Skip to content

sentiment_analysis.md

Latest commit

 

History

History
44 lines (30 loc) · 1.99 KB

sentiment_analysis.md

File metadata and controls

44 lines (30 loc) · 1.99 KB

Example: Sentiment and Alternative Data

AlphaMind exposes alternative-data sources through the API and includes a sentiment model in the research library. This example shows both.

The alternative-data API

curl -s http://localhost:8000/api/v1/alternative-data/sources | python -m json.tool

Each source has a type (satellite, sentiment, sec, or social), a status, a dataPoints count, and a latency. This endpoint backs the Alternative Data screen in both clients.

The sentiment model (research library)

A text sentiment classifier lives at code/backend/analytics/alternative_data/sentiment_analysis.py as MarketSentimentAnalyzer. It is a small embedding-based neural model (TensorFlow). Run from code/backend so the analytics package is importable, with tensorflow installed.

from analytics.alternative_data.sentiment_analysis import MarketSentimentAnalyzer

texts = [
    "Earnings beat expectations and guidance was raised",
    "The company missed targets and cut its outlook",
    "Trading was flat with no major news",
]
labels = [1, 0, 1]  # example training labels

analyzer = MarketSentimentAnalyzer(vocab_size=500, embedding_dim=16, max_length=20)
analyzer.prepare_tokenizer(texts)
analyzer.train(texts, labels)             # see source for the full signature

scores = analyzer.get_sentiment_score([
    "Strong quarter, raising forecasts",
])
print("sentiment scores:", scores)

Public methods include prepare_tokenizer, preprocess_text, train, predict, get_sentiment_score, and save / load. Inspect the source for exact argument shapes.

Related modules

  • analytics/alternative_data/scrapers/sec_8k_monitor.py — an SEC 8-K monitor that computes a sentiment label per filing.
  • analytics/alternative_data/satellite_processing.py — a satellite feature extractor.

These are library modules for experimentation. The live /api/v1/alternative-data/sources response is a source registry and does not run these pipelines on each request.