Sentiment
Every news article and earnings release this system holds is scored independently by finbert-finbert (ProsusAI/finbert) and finbert-sentence-finbert (ProsusAI/finbert) and vader (unknown). The two score on different scales, so nothing on this page averages them together: every figure is either one line per analyzer or one analyzer at a time, and every document below carries the name and version of the model that scored it.
Coverage
A sentiment result is one (analyzer, document) pair, so a document scored by both models carries two rows. The document counts below are distinct documents, not results.
Counts computed 2026-11-22 15:33 EST, on a worker — not measured when this page loaded.
-
IR feed items are scored zero times today.
sentiment_resultscan attach to a news article, an earnings event or an IR feed item, and the third arm has never been used. Anir_feed_itemsrow has no company foreign key of its own, so a scored row could not be linked to a company safely. - The figures below cover the news grain only, bucketed by the article's own publication date. The earnings-release grain carries a calendar date on the event rather than a publication timestamp, so mixing the two on one weekly axis would put the two grains on different clocks. Earnings-release scores are readable in the document list below by switching the grain, and per company on each company page.
- Market-wide figures cover every scored article in the window, regardless of the filters on the document list further down this page. The two sections describe different populations on purpose, and neither is a subset of the other unless you make it one.
Market-wide
Figures computed 2026-11-22 15:33 EST, on a worker — not measured when this page loaded.
| Analyzer | Version | Type | Documents in window | Mean score | Positive | Neutral | Negative |
|---|---|---|---|---|---|---|---|
| finbert-finbert | ProsusAI/finbert | FINBERT | 143,377 | +0.2205 | 49,632 (34.6%) | 76,898 (53.6%) | 16,847 (11.8%) |
| finbert-sentence-finbert | ProsusAI/finbert | FINBERT | 74,328 | +0.1731 | 17,386 (23.4%) | 47,704 (64.2%) | 9,238 (12.4%) |
| vader | unknown | VADER | 485,433 | +0.4625 | 326,870 (67.3%) | 94,685 (19.5%) | 63,878 (13.2%) |
Market-wide sentiment over time
Mean sentiment score per ISO week over the trailing 26 weeks, bucketed by the ARTICLE's publication date, over every scored news article published in that window — every company, no filter. One line per analyzer and never a mean across them: vader and finbert-finbert score on different scales, so a single averaged line would be a number about nothing. A week an analyzer did not score is a gap in its line, not a zero.
| Week starting | finbert-finbert | finbert-sentence-finbert | vader |
|---|---|---|---|
| 2026-11-16 | — | — | +0.5989 |
| 2026-11-09 | — | — | +0.5428 |
| 2026-11-02 | — | — | +0.5211 |
| 2026-10-26 | +0.0244 | — | +0.5522 |
| 2026-10-19 | — | — | +0.5587 |
| 2026-10-12 | — | — | +0.5593 |
| 2026-10-05 | — | +0.1953 | +0.5828 |
| 2026-09-28 | — | +0.1628 | +0.5394 |
| 2026-09-21 | — | +0.1704 | +0.4855 |
| 2026-09-14 | +0.0894 | +0.1656 | +0.4428 |
| 2026-09-07 | +0.0879 | +0.1740 | +0.3839 |
| 2026-08-31 | +0.0944 | +0.1888 | +0.4309 |
| 2026-08-24 | +0.1930 | +0.2214 | +0.3254 |
| 2026-08-17 | +0.2354 | +0.0273 | +0.3151 |
| 2026-08-10 | +0.2457 | +0.0310 | +0.2746 |
| 2026-08-03 | +0.2349 | +0.4278 | +0.2425 |
| 2026-07-27 | +0.2210 | — | +0.2499 |
| 2026-07-20 | +0.2240 | — | +0.2609 |
| 2026-07-13 | +0.2243 | — | +0.2897 |
| 2026-07-06 | +0.2399 | — | +0.3196 |
| 2026-06-29 | +0.2300 | — | +0.3105 |
| 2026-06-22 | +0.2172 | +0.0184 | +0.2850 |
| 2026-06-15 | +0.2514 | — | +0.3206 |
| 2026-06-08 | +0.2286 | -0.0092 | +0.2941 |
| 2026-06-01 | +0.2056 | — | +0.2836 |
| 2026-05-25 | +0.2410 | — | +0.3413 |
The vertical axis includes zero and is otherwise scaled to the data, so the distance between the two lines is a difference of scale between two models, not a disagreement about the market.
