Quantify current market sentiment per asset
Describes the keyword sentiment of the most-covered assets in the recent news window. It reads article headlines and descriptions, attributes them to assets, and counts bullish and bearish keywords once per article. Each scored asset carries a score from -1 to 1, a bullish, neutral or bearish label, and its keyword and article counts. The run's published computation rule states how the counts become scores and labels. These are mechanical readings of news language, not confidence estimates or price forecasts.
Lexicon sentiment analysis - the oldest form of text classification, and the one with the best-documented failure modes.
THE METHOD. A bag-of-words approach: count each bullish or bearish keyword at most once per article across its headline and description. The score describes the balance of those hits. Read the run's published computation rule for its normalization, rounding and label boundaries. The published keyword evidence gives each asset's distinct words, hit counts and article count for this window, not a day-by-day series. A scale endpoint is not a confidence estimate or a price forecast. No grammar, no word order, no context - just membership in two lists.
THE ACADEMIC RESULT THAT MAKES IT WORTH DOING. Tetlock (2007) measured pessimism in a Wall Street Journal column using a general-purpose lexicon and found that measured pessimism predicted downward pressure on prices followed by reversion. Read the claim precisely: that is predictive CONTENT for short-run movements, and the design does not by itself establish that media tone carries fundamental information or causes the move - the reversion is equally consistent with temporary sentiment pressure. So news sentiment is a real object of study, not a gimmick.
THE RESULT THAT SHOWS WHY THE LEXICON MATTERS ENORMOUSLY. Loughran and McDonald (2011) found that general-purpose word lists systematically misclassify financial text, because words that are negative in ordinary English are neutral or technical in finance - LIABILITY, TAX, COST, CAPITAL, DEPRECIATION. Nearly three-quarters of the words the standard Harvard list marked negative were not negative in a financial filing. They built a finance-specific lexicon in response, and it is the standard reference for exactly this problem. Any keyword sentiment system inherits this weakness in proportion to how generic its lists are.
FOUR FAILURES A WORD LIST CANNOT FIX, and note that adding entries makes none of them better.
The honest test is whether a human would need to READ THE SENTENCE to decide. Here they plainly would - which is the signal that context-free string matching is inadequate. The remedy is any CONTEXT-SENSITIVE method: parsed rules, a supervised classifier trained on labelled financial text, or a language model. A model is one answer, not the only one.
WHY THIS FAMILY IS TAUGHT TOGETHER. Six analyses in this catalog score text by keyword. They share this ceiling, so their agreement is not independent confirmation - it is the same method applied to the same corpus.
The word lists, the -1 to +1 scale and the bullish/neutral/bearish cut-offs are our own convention.
Published reports are readable without an account. The newest is Ether (ETH) and XRP extend a month-long slide as Bitcoin falls below $75,000: trend, momentum and news (Nov 9, 2026). Create an account to run research on the stocks and crypto you choose.