Upload 3 files

README.md

@@ -1,12 +1,65 @@
+# 🧠 Cho's Risk Scoring App (Powered by DSPy & Gradio)
+
+A real-time risk evaluation tool for retail investors, built with LLM-based Chain-of-Thought (CoT) reasoning.  
+This app analyzes stock-related questions and visualizes risk scores across 6 dimensions using natural language inputs.
+
+---
+
+## 📌 How to Use
+
+1. Enter a natural language query like:
+   ```
+   Is Tesla risky these days?
+   삼성전자는 지금 고평가됐어?
+   ```
+2. Click `Submit`.
+3. The app will automatically:
+   - Detect the stock ticker (US or Korea)
+   - Fetch recent Yahoo Finance headlines
+   - Run CoT-based risk scoring via OpenAI GPT-4o
+   - Display results as:
+     - 📈 1-Month price trend (Plotly)
+     - 📰 Headline list (HTML)
+     - 📊 Risk breakdown bar chart (Matplotlib)
+     - 🧠 Detailed reasoning per risk category
+
+---
+
+## 🧠 Risk Factors Evaluated
+
+- Overvaluation  
+- Poor Earnings  
+- Financial Instability  
+- Theme Overheating  
+- Recurring Negatives  
+- FII Sell-off
+
 ---
-title: Cho Risk Scoring
-emoji: 😻
-colorFrom: red
-colorTo: yellow
-sdk: gradio
-sdk_version: 5.38.0
-app_file: app.py
-pinned: false
+
+## 🛠️ Tech Stack
+
+- `Gradio`: frontend UI
+- `DSPy`: CoT prompting & reasoning
+- `yfinance`: stock price and metadata
+- `Yahoo RSS`: finance news headlines
+- `matplotlib` / `plotly`: data visualization
+
+---
+
+## 🔐 Environment Variable (required)
+
+This app uses OpenAI's GPT-4o via the DSPy framework.  
+You **must** define the following secret in Hugging Face:
+
+```
+OPENAI_API_KEY=sk-...  ← your actual OpenAI API key
+```
+
+Add it in your Hugging Face Space → `Settings` → `Secrets`.
+
 ---
 
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+## 🏫 About
+
+Developed by HYU researcher Cho  
+Based on [DSPy](https://github.com/stanfordnlp/dspy) by Stanford NLP Group

