modules/explanation.py

import logging
import re
import ast
from utils.models import get_llm_model

logger = logging.getLogger("misinformation_detector")

def extract_most_relevant_evidence(evidence_results):
    """
    Intelligently extract the most relevant piece of evidence
    
    Args:
        evidence_results (list): List of evidence items
    
    Returns:
        str: Most relevant evidence piece
    """
    if not evidence_results:
        return None

    # If evidence is a dictionary with 'evidence' key
    if isinstance(evidence_results[0], dict):
        # Sort by confidence if available
        sorted_evidence = sorted(
            evidence_results, 
            key=lambda x: x.get('confidence', 0), 
            reverse=True
        )
        
        # Return the evidence from the highest confidence item
        for item in sorted_evidence:
            evidence = item.get('evidence')
            if evidence:
                return evidence

    # If plain list of evidence
    return next((ev for ev in evidence_results if ev and isinstance(ev, str)), None)

def generate_explanation(claim, evidence_results, truth_label, confidence=None):
    """
    Generate an explanation for the claim's classification
    
    Args:
        claim (str): The original claim
        evidence_results (list/str): Evidence supporting the classification
        truth_label (str): Classification of the claim
        confidence (float): Confidence level (0-1)
    
    Returns:
        str: Explanation of the claim's classification
    """
    logger.info(f"Generating explanation for claim with verdict: {truth_label}")
    
    try:
        # Normalize evidence_results to a list
        if not isinstance(evidence_results, list):
            try:
                evidence_results = ast.literal_eval(str(evidence_results)) if evidence_results else []
            except:
                evidence_results = [evidence_results] if evidence_results else []

        # Get the LLM model
        explanation_model = get_llm_model()

        # Extract most relevant evidence
        most_relevant_evidence = extract_most_relevant_evidence(evidence_results)

        # Prepare evidence text for prompt
        evidence_text = "\n".join([
            f"Evidence {i+1}: {str(ev)[:200] + '...' if len(str(ev)) > 200 else str(ev)}" 
            for i, ev in enumerate(evidence_results[:3])
        ])

        # Convert confidence to percentage and description
        confidence_desc = ""
        if confidence is not None:
            confidence_pct = int(confidence * 100)
            if confidence < 0.3:
                confidence_desc = f"very low confidence ({confidence_pct}%)"
            elif confidence < 0.5:
                confidence_desc = f"low confidence ({confidence_pct}%)"
            elif confidence < 0.7:
                confidence_desc = f"moderate confidence ({confidence_pct}%)"
            elif confidence < 0.9:
                confidence_desc = f"high confidence ({confidence_pct}%)"
            else:
                confidence_desc = f"very high confidence ({confidence_pct}%)"
        else:
            # Determine confidence context from label if not explicitly provided
            confidence_desc = (
                "high confidence" if "High Confidence" in truth_label else
                "moderate confidence" if "Likely" in truth_label else
                "low confidence"
            )

        # Create prompt with specific instructions based on the type of claim
        has_negation = any(neg in claim.lower() for neg in ["not", "no longer", "isn't", "doesn't", "won't", "cannot"])
        
        # For claims with "True" verdict
        if "True" in truth_label:
            prompt = f"""
            Claim: "{claim}"
            
            Verdict: {truth_label} (with {confidence_desc})

            Available Evidence:
            {evidence_text}

            Task: Generate a clear explanation that:
            1. Clearly states that the claim IS TRUE based on the evidence
            2. {"Pay special attention to the logical relationship since the claim contains negation" if has_negation else "Explains why the evidence supports the claim"}
            3. Uses confidence level of {confidence_desc}
            4. Highlights the most relevant supporting evidence
            5. Is factual and precise
            """

        # For claims with "False" verdict
        elif "False" in truth_label:
            prompt = f"""
            Claim: "{claim}"
            
            Verdict: {truth_label} (with {confidence_desc})

            Available Evidence:
            {evidence_text}

            Task: Generate a clear explanation that:
            1. Clearly states that the claim IS FALSE based on the evidence
            2. {"Pay special attention to the logical relationship since the claim contains negation" if has_negation else "Explains why the evidence contradicts the claim"}
            3. Uses confidence level of {confidence_desc}
            4. Highlights the contradicting evidence
            5. Is factual and precise
            
            IMPORTANT: If the claim contains negation (words like 'not', 'no longer', etc.), be extra careful with the logical relationship between the evidence and the claim.
            """

        # For uncertain claims
        else:
            prompt = f"""
            Claim: "{claim}"
            
            Verdict: {truth_label} (with {confidence_desc})

            Available Evidence:
            {evidence_text}

            Task: Generate a clear explanation that:
            1. Clearly states that there is insufficient evidence to determine if the claim is true or false
            2. Explains what information is missing or why the available evidence is insufficient
            3. Uses confidence level of {confidence_desc}
            4. Makes NO speculation about whether the claim might be true or false
            5. Mentions that the user should seek information from other reliable sources
            """

        # Generate explanation with multiple attempts
        max_attempts = 3
        for attempt in range(max_attempts):
            try:
                # Invoke the model
                response = explanation_model.invoke(prompt)
                explanation = response.content.strip()

                # Validate explanation length
                if explanation and len(explanation.split()) >= 5:
                    return explanation

            except Exception as attempt_error:
                logger.error(f"Explanation generation attempt {attempt+1} failed: {str(attempt_error)}")

        # Ultimate fallback explanation
        if "Uncertain" in truth_label:
            return f"The claim '{claim}' cannot be verified due to insufficient evidence. The available information does not provide clear support for or against this claim. Consider consulting reliable sources for verification."
        elif "True" in truth_label:
            return f"The claim '{claim}' is supported by the evidence with {confidence_desc}. {most_relevant_evidence or 'The evidence indicates this claim is accurate.'}"
        else:
            return f"The claim '{claim}' is contradicted by the evidence with {confidence_desc}. {most_relevant_evidence or 'The evidence indicates this claim is not accurate.'}"

    except Exception as e:
        logger.error(f"Comprehensive error in explanation generation: {str(e)}")
        # Final fallback
        return f"The claim is classified as {truth_label} based on the available evidence."