Label composition per week — finbert-finbert ProsusAI/finbert
How finbert-finbert classified each week's articles over the trailing 26 weeks, stacked as counts rather than shares so the height of a week is also its volume. One analyzer at a time: the label thresholds belong to the model, so stacking two models together would count the same article twice under two different definitions of positive.
| Week starting | Positive | Neutral | Negative | Total |
|---|---|---|---|---|
| 2026-10-26 | 0 | 1 | 0 | 1 |
| 2026-09-14 | 231 | 715 | 148 | 1,094 |
| 2026-09-07 | 790 | 2,825 | 495 | 4,110 |
| 2026-08-31 | 936 | 2,819 | 571 | 4,326 |
| 2026-08-24 | 3,180 | 4,368 | 1,373 | 8,921 |
| 2026-08-17 | 5,006 | 5,161 | 1,987 | 12,154 |
| 2026-08-10 | 6,742 | 7,946 | 2,311 | 16,999 |
| 2026-08-03 | 7,191 | 12,161 | 1,944 | 21,296 |
| 2026-07-27 | 4,823 | 8,342 | 1,467 | 14,632 |
| 2026-07-20 | 3,082 | 5,203 | 948 | 9,233 |
| 2026-07-13 | 2,076 | 3,373 | 643 | 6,092 |
| 2026-07-06 | 1,724 | 2,479 | 553 | 4,756 |
| 2026-06-29 | 1,610 | 2,772 | 464 | 4,846 |
| 2026-06-22 | 1,652 | 2,984 | 518 | 5,154 |
| 2026-06-15 | 1,563 | 2,440 | 415 | 4,418 |
| 2026-06-08 | 1,764 | 2,985 | 520 | 5,269 |
| 2026-06-01 | 1,799 | 3,598 | 550 | 5,947 |
| 2026-05-25 | 5,463 | 6,726 | 1,940 | 14,129 |
Extremes — finbert-finbert ProsusAI/finbert
Mean score per company over the trailing 26 weeks, from finbert-finbert alone. A company appears only if it carries at least 20 scored documents in the window: a company with one scored article is not a ranking, and the count is printed on every row so a mean over the floor and a mean over three hundred documents are visibly different measurements. Ranking across both analyzers is not offered, because ordering companies by an average of two scales is not an ordering of anything.
Rankings computed 2026-11-22 15:41 EST, on a worker — not measured when this page loaded.
Most positive
Highest mean score, ties broken by ticker.
| Company | Mean score | Documents |
|---|---|---|
| ELEK ELEKTROS INC | +0.7448 | 35 |
| OZSC OZOP ENERGY SOLUTIONS, INC. | +0.7169 | 21 |
| VIVK Vivakor, Inc. | +0.6794 | 21 |
| VJET Catheter Precision, Inc. | +0.6707 | 20 |
| ATEYY ADVANTEST CORP | +0.6681 | 20 |
| BLIN Bridgeline Digital, Inc. | +0.6557 | 22 |
| VRXA Veraxa Biotech Holding AG | +0.6534 | 20 |
| SPCB SuperCom Ltd | +0.6497 | 27 |
| MCS MARCUS CORP | +0.6368 | 22 |
| FE FIRSTENERGY CORP | +0.6347 | 29 |
| CMND Clearmind Medicine Inc. | +0.6285 | 25 |
| OPTX SYNTEC OPTICS HOLDINGS, INC. | +0.6269 | 27 |
Most negative
Lowest mean score, ties broken by ticker.
| Company | Mean score | Documents |
|---|---|---|
| FVRR Fiverr International Ltd. | -0.5857 | 20 |
| USO United States Oil Fund | -0.4611 | 52 |
| UWMC UWM Holdings Corp | -0.4318 | 43 |
| TTD Trade Desk, Inc. | -0.4289 | 42 |
| JBI Janus International Group, Inc. | -0.4184 | 22 |
| UAA Under Armour, Inc. | -0.4163 | 25 |
| HUCK DOMO, INC. | -0.3860 | 27 |
| XPOF Xponential Fitness, Inc. | -0.3842 | 24 |
| XXI Twenty One Capital, Inc. | -0.3704 | 20 |
| ACM AECOM | -0.3507 | 25 |
| MYGN MYRIAD GENETICS INC | -0.3460 | 25 |
| FLGR Franklin FTSE Germany ETF | -0.3409 | 23 |
Scored documents
One row per (analyzer, document): an article scored by both models appears twice, with two scores, which is the truth rather than a duplicate. Every row carries the text the model actually read, the article's publisher and publication date, and a link to the original. The window is always bounded — at most 366 days — because an unbounded ordering over 1.4 million results is the one query that would scan the corpus on every page view.