Asset-Specific News
Sentiment Classification
Sentiment is classified by a financial NLP lexicon — a fixed list of weighted keywords. No AI models are used.
Method: The title is lowercased and scored by summing the weights of all matching tokens.
Decision thresholds:
- Score ≥ +2 →
bullish - Score ≤ −2 →
bearish - Score between −1 and +1 →
neutral(article is discarded)
Bullish keywords (examples):
| Keyword | Weight |
|---|---|
| "surge", "soar", "record high" | +3 |
| "beat", "rally", "upgrade", "profit", "buyback", "approved", "better than expected" | +2 |
| "growth", "strong", "gains", "positive", "boost", "higher", "dividend" | +1 |
Bearish keywords (examples):
| Keyword | Weight |
|---|---|
| "bankruptcy", "bankrupt" | −4 |
| "plunge", "crash", "layoffs", "fraud", "default" | −3 |
| "miss", "decline", "downgrade", "loss", "cut", "shortfall" | −2 |
| "fall", "reduce", "debt", "warning", "risk", "weak", "drop" | −1 |
Retention: Articles older than 2 days are automatically removed at the end of each scraping cycle.