AI Sentiment Score (0-100)

Every news headline and X post that lands in the Top Tier Newswire database is scored 0-100 by a fine-tuned language model. 50 is neutral; values above 50 indicate bullish lean, below 50 bearish lean.

How the score is produced

We use a fine-tuned variant of an open-weights model (currently Qwen 2.5 7B AWQ, fine-tuned in-house on financial-text sentiment pairs). The model receives the headline plus the article summary and emits an integer 0-100. The fine-tune was done on a curated set of ~50,000 finance-text-to-sentiment examples covering earnings beats and misses, M&A announcements, FDA decisions, downgrades, recalls, fraud cases, and macro releases.

What the score is NOT

• Not a price prediction. A bullish-sentiment article doesn't mean the stock will rise. Sentiment captures the framing of an article, not the market's reaction.
• Not a recommendation. Two stories rated 80 might have very different implications for the underlying name. Sentiment doesn't replace reading the story.
• Not magnitude-aware. 80 vs 90 doesn't mean "twice as bullish" — the score is ordinal, not interval. Treat 80 and 90 as broadly equivalent strong-bullish reads.

Useful patterns

• Filter your watchlist: hide items below sentiment 65 to surface only meaningfully bullish news on stocks you care about
• Aggregate: a 24-hour average sentiment across all NVDA mentions tells you the dominant tone faster than reading each item
• Diverge: when a stock's sentiment is rising but price is falling, that's a notable disconnect (or vice versa)

How it interacts with AI Top Trades

The Top Trades model uses the rolling-window average sentiment plus the recency-weighted signal as inputs to its conviction calculation. A name with consistently high sentiment over 48 hours scores higher than one with a single 90 spike followed by neutrals.