app.py

import streamlit as st
import requests
import pandas as pd
from datetime import datetime, timedelta
from transformers import pipeline
from vaderSentiment.vaderSentiment import SentimentIntensityAnalyzer
from textblob import TextBlob
import nltk
from wordcloud import WordCloud
import base64
from io import BytesIO
import numpy as np
from sklearn.linear_model import LinearRegression
import plotly.graph_objects as go

# --------------------------
# CONFIG
# --------------------------
st.set_page_config(page_title="📰 SentimentSync NewsAI", layout="wide")
API_KEY = "88bc396d4eab4be494a4b86ec842db47"

# --------------------------
# UTILITIES
# --------------------------
@st.cache_resource
def load_models():
    st.info("Loading sentiment models...")
    bert_model = pipeline("sentiment-analysis", model="nlptown/bert-base-multilingual-uncased-sentiment")
    vader = SentimentIntensityAnalyzer()
    return bert_model, vader


def analyze_text(text, bert_model, vader):
    if not text.strip():
        return 0

    vader_score = vader.polarity_scores(text)["compound"]
    textblob_score = TextBlob(text).sentiment.polarity
    bert_result = bert_model(text[:512])[0]
    label_map = {
        "1 star": -1,
        "2 stars": -0.5,
        "3 stars": 0,
        "4 stars": 0.5,
        "5 stars": 1
    }
    bert_score = label_map.get(bert_result["label"], 0)
    return np.mean([vader_score, textblob_score, bert_score])


@st.cache_data(ttl=3600)
def fetch_financial_news(keyword, days=7):
    """ดึงข่าวย้อนหลังจาก NewsAPI.org ตามจำนวนวัน"""
    to_date = datetime.now().strftime('%Y-%m-%d')
    from_date = (datetime.now() - timedelta(days=days)).strftime('%Y-%m-%d')

    all_articles = []
    page = 1
    while True:
        url = (
            f"https://newsapi.org/v2/everything?"
            f"q={keyword}+finance+stock&"
            f"from={from_date}&to={to_date}&"
            f"language=en&sortBy=publishedAt&"
            f"pageSize=100&page={page}&apiKey={API_KEY}"
        )
        r = requests.get(url)
        data = r.json()

        if data.get("status") != "ok":
            st.error(f"API Error: {data}")
            break

        articles = data.get("articles", [])
        if not articles:
            break

        for a in articles:
            if a["description"]:
                all_articles.append({
                    "date": pd.to_datetime(a["publishedAt"]),
                    "text": f"{a['title']} {a['description']}",
                    "source": a["source"]["name"],
                    "url": a["url"]
                })

        if len(articles) < 100:
            break  # หมดแล้ว
        page += 1

    return pd.DataFrame(all_articles)


def generate_wordcloud(text):
    stopwords = nltk.corpus.stopwords.words('english')
    wordcloud = WordCloud(width=800, height=400, background_color="white", stopwords=stopwords).generate(text)
    buf = BytesIO()
    wordcloud.to_image().save(buf, format="PNG")
    return base64.b64encode(buf.getvalue()).decode()


# --------------------------
# MAIN APP
# --------------------------
def main():
    st.title("📰 SentimentSync NewsAI")
    st.markdown("วิเคราะห์แนวโน้มอารมณ์ของข่าวการเงินย้อนหลังตามจำนวนวันที่เลือก พร้อมพยากรณ์ในอนาคต")

    # Sidebar
    with st.sidebar:
        keyword = st.text_input("ค้นหาคำ (เช่น Tesla, Bitcoin, Inflation):", "")
        days = st.slider("จำนวนวันย้อนหลัง:", 7, 30, 7)
        analyze_btn = st.button("วิเคราะห์เลย")

    if not analyze_btn:
        st.info("กรอกคำค้นแล้วกด 'วิเคราะห์เลย' เพื่อเริ่มต้น")
        return

    bert_model, vader = load_models()

