tasq/node_modules/agentdb/simulation/scenarios/domain-examples/scientific-research.ts

/**
 * Scientific Research: Research Paper Similarity and Discovery
 *
 * Use Case: Find similar research papers, identify research trends,
 * and discover cross-domain connections.
 *
 * Optimization Priority: QUALITY + CROSS-DOMAIN DISCOVERY
 */

import { UnifiedMetrics } from '../../types';

export const RESEARCH_ATTENTION_CONFIG = {
  heads: 12,                       // High-quality attention (more than 8-head optimal)
  forwardPassTargetMs: 100,        // 100ms acceptable (research workflow)
  batchSize: 128,                  // Large batches for corpus processing
  precision: 'float32' as const,   // Full precision for scientific data
  crossDomainDiscovery: true,      // Enable cross-field connections

  // Hierarchical clustering for research topics
  clustering: {
    algorithm: 'hierarchical' as const, // Better for taxonomy
    linkage: 'ward' as const,      // Ward's minimum variance
    hierarchicalLevels: 5,         // Deep taxonomy (field > subfield > topic > subtopic > specific)
    semanticPurity: 0.90           // High purity for research domains
  },

  // High-recall for comprehensive literature review
  recallTarget: 0.98,              // 98% recall (don't miss relevant papers)

  // Self-organizing for evolving research landscape
  selfHealing: {
    enabled: true,
    adaptationIntervalMs: 60000,   // 1-minute adaptation (not real-time)
    degradationThreshold: 0.03,    // 3% tolerance
    taxonomyUpdates: true          // Update research taxonomy
  }
};

// Research-specific metrics
export interface ResearchMetrics extends UnifiedMetrics {
  crossDomainConnections: number;  // Cross-field discoveries
  citationAccuracy: number;        // Citation network accuracy
  taxonomyQuality: number;         // Research taxonomy quality
  noveltyScore: number;            // Novel connection discovery rate
  expertAgreement: number;         // Agreement with domain experts
}

// Research connection interface
export interface ResearchConnection {
  paperId: string;
  title: string;
  authors: string[];
  similarity: number;
  domain: string;
  crossDomainConnection?: {
    targetDomain: string;
    connectionType: string;
    noveltyScore: number;
  };
  citations: number;
}

// Example: Research paper discovery with cross-domain connections
export async function discoverRelatedResearch(
  paperEmbedding: Float32Array,    // Paper abstract + citations embeddings
  researchCorpus: any,             // HNSWGraph type
  applyAttention: (data: Float32Array, config: any) => Promise<Float32Array>,
  buildResearchTaxonomy: (papers: any[], config: any) => Promise<any>,
  findCrossDomainConnections: (papers: any[], taxonomy: any) => Promise<any[]>,
  findDomain: (paper: any, taxonomy: any) => string,
  includeCrossDomain: boolean = true
): Promise<ResearchConnection[]> {
  const config = RESEARCH_ATTENTION_CONFIG;

  // 12-head attention for nuanced understanding
  const enhanced = await applyAttention(paperEmbedding, config);

  // High-recall search (k=100 for comprehensive review)
  const candidates = await researchCorpus.search(enhanced, 100);

  // Hierarchical clustering for taxonomy
  const taxonomy = await buildResearchTaxonomy(candidates, config.clustering);

  // Cross-domain discovery (find connections across fields)
  const crossDomain = includeCrossDomain
    ? await findCrossDomainConnections(candidates, taxonomy)
    : [];

  return candidates.map((p: any) => ({
    paperId: p.id,
    title: p.metadata.title,
    authors: p.metadata.authors,
    similarity: p.score,
    domain: findDomain(p, taxonomy),
    crossDomainConnection: crossDomain.find((c: any) => c.paperId === p.id),
    citations: p.metadata.citations
  }));
}

// Performance targets for research
export const RESEARCH_PERFORMANCE_TARGETS = {
  recallAt100: 0.98,               // 98% recall (comprehensive)
  p95LatencyMs: 200,               // 200ms p95 (batch processing)
  crossDomainRate: 0.15,           // 15% cross-domain discoveries
  expertAgreement: 0.85,           // 85% agreement with experts
  uptimePercent: 99.0              // 99% uptime (2 nines)
};

