app.py

import gradio as gr
import feedparser
import pandas as pd
import numpy as np
import faiss
import matplotlib.pyplot as plt
import re
from collections import Counter
from sentence_transformers import SentenceTransformer
from transformers import AutoTokenizer, AutoModelForSequenceClassification, pipeline
import torch
import warnings

warnings.filterwarnings('ignore')

# Global değişkenler
model = None
sentiment_analyzer = None
summarizer = None
df = None
index = None
embeddings = None

def initialize_models():
    """FinBERT ve özetleme modellerini yükle"""
    global model, sentiment_analyzer, summarizer
    
    try:
        if model is None:
            model = SentenceTransformer('all-MiniLM-L6-v2')
        
        if sentiment_analyzer is None:
            # FinBERT modelini yükle
            tokenizer = AutoTokenizer.from_pretrained("ProsusAI/finbert")
            finbert_model = AutoModelForSequenceClassification.from_pretrained("ProsusAI/finbert")
            
            sentiment_analyzer = pipeline(
                "sentiment-analysis",
                model=finbert_model,
                tokenizer=tokenizer,
                device=-1  # CPU kullan (GPU için 0)
            )
        
        if summarizer is None:
            # Özetleme modeli
            summarizer = pipeline(
                "summarization",
                model="facebook/bart-large-cnn",
                device=-1
            )
        
        return "✅ FinBERT ve özetleme modelleri başarıyla yüklendi!"
    except Exception as e:
        return f"❌ Model yükleme hatası: {str(e)}"

def fetch_news():
    """RSS feedlerinden haberleri topla"""
    global df, index, embeddings
    
    RSS_URLS = [
        "https://cointelegraph.com/rss",
        "https://cryptonews.com/news/feed",
        "https://www.coindesk.com/arc/outboundfeeds/rss/",
        "https://tr.investing.com/rss/302.rss"
    ]
    
    all_entries = []
    status_messages = []
    
    for url in RSS_URLS:
        try:
            feed = feedparser.parse(url)
            for entry in feed.entries[:30]:
                all_entries.append({
                    "title": entry.get("title", ""),
                    "link": entry.get("link", ""),
                    "summary": entry.get("summary", ""),
                    "published": entry.get("published", "")
                })
            status_messages.append(f"✓ {url.split('/')[2]} - {len(feed.entries[:30])} haber")
        except Exception as e:
            status_messages.append(f"✗ {url.split('/')[2]} - Hata")
    
    df = pd.DataFrame(all_entries).drop_duplicates(subset="title").reset_index(drop=True)
    
    # FinBERT ile sentiment analizi
    if df is not None and len(df) > 0:
        def analyze_sentiment_finbert(text):
            try:
                # FinBERT maksimum 512 token kabul eder
                text = text[:512]
                result = sentiment_analyzer(text)[0]
                
                # FinBERT çıktıları: positive, negative, neutral
                label = result['label'].lower()
                score = result['score']
                
                return label, score
            except Exception as e:
                return "neutral", 0.5
        
        df["sentiment_label"], df["sentiment_score"] = zip(*df["title"].apply(analyze_sentiment_finbert))
        
        # FAISS index oluştur
        corpus = df['title'].tolist()
        embeddings = model.encode(corpus)
        dimension = embeddings.shape[1]
        index = faiss.IndexFlatL2(dimension)
        index.add(embeddings.astype('float32'))
    
    status = "\n".join(status_messages)
    status += f"\n\n✅ Toplam {len(df)} benzersiz haber toplandı ve FinBERT ile analiz edildi!"
    
    return status, df[["title", "sentiment_label", "published"]].head(10)

def generate_news_summary():
    """Haberlerin genel özetini oluştur"""
    global df
    
    if df is None or len(df) == 0:
        return "⚠️ Önce haberleri toplamalısınız!", None, None
    
    try:
        # Sentiment istatistikleri
        sentiment_counts = df["sentiment_label"].value_counts()
        total_news = len(df)
        
        positive_count = sentiment_counts.get('positive', 0)
        negative_count = sentiment_counts.get('negative', 0)
        neutral_count = sentiment_counts.get('neutral', 0)
        
        positive_pct = (positive_count / total_news) * 100
        negative_pct = (negative_count / total_news) * 100
        neutral_pct = (neutral_count / total_news) * 100
        