app.py

@@ -0,0 +1,289 @@
+import dspy
+import yfinance as yf
+import os
+import requests
+import re
+import matplotlib.pyplot as plt
+import gradio as gr
+from bs4 import BeautifulSoup
+from datetime import datetime, timedelta, timezone
+import io
+import textwrap
+import plotly.graph_objects as go
+import plotly.io as pio
+
+# === 환경 설정 ===
+os.environ["USER_AGENT"] = "Mozilla/5.0"
+original_get = requests.get
+def patched_get(url, *args, **kwargs):
+    headers = kwargs.get("headers", {})
+    headers.update({"User-Agent": os.environ["USER_AGENT"]})
+    kwargs["headers"] = headers
+    return original_get(url, *args, **kwargs)
+requests.get = patched_get
+
+# === LLM 설정 ===
+api_key = os.environ.get("OPENAI_API_KEY")
+lm = dspy.LM("openai/gpt-4o", api_key=api_key)
+dspy.configure(lm=lm)
+
+# === 티커 추출 서명 ===
+class ExtractTicker(dspy.Signature):
+    user_input: str = dspy.InputField()
+    suggested_ticker: str = dspy.OutputField()
+
+ticker_extractor = dspy.Predict(ExtractTicker)
+
+# === 한국 종목 티커 추출 서명 ===
+class ExtractKRStock(dspy.Signature):
+    user_input: str = dspy.InputField()
+    suggested_kr_ticker: str = dspy.OutputField()
+
+kr_extractor = dspy.Predict(ExtractKRStock)
+
+def extract_ticker(user_input: str) -> str:
+    result = ticker_extractor(user_input=user_input)
+    ticker = result.suggested_ticker.strip().upper()
+
+    # 미국식 또는 한국식 티커면 그대로
+    if re.fullmatch(r"[A-Z]{1,5}(\.[A-Z]{2})?", ticker) or re.fullmatch(r"\d{6}\.KS", ticker):
+        return ticker
+
+    # 아니면 한글 기업명 추정 → 한국 티커 추출
+    kr_result = kr_extractor(user_input=user_input)
+    return kr_result.suggested_kr_ticker.strip().upper()
+
+# === 뉴스 수집 ===
+def get_yahoo_news(ticker: str):
+    try:
+        url = f"https://feeds.finance.yahoo.com/rss/2.0/headline?s={ticker}&region=US&lang=en-US"
+        soup = BeautifulSoup(requests.get(url).content, "xml")
+        one_week_ago = datetime.now(timezone.utc) - timedelta(days=7)
+        return [
+            {
+                "title": i.title.text,
+                "source": "Yahoo Finance",
+                "link": i.link.text
+            }
+            for i in soup.find_all("item")
+            if datetime.strptime(i.pubDate.text, "%a, %d %b %Y %H:%M:%S %z") >= one_week_ago
+        ][:6]
+    except:
+        return []
+
+# === 주가 정보 수집 ===
+def fetch_stock_info(ticker: str):
+    try:
+        stock = yf.Ticker(ticker)
+        hist = stock.history(period="1d")
+        info = stock.info
+        if hist.empty or "longName" not in info:
+            return None
+        return {
+            "company": info["longName"],
+            "price": round(hist["Close"].iloc[-1], 2),
+            "change_percent": round((hist["Close"].iloc[-1] - info.get("previousClose", 0)) / max(info.get("previousClose", 1), 1) * 100, 2)
+        }
+    except:
+        return None
+
+# === 리스크 스코어링 서명 ===
+class StructuredRiskScoringSignature(dspy.Signature):
+    stock_info: str = dspy.InputField()
+    news: str = dspy.InputField()
+    overvaluation_score: str = dspy.OutputField()
+    overvaluation_reasoning: str = dspy.OutputField()
+    poor_earnings_score: str = dspy.OutputField()
+    poor_earnings_reasoning: str = dspy.OutputField()
+    financial_instability_score: str = dspy.OutputField()
+    financial_instability_reasoning: str = dspy.OutputField()
+    theme_overheating_score: str = dspy.OutputField()
+    theme_overheating_reasoning: str = dspy.OutputField()
+    recurring_negatives_score: str = dspy.OutputField()
+    recurring_negatives_reasoning: str = dspy.OutputField()
+    selloff_score: str = dspy.OutputField()
+    selloff_reasoning: str = dspy.OutputField()
+    total_score: str = dspy.OutputField()
+    risk_level: str = dspy.OutputField()
+    investment_message: str = dspy.OutputField()
+
+risk_model = dspy.ChainOfThought(StructuredRiskScoringSignature)
+
+def create_price_plot(ticker: str):
+    try:
+        stock = yf.Ticker(ticker)
+        hist = stock.history(period="1mo")
+
+        if hist.empty:
+            return None
+
+        fig = go.Figure()
+        fig.add_trace(go.Scatter(
+            x=hist.index,
+            y=hist["Close"],
+            mode='lines+markers',
+            name='Close Price',
+            line=dict(color='royalblue'),
+            marker=dict(size=6)
+        ))
+
+        fig.update_layout(
+            title=f"{ticker} Price Trend (1mo)",
+            xaxis_title="Date",
+            yaxis_title="Close Price ($)",
+            font=dict(
+                family="Arial, sans-serif",
+                size=14,
+                 color="#333333"
+            ),
+            template="plotly_white",
+            height=300,
+            margin=dict(l=20, r=20, t=40, b=20),
+            plot_bgcolor="#fefbd8",     # 내부
+            paper_bgcolor="#e0f7fa"     # 외부
+        )
+
+        return fig  # 직접 반환 (Gradio가 Plotly 지원함)
+
+    except Exception as e:
+        print("Plotly price chart error:", e)
+        return None
+
+# === 시각화 ===
+from PIL import Image
+
+def create_risk_plot(result, company_name, ticker, total_score, risk_level, investment_message):
