app.py

import json
import os
import gradio as gr
import numpy
from transformers import pipeline, AutoTokenizer, AutoModelForSequenceClassification

# ==============================
# MODELLO ITALIAN SENTIMENT
# ==============================
print("🔄 Caricamento modello...")
tokenizer = AutoTokenizer.from_pretrained(
    "MilaNLProc/feel-it-italian-sentiment",
    use_fast=False
)

model = AutoModelForSequenceClassification.from_pretrained(
    "MilaNLProc/feel-it-italian-sentiment"
)

sentiment_pipeline = pipeline(
    "text-classification",
    model=model,
    tokenizer=tokenizer,
    truncation=True,
    max_length=512
)
print("✅ Modello caricato!")

# ==============================
# CHUNKING
# ==============================
def chunk_text(text, max_chars=400):
    if not text or not text.strip():
        return []
    
    sentences = text.replace("\n", " ").split(". ")
    chunks, current = [], ""
    
    for s in sentences:
        if len(current) + len(s) <= max_chars:
            current += s + ". "
        else:
            chunks.append(current.strip())
            current = s + ". "
    
    if current.strip():
        chunks.append(current.strip())
    
    return chunks

# ==============================
# ANALISI SENTIMENT PER ARTICOLO
# ==============================
def analyze_article_sentiment(article):
    text = article.get("content", "")
    
    if not text.strip():
        article["sentimentAnalysis"] = "UNKNOWN"
        article["sentimentScore"] = None
        article["chunksAnalyzed"] = 0
        return article
    
    chunks = chunk_text(text)
    
    if not chunks:
        article["sentimentAnalysis"] = "UNKNOWN"
        article["sentimentScore"] = None
        article["chunksAnalyzed"] = 0
        return article
    
    results = sentiment_pipeline(chunks)
    
    scores = []
    for r in results:
        label = r["label"].lower()
        
        if label == "negative":
            scores.append(0.0)
        elif label == "neutral":
            scores.append(0.5)
        elif label == "positive":
            scores.append(1.0)
    
    if not scores:
        article["sentimentAnalysis"] = "UNKNOWN"
        article["sentimentScore"] = None
        article["chunksAnalyzed"] = len(chunks)
        return article
    
    avg_score = sum(scores) / len(scores)
    
    if avg_score <= 0.4:
        sentiment = "NEGATIVE"
    elif avg_score < 0.66:
        sentiment = "NEUTRAL"
    else:
        sentiment = "POSITIVE"
    
    article["sentimentAnalysis"] = sentiment
    article["sentimentScore"] = round(avg_score, 3)
    article["chunksAnalyzed"] = len(chunks)
    
    return article

# ==============================
# FUNZIONE PER GRADIO
# ==============================
def process_input_json():
    """
    Legge input.json dal repository, processa gli articoli e restituisce output.json
    """
    input_file = "input.json"
    output_file = "/tmp/output.json"
    
    # Verifica esistenza file
    if not os.path.exists(input_file):
        available_files = os.listdir()
        error_msg = f"❌ File {input_file} non trovato!\n\n📂 File disponibili nella directory:\n"
        error_msg += "\n".join(f"  - {f}" for f in available_files)
        return None, error_msg
    
    try:
        # Leggi input
        with open(input_file, 'r', encoding='utf-8') as f:
            articles = json.load(f)
        
        if not isinstance(articles, list):
            return None, "❌ Errore: il file JSON deve contenere un array di articoli"
        
        log = f"📊 Trovati {len(articles)} articoli da analizzare\n\n"
        
        processed_articles = []
        
        for i, article in enumerate(articles, 1):
            title = article.get('title', 'N/A')[:60]
            log += f"⏳ [{i}/{len(articles)}] {title}...\n"
            
            processed_article = analyze_article_sentiment(article)
            processed_articles.append(processed_article)
            
            log += f"   ✅ {processed_article['sentimentAnalysis']} "
            log += f"(score: {processed_article['sentimentScore']}, "
            log += f"chunks: {processed_article['chunksAnalyzed']})\n\n"
        
        # Scrivi output
        with open(output_file, 'w', encoding='utf-8') as f:
            json.dump(processed_articles, f, ensure_ascii=False, indent=2)
        
        log += f"\n✅ COMPLETATO! {len(processed_articles)} articoli processati.\n"
        log += f"💾 Output salvato e pronto per il download"
        
        return output_file, log
        
    except Exception as e:
        import traceback
        error_msg = f"❌ Errore durante l'elaborazione:\n{str(e)}\n\n"
        error_msg += f"Traceback:\n{traceback.format_exc()}"
        return None, error_msg


def clear_outputs():
    """
    Pulisce l'output file e il log
    """
    return None, "🧹 Interfaccia pulita! Puoi modificare input.json e riprocessare."

# ==============================
# INTERFACCIA GRADIO
# ==============================
with gr.Blocks(title="Italian Sentiment Analyzer", theme=gr.themes.Soft()) as demo:
    gr.Markdown("# 🧠 Italian Sentiment Analyzer")
    gr.Markdown("Analizza il sentiment di articoli italiani usando il modello Feel-IT")
    gr.Markdown("### 📋 Istruzioni:")
    gr.Markdown("1. Assicurati che `input.json` sia presente nel repository")
    gr.Markdown("2. Clicca il pulsante per avviare l'analisi")
    gr.Markdown("3. Scarica `output.json` con i risultati")
    
    with gr.Row():
        process_btn = gr.Button("🚀 Processa input.json", variant="primary", size="lg", scale=2)
        clear_btn = gr.Button("🧹 Clear", variant="secondary", size="lg", scale=1)
        
    with gr.Row():
        with gr.Column(scale=1):
            output_file = gr.File(label="💾 Scarica output.json")
        
        with gr.Column(scale=2):
            log_box = gr.Textbox(
                label="📋 Log elaborazione", 
                lines=20, 
                max_lines=30
            )
    
    process_btn.click(
        fn=process_input_json,
        inputs=None,
        outputs=[output_file, log_box]
    )
    clear_btn.click(
        fn=clear_outputs,
        inputs=None,
        outputs=[output_file, log_box]
    )
    
    gr.Markdown("---")
    gr.Markdown("### 📊 Logica di classificazione:")
    gr.Markdown("- **NEGATIVE** 🔴: score ≤ 0.4")
    gr.Markdown("- **NEUTRAL** 🟡: 0.4 < score < 0.66")
    gr.Markdown("- **POSITIVE** 🟢: score ≥ 0.66")
    
    gr.Markdown("### 🎯 Campi aggiunti nell'output:")
    gr.Markdown("- **sentimentAnalysis**: POSITIVE / NEUTRAL / NEGATIVE / UNKNOWN")
    gr.Markdown("- **sentimentScore**: valore numerico da 0.0 (negativo) a 1.0 (positivo)")
    gr.Markdown("- **chunksAnalyzed**: numero di chunks in cui è stato suddiviso il testo")
    
    

# ==============================
# LAUNCH
# ==============================
if __name__ == "__main__":
    demo.launch(server_name="0.0.0.0", server_port=7860)