MediaPulse
Agents/Agent Types/Analysis agent

Analysis Agent

ID: analysis

Purpose

Analyze collected media data to generate insights using a plugin-based architecture. The agent dynamically loads and executes registered analysis types, allowing for easy addition, removal, or modification of analysis capabilities without code changes.

Analysis Types

The Analysis Agent uses a plugin-based system where analysis types are registered in the database and loaded dynamically. By default, the system includes three built-in analysis types:

  1. Event/Context Analysis (event): Identifies and analyzes external events that could affect the company, such as natural disasters, political discourse, regulatory changes, economic shifts, social movements, and industry disruptions. Assesses the potential impact of these events on the company's operations, reputation, and strategic positioning.

  2. Competitive Analysis (competitive): Monitors media coverage of competitors and compares it to the company's coverage. Identifies industry trends, tracks competitive positioning in the media, and provides insights into how competitors are being perceived relative to the company.

  3. Sentiment Analysis (sentiment): Measures public perception and emotional tone of media coverage about the company. Analyzes how positively or negatively the company is being discussed across news and social media, tracks sentiment trends over time, and identifies key drivers of sentiment changes.

New analysis types can be added by registering them in the AnalysisTypeRegistry table. See Extending Analysis Types for details.

Inputs

interface AnalysisInput {
  tickerId: number; // Required: Foreign key to the Ticker table
  jobId?: string; // Optional: Job ID for tracking this run
}

Analysis Plugin Interface

All analysis plugins must implement the standard interface to be registered and executed by the Analysis Agent:

interface AnalysisPlugin {
  id: string                    // Unique identifier (e.g., 'sentiment', 'competitive')
  name: string                  // Human-readable name
  version: string               // Plugin version
  execute: (
    data: CollectedData,
    config: AnalysisConfig
  ) => Promise<AnalysisResult>
  validateConfig: (config: any) => boolean
  getOutputSchema: () => ZodSchema
  getSectionTemplate?: () => string // Optional: template for content generation
}

Plugin Registration

Plugins are registered in the AnalysisTypeRegistry database table with:

  • Metadata (id, name, version, description)
  • Configuration schema (Zod schema for validation)
  • Output schema (Zod schema for result validation)
  • Optional section template for content generation

See Extending Analysis Types for a complete guide on creating and registering new analysis plugins.

Configurations

Analysis configurations are stored in AgentConfig with identifier analysis and a flexible structure that supports any registered analysis type:

{
  analysis: {
    enabledTypes: string[],           // Which analysis types to run by default
    [analysisTypeId: string]: {        // Dynamic config per analysis type
      enabled: boolean,
      // Type-specific configuration (validated by plugin's configSchema)
      // Example for 'sentiment':
      model: 'gpt-4',
      sources: ['news', 'social'],
      aggregation: 'weighted_average',
      timeDecay: true
      // Example for 'competitive':
      peerGroupSize: 5,
      comparisonMetrics: ['mediaMentions', 'sentiment', 'coverageVolume'],
      industryTrends: true,
      marketShare: true
      // Example for 'event':
      eventTypes: ['natural_disaster', 'political', 'regulatory', 'economic', 'social_movement', 'industry_disruption', 'crisis'],
      impactAssessment: boolean,
      earlyWarning: boolean,
      minRelevanceScore: number
    }
  },
  ai: {
    model: 'gpt-4',
    temperature: 0.3,
    maxTokens: 2000
  }
}

Configuration Flexibility:

  • Each analysis type has its own configuration section, validated by the plugin's configSchema
  • Configurations can be overridden per ticker or user
  • New analysis types automatically get their own config section when registered
  • Configurations are hot-reloadable without agent restart

Outputs

The output structure is generic and supports any registered analysis type:

{
  agentId: 'analysis',
  agentVersion: string,                 // Semantic version (e.g., "1.2.3") of the agent that generated this output
  tickerId: number,
  jobId: string,
  timestamp: Date,
  executionTime: number,
  analyses: {
    [analysisType: string]: {          // Dynamic: one entry per executed analysis type
      type: string,                     // Analysis type ID
      results: any,                     // Type-safe results (validated by plugin's outputSchema)
      metadata: {
        executionTime: number,
        confidence: number,
        dataQuality: number
      }
    }
  },
  insights: Array<{
    type: string,                       // Analysis type ID
    priority: 'high' | 'medium' | 'low',
    title: string,
    description: string,
    dataPoints: string[]
  }>,
  metadata: {
    dataQuality: number,
    analysisConfidence: number,
    executedTypes: string[]             // Which analysis types were executed
  }
}

Example output (for built-in analysis types):

{
  agentId: 'analysis',
  tickerId: number,
  jobId: string,
  timestamp: Date,
  executionTime: 45000,
  analyses: {
    sentiment: {
      type: 'sentiment',
      results: {
        overall: 0.65,                  // -1 to 1
        breakdown: {
          news: 0.7,
          social: 0.6
        },
        trend: 'improving',
        keyDrivers: [...],
        summary: '...'
      },
      metadata: { ... }
    },
    competitive: {
      type: 'competitive',
      results: {
        marketPosition: '...',
        peerComparison: [...],
        industryTrends: [...],
        summary: '...'
      },
      metadata: { ... }
    },
    event: {
      type: 'event',
      results: {
        identifiedEvents: [...],
        riskLevel: 'medium',
        opportunities: [...],
        summary: '...'
      },
      metadata: { ... }
    }
  },
  insights: [...],
  metadata: {
    dataQuality: 0.92,
    analysisConfidence: 0.88,
    executedTypes: ['sentiment', 'competitive', 'event']
  }
}

Type Safety: Each analysis plugin's outputSchema ensures type-safe results. The Content Generation Agent uses these schemas to properly render analysis results in newsletters.

Process

  1. Initialize:

    • Load analysis config from database
    • Query AnalysisTypeRegistry for enabled analysis types
    • Load plugin configurations for requested analysisTypes
    • Validate inputs and configurations

  2. Plugin Discovery:

    • Look up each requested analysis type in AnalysisTypeRegistry
    • Verify plugins are enabled and configurations are valid
    • Load plugin implementations (built-in or custom)

  3. Read Data: Query database for collected data for the specified ticker (independent of Data Collection Agent)

  4. Parallel Analysis Execution:

    • Execute each requested analysis type in parallel
    • Each plugin's execute() method is called with collected data and its configuration
    • Results are validated against the plugin's outputSchema

  5. Insight Extraction:

    • Use AI to identify top insights across all executed analyses
    • Cross-reference insights from different analysis types
    • Prioritize insights based on user preferences and focus areas

  6. Return: Structured analysis results with all executed analysis types

Dynamic Execution: The number of parallel workers is determined by the number of requested analysis types, not a fixed value. New analysis types are automatically executed when registered and enabled.

Sequence Diagram

Data Collection Agent

Previous Page

Extending Analysis Types

Next Page