        # Genel duygu
        if positive_pct > 50:
            overall_sentiment = "📈 POZITIF"
            sentiment_emoji = "🟢"
        elif negative_pct > 50:
            overall_sentiment = "📉 NEGATIF"
            sentiment_emoji = "🔴"
        else:
            overall_sentiment = "➡️ NÖTR"
            sentiment_emoji = "🟡"
        
        # En çok bahsedilen kelimeler
        all_titles = " ".join(df["title"].tolist())
        potential_coins = re.findall(r'\b[A-Z][a-zA-Z]+\b', all_titles)
        stop_words = {
            "Coin", "Price", "Market", "News", "Today", "Crypto", "Token",
            "The", "This", "That", "What", "When", "Where", "How", "Why",
            "And", "But", "For", "From", "With", "About", "After", "Before"
        }
        filtered_tokens = [t for t in potential_coins if t not in stop_words]
        trending = Counter(filtered_tokens).most_common(5)
        
        # En önemli haberler
        top_positive = df.nlargest(3, "sentiment_score")[["title", "sentiment_score"]]
        top_negative = df.nsmallest(3, "sentiment_score")[["title", "sentiment_score"]]
        
        # Özet metni oluştur
        summary_text = f"""
# 🌐 GÜNCEL KRİPTO HABER ÖZETİ

## {sentiment_emoji} Genel Durum: {overall_sentiment}

### 📊 Sentiment Dağılımı
- **Toplam Haber:** {total_news}
- 🟢 **Pozitif:** {positive_count} (%{positive_pct:.1f})
- 🔴 **Negatif:** {negative_count} (%{negative_pct:.1f})
- 🟡 **Nötr:** {neutral_count} (%{neutral_pct:.1f})

### 🔥 En Çok Bahsedilen Konular
"""
        for i, (term, count) in enumerate(trending, 1):
            summary_text += f"{i}. **{term}** - {count} kez bahsedildi\n"
        
        summary_text += "\n### ⭐ En Pozitif Haberler\n"
        for idx, row in top_positive.iterrows():
            summary_text += f"- {row['title'][:100]}... (Skor: {row['sentiment_score']:.3f})\n"
        
        summary_text += "\n### ⚠️ En Negatif Haberler\n"
        for idx, row in top_negative.iterrows():
            summary_text += f"- {row['title'][:100]}... (Skor: {row['sentiment_score']:.3f})\n"
        
        # Grafik oluştur
        fig, axes = plt.subplots(1, 2, figsize=(14, 5))
        
        # Sentiment dağılımı
        ax1 = axes[0]
        color_map = {
            'positive': '#2ecc71',
            'negative': '#e74c3c',
            'neutral': '#95a5a6'
        }
        colors = [color_map.get(x, '#95a5a6') for x in sentiment_counts.index]
        
        ax1.pie(sentiment_counts.values, labels=sentiment_counts.index, autopct='%1.1f%%',
                colors=colors, startangle=90, textprops={'fontsize': 12, 'fontweight': 'bold'})
        ax1.set_title('Genel Sentiment Dağılımı', fontweight='bold', fontsize=14)
        
        # Trend grafiği
        ax2 = axes[1]
        if len(trending) > 0:
            terms, counts = zip(*trending)
            ax2.barh(terms, counts, color='#3498db', alpha=0.7)
            ax2.set_xlabel('Frekans', fontweight='bold', fontsize=11)
            ax2.set_title('En Popüler 5 Konu', fontweight='bold', fontsize=14)
            ax2.invert_yaxis()
            ax2.grid(axis='x', alpha=0.3)
        
        plt.tight_layout()
        
        # Özet tablo
        summary_table = pd.DataFrame({
            'Metric': ['Toplam Haber', 'Pozitif', 'Negatif', 'Nötr', 'Genel Durum'],
            'Value': [total_news, positive_count, negative_count, neutral_count, overall_sentiment]
        })
        
        return summary_text, fig, summary_table
        
    except Exception as e:
        return f"❌ Özet oluşturma hatası: {str(e)}", None, None

def search_similar_news(query, top_k=5):
    """Semantik arama"""
    global df, index, model
    
    if df is None or index is None:
        return "⚠️ Önce haberleri toplamalısınız!", None
    
    try:
        q_embedding = model.encode([query])
        distances, indices = index.search(q_embedding.astype('float32'), k=min(top_k, len(df)))
        