// Research field-specific configurations
export const RESEARCH_CONFIG_VARIATIONS = {
  // Computer science (fast-moving field, many papers)
  computerScience: {
    ...RESEARCH_ATTENTION_CONFIG,
    heads: 10,
    forwardPassTargetMs: 50,       // Faster for large corpus
    batchSize: 256,
    recallTarget: 0.97             // Slightly lower for speed
  },

  // Medicine (high precision required, slower pace)
  medicine: {
    ...RESEARCH_ATTENTION_CONFIG,
    heads: 16,                     // Maximum quality
    forwardPassTargetMs: 150,
    recallTarget: 0.99,            // Highest recall
    precision: 'float32' as const
  },

  // Physics (mathematical precision, cross-domain links)
  physics: {
    ...RESEARCH_ATTENTION_CONFIG,
    heads: 14,
    crossDomainDiscovery: true,
    clustering: {
      ...RESEARCH_ATTENTION_CONFIG.clustering,
      hierarchicalLevels: 6        // Deeper for specialized subfields
    }
  },

  // Social sciences (qualitative, broader connections)
  socialSciences: {
    ...RESEARCH_ATTENTION_CONFIG,
    heads: 12,
    crossDomainDiscovery: true,
    clustering: {
      ...RESEARCH_ATTENTION_CONFIG.clustering,
      semanticPurity: 0.85         // More flexible clustering
    }
  }
};

// Search mode adaptations
export function adaptConfigToSearchMode(
  baseConfig: typeof RESEARCH_ATTENTION_CONFIG,
  mode: 'literature-review' | 'novelty-discovery' | 'citation-tracing' | 'interdisciplinary'
): typeof RESEARCH_ATTENTION_CONFIG {
  switch (mode) {
    case 'literature-review':
      return {
        ...baseConfig,
        heads: 12,
        recallTarget: 0.99,        // Comprehensive recall
        crossDomainDiscovery: false // Stay within field
      };
    case 'novelty-discovery':
      return {
        ...baseConfig,
        heads: 16,                 // Maximum quality
        crossDomainDiscovery: true,
        clustering: {
          ...baseConfig.clustering,
          semanticPurity: 0.80     // More flexible for novel connections
        }
      };
    case 'citation-tracing':
      return {
        ...baseConfig,
        heads: 10,
        forwardPassTargetMs: 50,
        recallTarget: 0.95,        // Fast, focused
        crossDomainDiscovery: false
      };
    case 'interdisciplinary':
      return {
        ...baseConfig,
        heads: 14,
        crossDomainDiscovery: true,
        clustering: {
          ...baseConfig.clustering,
          hierarchicalLevels: 6,   // Deep taxonomy for connections
          semanticPurity: 0.82     // Flexible for cross-field
        }
      };
  }
}

// Research stage adaptations
export function adaptConfigToResearchStage(
  baseConfig: typeof RESEARCH_ATTENTION_CONFIG,
  stage: 'initial-exploration' | 'hypothesis-formation' | 'validation' | 'writing'
): typeof RESEARCH_ATTENTION_CONFIG {
  switch (stage) {
    case 'initial-exploration':
      return {
        ...baseConfig,
        heads: 12,
        crossDomainDiscovery: true,
        recallTarget: 0.95         // Broad search
      };
    case 'hypothesis-formation':
      return {
        ...baseConfig,
        heads: 14,
        crossDomainDiscovery: true,
        recallTarget: 0.98         // Comprehensive
      };
    case 'validation':
      return {
        ...baseConfig,
        heads: 16,                 // Maximum precision
        recallTarget: 0.99,
        crossDomainDiscovery: false // Focused
      };
    case 'writing':
      return {
        ...baseConfig,
        heads: 10,
        forwardPassTargetMs: 50,   // Fast lookups
        recallTarget: 0.95
      };
  }
}

// Citation network analysis
export interface CitationNetworkMetrics {
  networkDensity: number;
  clusteringCoefficient: number;
  averagePathLength: number;
  communityModularity: number;
}

export function analyzeCitationNetwork(papers: ResearchConnection[]): CitationNetworkMetrics {
  // Placeholder implementation
  return {
    networkDensity: 0.15,
    clusteringCoefficient: 0.45,
    averagePathLength: 3.2,
    communityModularity: 0.78
  };
}