Skip to content

Emotion Detection

Overview

This document describes how AICO detects the user's emotional state from their inputs and exposes it as structured signals for other components (Emotion Simulation, AMS, crisis detection, memory, frontend).

For AICO's own internal emotional life (AppraisalCloudPCT/CPM), see emotion-simulation.md. For system-wide integration and believability, see emotion-integration.md.

Goals

  • Accurately infer the user's emotional state from multimodal inputs (initially text, later voice/vision).
  • Provide compact, structured signals that are easy for other components to consume.
  • Support real-time responsiveness and privacy-preserving, local-first processing.

Core Responsibilities

  • Analyze user inputs for:
  • Primary and secondary emotions.
  • Valence, arousal, and dominance.
  • Stress, fatigue, and crisis indicators.
  • Interaction intent related to emotion (e.g., venting, celebrating, seeking advice).
  • Publish these results as message bus events (topics TBD in Phase 2 implementation).
  • Provide stable REST endpoints (via the backend) for diagnostic and tooling purposes.

Modelservice Integration

  • Sentiment vs Emotion
  • Sentiment: Lightweight polarity/intensity (negative/neutral/positive) for coarse-grained signals.
  • Emotion: Richer, multi-dimensional affect (labels, valence, arousal, dominance, dimensional vectors).
  • Both are implemented as standard modelservice request/response topics and mirrored via REST endpoints (e.g. /api/v1/affective/sentiment, /api/v1/affective/emotion).

Message Bus Topics

  • Input (from other systems):
  • CONVERSATION_USER_INPUT – current user messages from conversation engine.
  • Conversation context from memory/AMS systems.
  • Output (produced by detection):
  • User emotion signals published to message bus for consumption by EmotionEngine and other components.
  • Voice analysis signals (future multimodal).

Detailed JSON examples for these messages are defined in emotion-messages.md.

Downstream Consumers

  • Emotion Simulation – uses user emotion as input to appraisal.
  • AMS/Memory – records emotional context and patterns to learn effective strategies.
  • Crisis Detection – monitors for high-risk emotional states and triggers protocols.
  • Frontend – may show user-facing indicators or adapt UI subtly (e.g., warning banners for crisis).

Design Principles

  • Local-first, privacy-preserving processing.
  • Compact, stable schemas for easy consumption.
  • Clear separation from AICO's simulated emotion while remaining tightly integrated through the dual emotion system.