        results = []
        for idx, dist in zip(indices[0], distances[0]):
            news = df.iloc[idx]
            results.append({
                "Başlık": news['title'],
                "Sentiment": news['sentiment_label'],
                "Skor": f"{news['sentiment_score']:.3f}",
                "Link": news['link']
            })
        
        results_df = pd.DataFrame(results)
        message = f"🔎 '{query}' için {len(results)} sonuç bulundu"
        
        return message, results_df
    
    except Exception as e:
        return f"❌ Hata: {str(e)}", None

def analyze_coin_sentiment(coin_name):
    """Coin bazlı sentiment analizi"""
    global df
    
    if df is None:
        return "⚠️ Önce haberleri toplamalısınız!", None, None
    
    filtered = df[df["title"].str.contains(coin_name, case=False, na=False)]
    
    if len(filtered) == 0:
        return f"⚠️ '{coin_name}' hakkında haber bulunamadı.", None, None
    
    # Sentiment dağılımı
    sentiment_dist = filtered["sentiment_label"].value_counts()
    
    # Grafik oluştur
    fig, (ax1, ax2) = plt.subplots(1, 2, figsize=(12, 4))
    
    # Renk haritası (FinBERT için)
    color_map = {
        'positive': '#2ecc71',
        'negative': '#e74c3c',
        'neutral': '#95a5a6'
    }
    colors = [color_map.get(x, '#95a5a6') for x in sentiment_dist.index]
    
    # Bar chart
    ax1.bar(sentiment_dist.index, sentiment_dist.values, color=colors, alpha=0.7)
    ax1.set_title(f'{coin_name} - Sentiment Dağılımı (FinBERT)', fontweight='bold')
    ax1.set_ylabel('Haber Sayısı')
    ax1.grid(axis='y', alpha=0.3)
    
    # Pie chart
    ax2.pie(sentiment_dist.values, labels=sentiment_dist.index, autopct='%1.1f%%',
            colors=colors, startangle=90)
    ax2.set_title(f'{coin_name} - Yüzdelik Dağılım', fontweight='bold')
    
    plt.tight_layout()
    
    # Metin rapor
    total = len(filtered)
    report = f"""
📊 **{coin_name.upper()} ANALİZ RAPORU (FinBERT)**
📰 Toplam Haber: {total}
💭 **Sentiment Dağılımı:**
"""
    for label, count in sentiment_dist.items():
        percentage = (count / total) * 100
        report += f"\n• {label.capitalize()}: {count} haber (%{percentage:.1f})"
    
    # En pozitif haberler
    top_positive = filtered.nlargest(3, "sentiment_score")[["title", "sentiment_score"]]
    report += "\n\n⭐ **En Pozitif 3 Haber:**\n"
    for idx, row in top_positive.iterrows():
        report += f"\n• {row['title'][:80]}...\n  Skor: {row['sentiment_score']:.3f}"
    
    # Tablo
    table_data = filtered[["title", "sentiment_label", "sentiment_score", "link"]].head(10)
    
    return report, fig, table_data

def get_trending_topics():
    """Trend analizi"""
    global df
    
    if df is None:
        return "⚠️ Önce haberleri toplamalısınız!", None, None
    
    all_titles = " ".join(df["title"].tolist())
    potential_coins = re.findall(r'\b[A-Z][a-zA-Z]+\b', all_titles)
    
    stop_words = {
        "Coin", "Price", "Market", "News", "Today", "Crypto", "Token",
        "The", "This", "That", "What", "When", "Where", "How", "Why",
        "And", "But", "For", "From", "With", "About", "After", "Before"
    }
    filtered_tokens = [t for t in potential_coins if t not in stop_words]
    trending = Counter(filtered_tokens).most_common(10)
    
    # Grafik
    fig, ax = plt.subplots(figsize=(10, 6))
    terms, counts = zip(*trending)
    ax.barh(terms, counts, color='#3498db', alpha=0.7)
    ax.set_xlabel('Frekans', fontweight='bold')
    ax.set_title('En Çok Bahsedilen 10 Terim', fontweight='bold', fontsize=14)
    ax.invert_yaxis()
    ax.grid(axis='x', alpha=0.3)
    plt.tight_layout()
    
