sentiment_analyzer.py

from transformers import pipeline
import torch
import re

class NewsAnalyzer:
    def __init__(self, model_name=None):
        """
        Initialize news analyzer with fast, CPU-friendly Zero-Shot pipelines
        """
        print("Initializing Zero-Shot News Analyzer...")
        
        self.device = 0 if torch.cuda.is_available() else -1
        print(f"Using device: {'cuda' if self.device == 0 else 'cpu'}")

        try:
            # Pipeline 1: For Sentiment Analysis
            print("Loading Sentiment model...")
            self.sentiment_pipeline = pipeline(
                "sentiment-analysis",
                model="distilbert-base-uncased-finetuned-sst-2-english",
                device=self.device
            )

            # Pipeline 2: For Zero-Shot Classification (Theme & Impact)
            print("Loading Zero-Shot model...")
            # --- MODIFIED: Corrected model name ---
            self.classifier_pipeline = pipeline(
                "zero-shot-classification",
                model="facebook/bart-large-mnli", # นี่คือโมเดลมาตรฐานที่ถูกต้อง
                device=self.device
            )
            # --- End of modification ---
            
            print("Models loaded successfully!")

            # Define the labels for classification
            self.theme_labels = [
                "Earnings/Finance", "Product/Service", "Legal/Regulation", 
                "Management/M&A", "Market/Economy", "Other"
            ]
            self.impact_labels = ["Opportunity", "Risk", "Neutral"]

        except Exception as e:
            print(f"Fatal error loading models: {e}")
            self.sentiment_pipeline = None
            self.classifier_pipeline = None

    def analyze_news_item(self, text):
        """
        วิเคราะห์ข่าว (Sentiment, Theme, Impact) โดยใช้ Zero-Shot
        """
        # ตรวจสอบว่า pipeline โหลดสำเร็จหรือไม่
        if not self.classifier_pipeline or not self.sentiment_pipeline:
            print("Error: Pipelines are not loaded.")
            return {
                "sentiment": "Neutral", "score": 0.5, "theme": "N/A",
                "impact": "N/A", "explanation": "Model loading failed"
            }

        if not text or len(text.strip()) == 0:
            return {
                "sentiment": "Neutral", "score": 0.5, "theme": "N/A",
                "impact": "N/A", "explanation": "No text"
            }

        try:
            # 1. Analyze Sentiment
            sentiment_result = self.sentiment_pipeline(text[:512])[0]
            sentiment = sentiment_result['label'].capitalize()
            score = sentiment_result['score']
            
            # 2. Analyze Theme
            theme_result = self.classifier_pipeline(
                text[:512], 
                candidate_labels=self.theme_labels
            )
            theme = theme_result['labels'][0]

            # 3. Analyze Impact
            impact_result = self.classifier_pipeline(
                text[:512],
                candidate_labels=self.impact_labels
            )
            impact = impact_result['labels'][0]

            # 4. Create an explanation
            explanation = f"Classified as '{theme}' (Impact: {impact}) via zero-shot analysis."

            # แปลง Sentiment จาก 'Positive'/'Negative' ของ distilbert
            # เป็น 'Positive'/'Negative'/'Neutral' (แต่โมเดลนี้ไม่มี Neutral)
            final_sentiment = "Positive" if sentiment == "Positive" else "Negative"

            return {
                "sentiment": final_sentiment,
                "score": score,
                "theme": theme,
                "impact": impact,
                "explanation": explanation
            }

        except Exception as e:
            print(f"Error in analysis: {e}")
            return {
                "sentiment": "Neutral", "score": 0.5, "theme": "N/A",
                "impact": "N/A", "explanation": "Analysis failed"
            }

    def analyze_batch(self, news_list):
        """
        วิเคราะห์ sentiment หลายข่าวพร้อมกัน
        """
        results = []
        for news in news_list:
            combined_text = f"{news.get('title', '')} {news.get('summary', '')}"
            sentiment_result = self.analyze_news_item(combined_text)
            results.append({
                **news,
                **sentiment_result
            })
        return results