chat/comparator.py

# chat/comparator.py
"""

Advanced cross-paper comparison and methodology analysis

Compares techniques, results, and approaches across multiple studies

"""

from typing import List, Dict, Any, Tuple
from llm.llm_provider import GrokLLM
from llm.prompt_templates import MedicalResearchPrompts, ResponseFormatter
import re


class CrossPaperComparator:
    """

    Advanced comparator that analyzes and compares research across multiple papers

    Focuses on methodologies, results, and performance metrics

    """

    def __init__(self, llm=None):
        self.llm = llm or GrokLLM(model="model")  # Use shared LLM
        self.prompts = MedicalResearchPrompts()
        self.formatter = ResponseFormatter()

    def compare_methods(self, papers: List[Dict], method1: str, method2: str, domain: str) -> Dict[str, Any]:
        """

        Compare two methods across multiple research papers

        """
        print(f"🔍 Comparing {method1} vs {method2} across {len(papers)} papers")

        # Filter papers that mention either method
        relevant_papers = self._filter_relevant_papers(papers, [method1, method2])

        if len(relevant_papers) < 2:
            return self._create_insufficient_data_response(method1, method2, domain, len(relevant_papers))

        try:
            # Generate detailed comparison
            comparison_prompt = self.prompts.cross_paper_comparison(relevant_papers, method1, method2, domain)

            response = self.llm.generate(
                comparison_prompt,
                system_message=self.prompts.SYSTEM_MESSAGES["methodology_expert"],
                temperature=0.1,
                max_tokens=2000
            )

            # Extract performance metrics and structured data
            performance_data = self._extract_performance_metrics(relevant_papers, method1, method2)
            trend_analysis = self._analyze_trends(relevant_papers, method1, method2)

            return {
                "detailed_comparison": response,
                "performance_metrics": performance_data,
                "trend_analysis": trend_analysis,
                "method1": method1,
                "method2": method2,
                "domain": domain,
                "papers_analyzed": len(relevant_papers),
                "relevant_papers": [self.formatter.format_citation(paper, i + 1) for i, paper in
                                    enumerate(relevant_papers)]
            }

        except Exception as e:
            print(f"❌ Comparison error: {e}")
            return self._create_fallback_comparison(relevant_papers, method1, method2, domain)

    def _filter_relevant_papers(self, papers: List[Dict], methods: List[str]) -> List[Dict]:
        """Filter papers that mention the methods of interest"""
        relevant_papers = []

        for paper in papers:
            abstract = paper.get('abstract', '').lower()
            title = paper.get('title', '').lower()

            # Check if paper mentions any of the methods
            for method in methods:
                method_terms = self._get_method_variations(method)
                if any(term in abstract or term in title for term in method_terms):
                    relevant_papers.append(paper)
                    break

        return relevant_papers

    def _get_method_variations(self, method: str) -> List[str]:
        """Get common variations of method names"""
        method = method.lower()
        variations = [method]

        # Common method variations
        method_variations = {
            'cnn': ['cnn', 'convolutional neural network', 'convolutional network'],
            'transformer': ['transformer', 'attention mechanism', 'self-attention'],
            'random forest': ['random forest', 'rf classifier'],
            'svm': ['svm', 'support vector machine'],
            'knn': ['knn', 'k-nearest neighbor'],
            'logistic regression': ['logistic regression', 'logit model']
        }

        if method in method_variations:
            variations.extend(method_variations[method])

        return variations

    def _extract_performance_metrics(self, papers: List[Dict], method1: str, method2: str) -> Dict[str, Any]:
        """Extract performance metrics from papers"""
        metrics = {
            method1: {"papers_count": 0, "performance_mentions": []},
            method2: {"papers_count": 0, "performance_mentions": []},
            "comparison_mentions": []
        }

        performance_patterns = [
            r'(\d+\.?\d*)%',  # Percentage metrics
            r'accuracy of (\d+\.?\d*)',
            r'precision of (\d+\.?\d*)',
            r'recall of (\d+\.?\d*)',
            r'f1 score of (\d+\.?\d*)',
            r'auc of (\d+\.?\d*)'
        ]

        for paper in papers:
            abstract = paper.get('abstract', '').lower()
            title = paper.get('title', '').lower()