    # Tablo
    trend_df = pd.DataFrame(trending, columns=["Terim", "Frekans"])
    
    message = f"🔥 En popüler terim: **{trending[0][0]}** ({trending[0][1]} kez bahsedildi)"
    
    return message, fig, trend_df

def create_overview_chart():
    """Genel sentiment grafiği"""
    global df
    
    if df is None:
        return "⚠️ Önce haberleri toplamalısınız!"
    
    fig, axes = plt.subplots(2, 2, figsize=(12, 10))
    fig.suptitle('Kripto Haber Analizi - Genel Bakış (FinBERT)', fontsize=16, fontweight='bold')
    
    # Renk haritası
    color_map = {
        'positive': '#2ecc71',
        'negative': '#e74c3c',
        'neutral': '#95a5a6'
    }
    
    # 1. Genel sentiment
    ax1 = axes[0, 0]
    sentiment_counts = df["sentiment_label"].value_counts()
    colors = [color_map.get(x, '#95a5a6') for x in sentiment_counts.index]
    ax1.bar(sentiment_counts.index, sentiment_counts.values, color=colors, alpha=0.7)
    ax1.set_title('Tüm Haberlerde Sentiment', fontweight='bold')
    ax1.set_ylabel('Haber Sayısı')
    ax1.grid(axis='y', alpha=0.3)
    
    # 2. Skor dağılımı
    ax2 = axes[0, 1]
    positive_scores = df[df["sentiment_label"] == "positive"]["sentiment_score"]
    negative_scores = df[df["sentiment_label"] == "negative"]["sentiment_score"]
    neutral_scores = df[df["sentiment_label"] == "neutral"]["sentiment_score"]
    
    ax2.hist([positive_scores, negative_scores, neutral_scores], bins=15, 
             label=['Positive', 'Negative', 'Neutral'],
             color=['#2ecc71', '#e74c3c', '#95a5a6'], alpha=0.6)
    ax2.set_title('Sentiment Skor Dağılımı', fontweight='bold')
    ax2.set_xlabel('Güven Skoru')
    ax2.legend()
    ax2.grid(axis='y', alpha=0.3)
    
    # 3. Kaynak dağılımı
    ax3 = axes[1, 0]
    sources = df["link"].apply(lambda x: x.split('/')[2] if '/' in x else 'Unknown')
    source_counts = sources.value_counts().head(5)
    ax3.barh(source_counts.index, source_counts.values, color='#9b59b6', alpha=0.7)
    ax3.set_title('En Çok Haber Kaynağı', fontweight='bold')
    ax3.set_xlabel('Haber Sayısı')
    ax3.invert_yaxis()
    
    # 4. Pie chart
    ax4 = axes[1, 1]
    ax4.pie(sentiment_counts.values, labels=sentiment_counts.index, autopct='%1.1f%%',
            colors=colors, startangle=90)
    ax4.set_title('Sentiment Yüzdesi', fontweight='bold')
    
    plt.tight_layout()
    return fig

# Gradio arayüzü
with gr.Blocks(theme=gr.themes.Soft(), title="Kripto Haber Analiz Platformu - FinBERT") as app:
    
    gr.Markdown("""
    # 🚀 Kripto Haber Analiz Platformu (FinBERT)
    
    Bu uygulama kripto para haberlerini toplar ve **FinBERT** modeli ile finansal sentiment analizi yapar.
    
    ### 📋 Kullanım Adımları:
    1. **"FinBERT Modelini Başlat"** butonuna tıklayın
    2. **"Haberleri Topla ve Analiz Et"** butonuna tıklayın  
    3. **"Genel Özet"** sekmesinden haberlerin özetini görün
    4. Diğer sekmeleri kullanarak detaylı analiz yapın
    
    ### 🎯 FinBERT Nedir?
    FinBERT, finansal metinler üzerinde eğitilmiş özel bir BERT modelidir. 
    Kripto ve finans haberlerinde daha doğru sentiment analizi sağlar.
    """)
    
    with gr.Tab("🏠 Başlangıç"):
        gr.Markdown("### Sistemi Başlatın")
        
        init_btn = gr.Button("🔧 FinBERT Modelini Başlat (İlk adım)", variant="primary", size="lg")
        init_output = gr.Textbox(label="Durum", lines=2)
        
        gr.Markdown("---")
        
        fetch_btn = gr.Button("📰 Haberleri Topla ve Analiz Et (İkinci adım)", variant="primary", size="lg")
        fetch_output = gr.Textbox(label="Toplama Durumu", lines=10)
        fetch_table = gr.Dataframe(label="İlk 10 Haber")
        
        init_btn.click(initialize_models, outputs=init_output)
        fetch_btn.click(fetch_news, outputs=[fetch_output, fetch_table])
    
    with gr.Tab("📰 Genel Özet"):
        gr.Markdown("### Tüm Haberlerin Genel Özeti")
        gr.Markdown("Toplanan tüm haberlerin sentiment analizi, en önemli haberler ve trend konuları.")
        
