Sentiment Analysis Agent

The Sentiment Analysis Agent is a specialized component of the trading system designed to gauge market psychology. It processes real-time news feeds, social sentiment, and volatility metrics to determine if the market is driven by irrational fear, excessive greed, or neutral fundamentals.

Overview

The agent leverages a Retrieval-Augmented Generation (RAG) architecture to combine live data with curated trading wisdom. By analyzing news headlines and the Crypto Fear & Greed Index, it identifies potential trend reversals and provides a "psychological" weight to the final trading decision.

Key Capabilities

• News Aggregation: Fetches and cleans news articles from multiple RSS feeds and the NewsAPI.
• Fear & Greed Interpretation: Analyzes market sentiment indices to identify overbought or oversold psychological conditions.
• Contrarian Signaling: Identifies scenarios where extreme sentiment may signal a reversal (e.g., "Buying the fear").
• RAG Integration: Accesses a local vector database of sentiment-specific trading strategies and historical papers.

Data Sources

The agent primarily consumes data through the NewsAPIConnector, which implements a tiered fetching strategy to ensure high availability:

| Source | Method | Description | | :--- | :--- | :--- | | Crypto RSS | Web Scraping | Fetches from CoinTelegraph and CoinDesk (includes full descriptions). | | Google News | RSS | Provides broad market coverage without requiring an API key. | | NewsAPI | REST API | Official fallback for high-quality, filtered financial news (requires NEWSAPI_KEY). |

Response Schema

The Sentiment Agent returns a structured SentimentAnalysis object. This ensures the TradingOrchestrator can programmatically parse its reasoning.

SentimentAnalysis Model

Inherits from AgentRecommendation.

| Field | Type | Description | | :--- | :--- | :--- | | recommendation | BUY \| SELL \| HOLD | The sentiment-driven action. | | confidence | int (0-100) | Confidence level in the analysis. | | reasoning | str | Detailed explanation of the current market mood. | | metadata | SentimentMetadata | Specific sentiment metrics (see below). |

SentimentMetadata

| Field | Type | Description | | :--- | :--- | :--- | | news_sentiment | str | POSITIVE, NEGATIVE, or NEUTRAL. | | fear_greed_interpretation | str | Interpretation of the Fear & Greed Index. | | contrarian_signal | bool | True if the agent suggests acting against the crowd. |

Usage Example

The Sentiment Agent is typically invoked via the TradingOrchestrator, but it can be used independently for market research.

from agents.schemas import SentimentAnalysis
from data_connectors.newsapi_connector import NewsAPIConnector

# 1. Fetch live news data
connector = NewsAPIConnector()
news_data = connector.get_bitcoin_news(limit=5)

# 2. Example of the agent's structured output
# This is generated by the LLM via the SentimentAnalysis Pydantic schema
analysis = SentimentAnalysis(
    recommendation="BUY",
    confidence=75,
    reasoning="Market is in a state of 'Extreme Fear' (20/100). News headlines show stabilization after a major dip. Historical RAG data suggests this is a high-probability accumulation zone.",
    metadata={
        "news_sentiment": "NEUTRAL",
        "fear_greed_interpretation": "EXTREME_FEAR",
        "contrarian_signal": True
    }
)

print(f"Action: {analysis.recommendation} (Confidence: {analysis.confidence}%)")
print(f"Contrarian Play: {analysis.metadata.contrarian_signal}")

Knowledge Indexing (RAG)

The Sentiment Agent is "trained" on specific domain knowledge stored in knowledge/sentiment/. The KnowledgeIndexer processes these files into a vector store to help the agent ground its decisions in proven sentiment-trading theories.

To index new sentiment research or strategy PDFs:

1. Place the files in knowledge/sentiment/ or knowledge/papers/.
2. Run the indexer:

from agents.rag.indexer import KnowledgeIndexer

indexer = KnowledgeIndexer()
results = indexer.index_domain("sentiment", "knowledge/sentiment")
print(f"Indexed {results} sentiment knowledge chunks.")

Internal Workflow

1. Context Collection: The agent receives a query containing the latest price action and Fear & Greed values.
2. Document Retrieval: The RAG system searches the sentiment vector store for relevant psychological trading rules.
3. Sentiment Extraction: The LLM parses recent news headlines from the NewsAPIConnector.
4. Synthesis: The agent compares the live news mood against historical sentiment cycles and outputs the structured SentimentAnalysis schema.