Why simple keyword counting fails in finance—and how AI is changing the game.

Hinzugefügt: 2026-04-29

[00:00:00]Uh, you know, one of the primary areas where where ML has helped so far is processing natural language.

[00:00:06]Uh, being able to tell if something is good or bad news.

[00:00:09]And the quants have done this for a million years. Um, the old-fashioned way to do it is to build up keywords and phrases and assign numerical scores.

[00:00:18]Way oversimplifying. Do you see the word increasing? Plus one.

[00:00:23]Count up all you know, similar words.

[00:00:26]And of course, you you immediately see the problem if the actual sentence was massive embezzlement is increasing, plus one was probably not your best guess.

[00:00:36]Now, quants can survive looking silly 47% of the time if we get it right 53% of the time. So, I'm not knocking this old approach. I'm just saying it's very noisy.

[00:00:46]It turns out that applying uh, natural language processing to to language data, language is a very non-linear thing.

[00:00:54]Whether a certain word is good or bad, um, might depend on this sentence like on the simple one I just gave you, but it might depend on three sentences ago.

[00:01:02]Nothing is perfect. Language is very complicated. But it turns out NLP applied to to to textual data to forecast returns, we found to be very powerful. We've had some great success with it. Uh, I can tell you in general what it's doing. It's doing a fundamental formula of momentum. Are good things happening?

[00:01:22]But can I tell you precisely why it likes this stock?