            # Count method mentions
            method1_terms = self._get_method_variations(method1)
            method2_terms = self._get_method_variations(method2)

            method1_mentioned = any(term in abstract or term in title for term in method1_terms)
            method2_mentioned = any(term in abstract or term in title for term in method2_terms)

            if method1_mentioned:
                metrics[method1]["papers_count"] += 1
            if method2_mentioned:
                metrics[method2]["papers_count"] += 1

            # Extract performance numbers
            for pattern in performance_patterns:
                matches = re.findall(pattern, abstract)
                for match in matches:
                    try:
                        value = float(match)
                        if method1_mentioned:
                            metrics[method1]["performance_mentions"].append(value)
                        if method2_mentioned:
                            metrics[method2]["performance_mentions"].append(value)
                    except ValueError:
                        continue

            # Look for direct comparisons
            comparison_terms = ['compared to', 'versus', 'vs', 'outperform', 'better than', 'worse than']
            if any(term in abstract for term in comparison_terms) and (method1_mentioned or method2_mentioned):
                metrics["comparison_mentions"].append(paper.get('title', 'Unknown'))

        # Calculate average performance if we have data
        for method in [method1, method2]:
            mentions = metrics[method]["performance_mentions"]
            if mentions:
                metrics[method]["average_performance"] = sum(mentions) / len(mentions)
                metrics[method]["max_performance"] = max(mentions)
                metrics[method]["min_performance"] = min(mentions)

        return metrics

    def _analyze_trends(self, papers: List[Dict], method1: str, method2: str) -> Dict[str, Any]:
        """Analyze publication trends for each method"""
        trends = {
            method1: {"recent_papers": 0, "total_papers": 0},
            method2: {"recent_papers": 0, "total_papers": 0},
            "trend_direction": "neutral"
        }

        for paper in papers:
            # Simple date-based trend analysis
            date = paper.get('publication_date', '')
            abstract = paper.get('abstract', '').lower()

            method1_mentioned = any(term in abstract for term in self._get_method_variations(method1))
            method2_mentioned = any(term in abstract for term in self._get_method_variations(method2))

            if method1_mentioned:
                trends[method1]["total_papers"] += 1
                if '2024' in date or '2023' in date:
                    trends[method1]["recent_papers"] += 1

            if method2_mentioned:
                trends[method2]["total_papers"] += 1
                if '2024' in date or '2023' in date:
                    trends[method2]["recent_papers"] += 1

        # Determine trend direction
        method1_recent_ratio = trends[method1]["recent_papers"] / max(1, trends[method1]["total_papers"])
        method2_recent_ratio = trends[method2]["recent_papers"] / max(1, trends[method2]["total_papers"])

        if method1_recent_ratio > method2_recent_ratio + 0.2:
            trends["trend_direction"] = f"{method1} gaining"
        elif method2_recent_ratio > method1_recent_ratio + 0.2:
            trends["trend_direction"] = f"{method2} gaining"
        else:
            trends["trend_direction"] = "both stable"

        return trends

    def _create_insufficient_data_response(self, method1: str, method2: str, domain: str, relevant_count: int) -> Dict[
        str, Any]:
        """Create response when insufficient data is available"""
        return {
            "detailed_comparison": f"Insufficient data for comparison. Only {relevant_count} papers mention {method1} or {method2} in the {domain} domain.",
            "performance_metrics": {},
            "trend_analysis": {},
            "method1": method1,
            "method2": method2,
            "domain": domain,
            "papers_analyzed": relevant_count,
            "relevant_papers": [],
            "insufficient_data": True
        }

    def _create_fallback_comparison(self, papers: List[Dict], method1: str, method2: str, domain: str) -> Dict[
        str, Any]:
        """Create basic comparison when LLM fails"""
        performance_data = self._extract_performance_metrics(papers, method1, method2)
        trend_analysis = self._analyze_trends(papers, method1, method2)

        basic_comparison = f"""