        summary_btn = gr.Button("📰 Genel Özet Oluştur", variant="primary", size="lg")
        summary_text = gr.Markdown(label="Özet Rapor")
        summary_chart = gr.Plot(label="Özet Grafikler")
        summary_table = gr.Dataframe(label="İstatistikler")
        
        summary_btn.click(generate_news_summary, outputs=[summary_text, summary_chart, summary_table])
    
    with gr.Tab("🔍 Haber Arama"):
        gr.Markdown("### Semantik Arama ile Benzer Haberleri Bulun")
        
        with gr.Row():
            search_input = gr.Textbox(label="Arama Sorgusu", placeholder="Örn: Bitcoin, Ethereum, NFT...")
            search_k = gr.Slider(minimum=1, maximum=10, value=5, step=1, label="Sonuç Sayısı")
        
        search_btn = gr.Button("🔎 Ara", variant="primary")
        search_status = gr.Textbox(label="Durum")
        search_results = gr.Dataframe(label="Sonuçlar")
        
        search_btn.click(search_similar_news, inputs=[search_input, search_k], 
                        outputs=[search_status, search_results])
    
    with gr.Tab("📊 Coin Analizi"):
        gr.Markdown("### Belirli Bir Coin için Detaylı Sentiment Analizi (FinBERT)")
        
        coin_input = gr.Textbox(label="Coin Adı", placeholder="Örn: Bitcoin, Ethereum, Solana...")
        analyze_btn = gr.Button("📊 Analiz Et", variant="primary")
        
        coin_report = gr.Markdown(label="Rapor")
        coin_chart = gr.Plot(label="Grafikler")
        coin_table = gr.Dataframe(label="Detaylı Haberler")
        
        analyze_btn.click(analyze_coin_sentiment, inputs=coin_input,
                         outputs=[coin_report, coin_chart, coin_table])
    
    with gr.Tab("🔥 Trend Analizi"):
        gr.Markdown("### En Çok Bahsedilen Konular")
        
        trend_btn = gr.Button("🔥 Trendleri Göster", variant="primary")
        trend_status = gr.Markdown(label="Özet")
        trend_chart = gr.Plot(label="Trend Grafiği")
        trend_table = gr.Dataframe(label="Trend Tablosu")
        
        trend_btn.click(get_trending_topics, outputs=[trend_status, trend_chart, trend_table])
    
    with gr.Tab("📈 Genel Bakış"):
        gr.Markdown("### Tüm Verilerin Genel Görünümü")
        
        overview_btn = gr.Button("📈 Genel İstatistikleri Göster", variant="primary")
        overview_chart = gr.Plot(label="Genel Grafikler")
        
        overview_btn.click(create_overview_chart, outputs=overview_chart)
    
    gr.Markdown("""
    ---
    ### 💡 İpuçları:
    - FinBERT ilk çalıştırmada yüklendiği için biraz zaman alabilir
    - Haberler RSS feedlerinden gerçek zamanlı olarak çekilir
    - **Genel Özet** sekmesi tüm haberlerin analizini sunar
    - FinBERT finansal metinler için özelleştirilmiştir
    - FAISS kullanarak semantik benzerlik hesaplanır
    
    ### 🛠️ Kullanılan Teknolojiler:
    - **FinBERT (ProsusAI/finbert)**: Finansal sentiment analizi
    - **Sentence Transformers**: Semantik embedding
    - **FAISS**: Hızlı benzerlik araması
    - **Gradio**: Kullanıcı arayüzü
    
    ### 📊 FinBERT Çıktıları:
    - **Positive**: Pozitif finansal sentiment
    - **Negative**: Negatif finansal sentiment
    - **Neutral**: Nötr finansal sentiment
    """)

# Uygulamayı başlat
if __name__ == "__main__":
    app.launch()