najia-afk/aria_skills/sentiment_analysis

---
name: aria-sentiment-analysis
description: "💭 Multi-dimensional sentiment analysis with adaptive tone"
metadata: {"aria": {"emoji": "💭"}}
---

# aria-sentiment-analysis

Multi-dimensional sentiment analysis engine. Detects valence, arousal, and
dominance across messages. Derives frustration, satisfaction, and confusion
scores. Blends fast lexicon analysis with LLM classification. Tracks
conversation trajectories and recommends adaptive response tones.

## Architecture

User text ↓ SentimentLexicon (fast, ~30% weight) + LLMSentimentClassifier (kimi model, ~70% weight when ambiguous) ↓ SentimentAnalyzer → Sentiment(valence, arousal, dominance, primary_emotion) ↓ ↓ ConversationAnalyzer ResponseTuner (trajectory, volatility, (empathetic / step-by-step / turning points) celebratory / neutral) ↓ SemanticMemory (category: sentiment / sentiment_conversation)

## Integration

Hooked into `aria_mind/cognition.py` at Step 2.1 — every user message gets
automatic sentiment analysis. The derived sentiment and tone recommendation
are injected into the context dict for downstream agent use.

## Usage

```bash
# Analyze a single message
exec python3 /app/skills/run_skill.py sentiment_analysis analyze_message '{"text": "This is really frustrating, nothing works!"}'

# Analyze full conversation trajectory
exec python3 /app/skills/run_skill.py sentiment_analysis analyze_conversation '{"messages": [{"role": "user", "content": "help me"}, {"role": "assistant", "content": "sure!"}]}'

# Get tone recommendation for response tuning
exec python3 /app/skills/run_skill.py sentiment_analysis get_tone_recommendation '{"text": "I keep getting errors"}'

# View sentiment history
exec python3 /app/skills/run_skill.py sentiment_analysis get_sentiment_history '{"limit": 20}'

Functions

analyze_message

Analyze sentiment of a single message. Returns valence (-1 to 1), arousal, dominance, primary emotion, derived frustration/satisfaction/confusion scores, and an adaptive tone recommendation. Stores result in semantic memory.

analyze_conversation

Analyze sentiment trajectory of a full conversation. Detects whether the conversation is improving, declining, stable, or volatile. Identifies turning points and peak sentiment moments.

get_tone_recommendation

Quick tone recommendation without storing — returns one of 4 profiles:

• empathetic: High frustration → acknowledge + gentle guidance
• step-by-step: High confusion → structured walkthrough
• celebratory: High satisfaction → reinforce success
• neutral: Default balanced tone

get_sentiment_history

Return recent sentiment events from both the in-memory session history and the persistent semantic memory storage.

Dependencies

• api_client (semantic memory storage)
• LiteLLM proxy (kimi model for LLM classification)