When you're reading a NLP paper that cites Aristotle, you know some serious shit is gonna go down.

Serious bullshit?

Sentiment analysis stuff, it's "how do we find relationships like 'Aristotle is human, all humans are mortal, therefore Aristotle is mortal'" kinda stuff.

https://sciencedirect.com/science/...

It didn’t mention “Transformers”?

@amoux nvm, it's an old paper. But damn, just a few years ago, researchers made “semantic search” sound more complex than it is lol

@atheist
Sentiment analysis sounds complicated.
But hadn't that been solved with prolog?

@scor it's actually kinda similar to prolog, you're right, I hadn't thought about that. The paper is talking about fuzzy word association with human text, the kinda stuff search engines use.

@atheist
So it's prolog reinvented / from scratch?
What are you using for such fuzzy word association?
I'd actually go into dictionaries of user groups or professions and build neuronal associations. Among the standard elastics & search term methods.
Wouldn't one?

@scor the paper above does fuzzy association for a given set of documents by creating an association between words and other words that appear in the same document as that word, then that association can be repeated to give n-order association, usually only interested in one or two orders of association.

The technique was used to identify fish oil as a possible treatment for Reynard's in the 90s based on text alone. Neither concept had appeared together and it wasn't something someone had tried, which was quite revolutionary at the time. Less so now, but it's a technique still in use.

Prolog on the other hand is much more explicit, parses clear rule statements then builds the associations from those rules. I think the paper describes one as inductive logic and the other as deductive logic.