Stock market prediction using natural language processing

US8285619, 'Stock market prediction using natural language processing,' was filed January 22, 2002, claiming priority to a provisional a year earlier, on January 22, 2001. The named inventors are a genuinely credentialed group for the subject: Frederick S. M. Herz, Lyle H. Ungar and Jason M. Eisner — the latter two well-known academic figures in machine learning and computational linguistics — together with Walter Paul Labys. Filing an NLP-reads-the-news trading patent in 2001–2002 was early; this was the bag-of-words era, when 'natural language processing' meant hand-built linguistic rules and word-frequency statistics.

What it claims

The patent describes automatically extracting company-relevant information from news feeds using natural language processing, filling standardized templates from that text, clustering similar announcements together, and correlating those events with subsequent stock-price movements to predict future behavior. Strip away the 2002 vocabulary and it is recognizably the skeleton of a modern news-sentiment trading signal — the exact thing that became a crowded commercial idea two decades later.

The long road to grant — and why 2022 is the wrong date

It is tempting to say this patent 'expired twenty years after filing,' on January 22, 2022. That is the naive estimate, and it is wrong. The application sat in prosecution for a decade, not granting until October 9, 2012. US law compensates applicants for that kind of Patent Office delay with a patent-term adjustment (PTA) that pushes the expiration date later. Here the adjustment was large: the USPTO lists an adjusted anticipated expiration of April 20, 2028 — more than six years past the naive twenty-from-filing date.

This is a clean illustration of a caveat this site repeats everywhere: the '20 years from filing' math can be wrong in both directions. An adjustment can push the real term-end years later than the estimate, and — as here — an early lapse can pull the actual death years earlier.

How it actually died

The patent never reached that 2028 adjusted date. The USPTO's current legal status for US8285619 is 'Expired — Fee Related,' meaning it lapsed for non-payment of a maintenance fee rather than running its full term. US patents carry escalating maintenance fees at 3.5, 7.5, and 11.5 years after grant; our records place the lapse in late 2024, consistent with a decision not to pay the largest, final maintenance fee due around the eleven-and-a-half-year mark.

The owner by then was Fred Herz Patents, LLC — a patent-holding entity, assigned the patent in 2014. It kept the patent alive for over a decade and then, near the end, let it go, walking away from roughly three and a half years of remaining adjusted term.

The timing, told straight

The lapse landed at an almost perverse moment. ChatGPT launched in November 2022; by 2024, running a language model over financial news and trading the sentiment had gone from a niche quant experiment to something close to a default idea. A patent whose claims read squarely on that concept was abandoned for the price of a maintenance fee, right as the concept it described became commercially obvious.

It is worth being careful about what that does and does not mean. The patent was still in force when ChatGPT launched; it did not 'expire before ChatGPT.' What happened is quieter and, arguably, more interesting: a prescient, hard-won patent was allowed to lapse early, so the mechanism it claimed became free to build years ahead of its own scheduled expiry. This patent also has a full working resurrection — see the grave at /graves/news-sentiment-signals.