        Basic Comparison: {method1} vs {method2} in {domain}



        Papers Analyzed: {len(papers)}

        {method1} mentioned in: {performance_data[method1]['papers_count']} papers

        {method2} mentioned in: {performance_data[method2]['papers_count']} papers



        Trend: {trend_analysis['trend_direction']}



        Note: Detailed AI comparison unavailable. Consider refining your search terms.

        """

        return {
            "detailed_comparison": basic_comparison,
            "performance_metrics": performance_data,
            "trend_analysis": trend_analysis,
            "method1": method1,
            "method2": method2,
            "domain": domain,
            "papers_analyzed": len(papers),
            "relevant_papers": [self.formatter.format_citation(paper, i + 1) for i, paper in enumerate(papers)],
            "fallback_used": True
        }

    def generate_comparison_table(self, comparison_data: Dict[str, Any]) -> str:
        """Generate a structured comparison table"""
        method1 = comparison_data['method1']
        method2 = comparison_data['method2']
        metrics = comparison_data['performance_metrics']
        trends = comparison_data['trend_analysis']

        table = f"**Comparison: {method1} vs {method2}**\n\n"
        table += "| Metric | {method1} | {method2} |\n".format(method1=method1, method2=method2)
        table += "|--------|-----------|-----------|\n"

        # Papers count
        table += "| Papers Mentioned | {count1} | {count2} |\n".format(
            count1=metrics[method1]['papers_count'],
            count2=metrics[method2]['papers_count']
        )

        # Recent papers
        table += "| Recent Papers (2023-2024) | {recent1} | {recent2} |\n".format(
            recent1=trends[method1]['recent_papers'],
            recent2=trends[method2]['recent_papers']
        )

        # Average performance if available
        if 'average_performance' in metrics[method1] and 'average_performance' in metrics[method2]:
            table += "| Avg Performance | {avg1:.1f}% | {avg2:.1f}% |\n".format(
                avg1=metrics[method1]['average_performance'],
                avg2=metrics[method2]['average_performance']
            )

        table += f"\n**Trend:** {trends['trend_direction']}\n"

        return table


# Quick test
def test_comparator():
    """Test the cross-paper comparator"""
    print("🧪 Testing Cross-Paper Comparator")
    print("=" * 50)

    test_papers = [
        {
            'title': 'CNN vs Transformer for Medical Images',
            'authors': ['Smith J', 'Lee K'],
            'abstract': 'We compare CNN and Transformer architectures for medical image classification. CNNs achieve 92% accuracy while Transformers reach 94% but require more data.',
            'source': 'Medical Image Analysis',
            'domain': 'medical_imaging',
            'publication_date': '2024-01-15'
        },
        {
            'title': 'Efficient Transformers in Radiology',
            'authors': ['Chen R', 'Wang L'],
            'abstract': 'This paper introduces efficient transformer variants for radiology applications. Our method maintains 93% accuracy with 50% fewer parameters compared to standard transformers.',
            'source': 'IEEE Transactions',
            'domain': 'medical_imaging',
            'publication_date': '2024-03-10'
        }
    ]

    comparator = CrossPaperComparator()

    try:
        comparison = comparator.compare_methods(
            test_papers,
            "CNN",
            "Transformer",
            "medical_imaging"
        )

        print(f"✅ Comparison generated successfully")
        print(f"📊 Papers analyzed: {comparison['papers_analyzed']}")
        print(f"📈 Trend: {comparison['trend_analysis']['trend_direction']}")

        table = comparator.generate_comparison_table(comparison)
        print(f"\n📋 Comparison Table:\n{table}")

    except Exception as e:
        print(f"❌ Comparison test failed: {e}")


if __name__ == "__main__":
    test_comparator()