    # ดึงข่าว
    st.info(f"กำลังดึงข่าวย้อนหลัง {days} วัน จาก NewsAPI.org สำหรับ '{keyword}' ...")
    news_df = fetch_financial_news(keyword, days=days)
    if news_df.empty:
        st.warning("ไม่พบบทความข่าวในช่วงเวลาที่เลือก")
        return

    # วิเคราะห์ sentiment
    st.info("กำลังวิเคราะห์อารมณ์ของข่าว...")
    news_df["sentiment"] = news_df["text"].apply(lambda x: analyze_text(x, bert_model, vader))
    news_df["date"] = pd.to_datetime(news_df["date"])

    avg_sentiment = news_df["sentiment"].mean()
    pos_pct = (news_df["sentiment"] > 0.1).mean() * 100
    neg_pct = (news_df["sentiment"] < -0.1).mean() * 100

    col1, col2, col3 = st.columns(3)
    col1.metric("ค่าเฉลี่ยอารมณ์ข่าว", f"{avg_sentiment:.2f}",
                "Positive" if avg_sentiment > 0 else "Negative" if avg_sentiment < 0 else "Neutral")
    col2.metric("ข่าวเชิงบวก", f"{pos_pct:.1f}%")
    col3.metric("ข่าวเชิงลบ", f"{neg_pct:.1f}%")

    # Wordcloud
    st.subheader("☁️ Word Cloud ของข่าว")
    all_text = " ".join(news_df["text"].tolist())
    img = generate_wordcloud(all_text)
    st.image(f"data:image/png;base64,{img}", use_column_width=True)

    # แนวโน้มและพยากรณ์ในกราฟเดียว
    st.subheader("📈 แนวโน้มและพยากรณ์อารมณ์ของข่าว")

    df_sorted = news_df.sort_values("date").copy()
    df_sorted["timestamp"] = (df_sorted["date"] - df_sorted["date"].min()).dt.days

    # Train model
    model = LinearRegression()
    model.fit(df_sorted[["timestamp"]], df_sorted["sentiment"])

    # Forecast next 7 days
    future_days = 7
    future_timestamps = np.arange(df_sorted["timestamp"].max() + 1, df_sorted["timestamp"].max() + future_days + 1)
    future_dates = [df_sorted["date"].max() + timedelta(days=i) for i in range(1, future_days + 1)]
    future_preds = model.predict(future_timestamps.reshape(-1, 1))

    # Plot both actual + prediction
    fig = go.Figure()

    fig.add_trace(go.Scatter(
        x=df_sorted["date"], y=df_sorted["sentiment"],
        mode="lines+markers", name="Actual Sentiment",
        line=dict(color="blue")
    ))

    fig.add_trace(go.Scatter(
        x=future_dates, y=future_preds,
        mode="lines+markers", name="Predicted Sentiment (7-day Forecast)",
        line=dict(color="orange", dash="dash")
    ))

    fig.add_trace(go.Scatter(
        x=future_dates + future_dates[::-1],
        y=list(future_preds + 0.1) + list((future_preds - 0.1)[::-1]),
        fill='toself', fillcolor='rgba(255,165,0,0.2)',
        line=dict(color='rgba(255,255,255,0)'),
        hoverinfo="skip",
        showlegend=False
    ))

    fig.update_layout(
        title=f"แนวโน้มและพยากรณ์อารมณ์ของข่าว '{keyword}'",
        xaxis_title="วันที่",
        yaxis_title="ค่าอารมณ์ (Sentiment)",
        hovermode="x unified",
        template="plotly_white"
    )
    st.plotly_chart(fig, use_container_width=True)

    st.subheader("📰 รายการข่าว")
    st.dataframe(news_df[["date", "source", "text", "sentiment", "url"]], use_container_width=True)


if __name__ == "__main__":
    nltk.download("stopwords", quiet=True)
    main()