Create app.py

app.py

@@ -0,0 +1,89 @@
+# ==========================================
+# 高雄市長選戰輕量化輿情分析 (HF Spaces)
+# ==========================================
+
+import snscrape.modules.twitter as sntwitter
+import pandas as pd
+from datetime import datetime, timedelta
+from transformers import pipeline
+import matplotlib.pyplot as plt
+import io
+import base64
+import gradio as gr
+
+# -----------------------------
+# 參數設定
+# -----------------------------
+candidates = ["許智傑", "邱議瑩", "賴瑞隆", "林岱樺", "柯志恩"]
+days_back = 7
+max_tweets_per_candidate = 100  # HF部署建議少量
+
+# -----------------------------
+# 中文情緒分析模型
+# -----------------------------
+sentiment = pipeline("sentiment-analysis", model="uer/roberta-base-finetuned-chinanews-chinese-sentiment")
+
+# -----------------------------
+# 主分析函數
+# -----------------------------
+def run_analysis():
+    since_date = (datetime.now() - timedelta(days=days_back)).strftime('%Y-%m-%d')
+    until_date = datetime.now().strftime('%Y-%m-%d')
+
+    # 1. 抓貼文
+    all_tweets = []
+    for candidate in candidates:
+        query = f'{candidate} since:{since_date} until:{until_date}'
+        for i, tweet in enumerate(sntwitter.TwitterSearchScraper(query).get_items()):
+            if i >= max_tweets_per_candidate:
+                break
+            all_tweets.append([tweet.date, tweet.user.username, tweet.content, candidate])
+    df_tweets = pd.DataFrame(all_tweets, columns=["日期", "使用者", "內容", "候選人"])
+
+    # 2. 情緒分析
+    df_tweets['情緒'] = df_tweets['內容'].apply(lambda x: sentiment(x)[0]['label'])
+    df_tweets['信心度'] = df_tweets['內容'].apply(lambda x: sentiment(x)[0]['score'])
+
+    summary = df_tweets.groupby(['候選人', '情緒']).size().unstack(fill_value=0)
+    summary['總貼文'] = summary.sum(axis=1)
+    summary['正面比率'] = summary.get('positive', 0) / summary['總貼文']
+    summary['負面比率'] = summary.get('negative', 0) / summary['總貼文']
+
+    # 3. 可視化
+    plt.figure(figsize=(8,5))
+    summary[['正面比率','負面比率']].plot(kind='bar', stacked=True, colormap='coolwarm')
+    plt.title("候選人社群情緒比例")
+    plt.ylabel("比例")
+    plt.xlabel("候選人")
+    plt.xticks(rotation=0)
+    plt.tight_layout()
+
+    buf = io.BytesIO()
+    plt.savefig(buf, format="png")
+    buf.seek(0)
+    img_b64 = base64.b64encode(buf.read()).decode("utf-8")
+    buf.close()
+
+    # 4. 即時輿情報告 (HTML)
+    html_report = f"""
+    <h2>高雄市長選戰輿情摘要 ({datetime.now().strftime('%Y-%m-%d')})</h2>
+    <h3>1. 社群貼文聲量與情緒</h3>
+    {summary.to_html()}
+    <h3>2. 情緒圖表</h3>
+    <img src='data:image/png;base64,{img_b64}' width='600'>
+    """
+    return html_report
+
+# -----------------------------
+# Gradio 前端
+# -----------------------------
+iface = gr.Interface(
+    fn=run_analysis,
+    inputs=[],
+    outputs=gr.HTML,
+    live=False,
+    title="高雄市長選戰輿情分析",
+    description="抓取 X（Twitter）貼文、分析情緒比例，並生成即時輿情報告"
+)
+
+iface.launch()