+    categories = [
+        "Overval.", "Earnings", "Fin. Instab.",
+        "Theme OH", "Neg. News", "FII Sell"
+    ]
+    scores = list(map(int, [
+        result.overvaluation_score,
+        result.poor_earnings_score,
+        result.financial_instability_score,
+        result.theme_overheating_score,
+        result.recurring_negatives_score,
+        result.selloff_score
+    ]))
+
+    # 📐 전체 배경색 (figure)
+    plt.figure(figsize=(10, 8), facecolor='#e0f7fa')
+
+    # 🎨 AXES 배경색
+    ax = plt.gca()
+    ax.set_facecolor('#fefbd8')  # 예: 흰색 배경, 혹은 '#fefbd8', 'aliceblue' 등
+
+    bars = plt.bar(categories, scores, color='pink')
+    plt.ylim(0, 10)
+    plt.title(f"{company_name} ({ticker})", fontsize=30)
+    plt.ylabel("Score")
+    plt.xlabel("Category")
+
+    for bar in bars:
+        height = bar.get_height()
+        plt.text(bar.get_x() + bar.get_width() / 2, height + 0.2, str(height), ha='center', fontsize=12)
+
+    plt.text(
+        0.5, 0.91,
+        f"Total Score: {total_score} / Risk Level: {risk_level}",
+        fontsize=20, ha='center', transform=plt.gca().transAxes
+    )
+
+    # 여백 조정
+    plt.subplots_adjust(top=0.85, bottom=0.10)
+
+    buf = io.BytesIO()
+    plt.savefig(buf, format='png')
+    buf.seek(0)
+    plt.close()
+
+    return Image.open(buf)
+
+# === Gradio 함수 ===
+def gradio_run(user_input: str):
+    ticker = extract_ticker(user_input)
+    if not ticker:
+        return "❌ Unable to recognize a valid stock ticker from your input.", None
+
+    stock = fetch_stock_info(ticker)
+    if not stock:
+        return f"❌ Failed to retrieve stock information for: {ticker}", None
+
+    news = get_yahoo_news(ticker)
+
+    if news:
+        news_html = "<h4>📰 Cho's Pick Headlines (Recent 7 Days)</h4><ul>" + "".join(
+             f"<li><a href='{n['link']}' target='_blank'>{n['title']}</a> ({n['source']})</li>"
+             for n in news
+        ) + "</ul>"
+    else:
+        news_html = "<h4>📰 Cho's Pick Headlines</h4><p>(No recent news available)</p>"
+    
+    stock_text = f"Company: {stock['company']}\nPrice: ${stock['price']} ({stock['change_percent']}%)"
+
+    result = risk_model(stock_info=stock_text, news=news_html)
+
+    summary = f"""📌 Ticker: {stock['company']} ({ticker})  
+🧮 Total Score: {result.total_score}  
+⚠️ Risk Level: {result.risk_level}  
+💬 Recommendation: {result.investment_message}"""
+
+    plot_img = create_risk_plot(
+        result, stock["company"], ticker, result.total_score,
+        result.risk_level, result.investment_message
+    )
+    price_img = create_price_plot(ticker)
+    # 🧠 리스크 항목별 설명
+    annotations = f"""🧠 Category-wise Reasoning:
+1️⃣ Overvaluation: {result.overvaluation_reasoning}
+2️⃣ Poor Earnings: {result.poor_earnings_reasoning}
+3️⃣ Financial Instability: {result.financial_instability_reasoning}
+4️⃣ Theme Overheating: {result.theme_overheating_reasoning}
+5️⃣ Recurring Negatives: {result.recurring_negatives_reasoning}
+6️⃣ FII Sell-off: {result.selloff_reasoning}
+"""
+
+    return price_img, news_html, summary, plot_img, annotations
+
+# === Gradio 인터페이스 실행 ===
+with gr.Blocks() as iface:
+    # ✅ 여기에 로고+제목 HTML 삽입
+    gr.HTML("""
+        <div style="display: flex; align-items: center; margin-bottom: 10px;">
+            <img src="https://www.hanyang.ac.kr/documents/20182/73809/HYU_logo_singlecolor_png.png/b8aabfbe-a488-437d-b4a5-bd616d1577da?t=1474070795276" style="height: 50px; margin-right: 10px;">
+            <h2 style="margin: 0;">HYU-Cho's 'Risk Scoring Model' for Retail Portfolios based on Chain-of-Thought</h2>
+        </div>
+        <p>Analyze and visualize the risk level of any stock using natural language input.</p>
+    """)
+
+    # ✅ 안내 문구 추가
+    with gr.Row():
+        gr.Markdown("📌 **Welcome to Cho's Risk Scoring Model. Enter a stock-related question to begin.**")
+
+    with gr.Row():
+        user_input = gr.Textbox(label="User Input", lines=2, placeholder="e.g. Is Tesla risky these days?")
+        submit_btn = gr.Button("Submit", variant="primary")
+        clear_btn = gr.Button("Clear")
+
+    # 📈 주가 흐름 제일 위로
+    output_price_plot = gr.Plot(label="📈 1-Month Price Trend")
+
+    # 📰 뉴스 (HTML 클릭 가능 + 타이틀)
+    output_news_only = gr.HTML(label=None)  # 헤드라인 타이틀은 HTML 내에서 직접 표현
+
+    # 📋 요약
+    output_summary = gr.Textbox(label="📋 Portfolio Risk Evaluation by CoT", lines=6)
+
+    # 📊 위험도 시각화 + 상세 설명
+    with gr.Row():
+        output_plot = gr.Image(label="📊 Risk Score Visualization")
+        output_detail = gr.Textbox(label="🧠 Detailed Risk Reasoning", lines=25)
+
+    # 실행
+    submit_btn.click(fn=gradio_run, inputs=user_input,
+                     outputs=[output_price_plot, output_news_only, output_summary, output_plot, output_detail])
+
+    clear_btn.click(lambda: ("", "", "", None, ""), inputs=[], 
+                    outputs=[output_price_plot, output_news_only, output_summary, output_plot, output_detail])
+
+    iface.launch()

requirements.txt

@@ -0,0 +1,11 @@
+gradio
+dspy
+yfinance
+requests
+beautifulsoup4
+plotly
+matplotlib
+pillow
+openai
+tiktoken
+httpx