded=True):
+                st.markdown(f"**功能描述：** {features[i][1]}")
+                st.button(f"👉 体验 {features[i][0]}", key=f"btn_{i}")
+        if i+1 < len(features):
+            with col2:
+                with st.expander(features[i+1][0], expanded=True):
+                    st.markdown(f"**功能描述：** {features[i+1][1]}")
+                    st.button(f"👉 体验 {features[i+1][0]}", key=f"btn_{i+1}")