@martinibuster in the article you wrote you "this algorithm may not be in use". Why so certain now?
You are selectively quoting the sentence, which distorts the meaning. Here's the entire quote:
"While this algorithm may not be in use, or maybe used as part of a group of algorithms, it does serve as an example of how a “neural matching” algorithm could work."
That is a standard statement I use to head off Danny Sullivan who uses a similar statement to hide what is in use when someone asks them if a particular algo is in use. Bill Slawski's well acquainted with Google's standard response. I just put that in there to beat Google to it.
That whole statement is an acknowledgement that an algorithm can be used in any way, either partially, as part of another algorithm or in a way that's altered and improved from what's described in the research paper. The algorithms most likely to be a good candidate for use in information retrieval are the ones that report significant gains over previous methods, particularly without exorbitant processing costs (like an extreme amount of hardware at scale).
The studies that report modest advances are the ones I typically ignore as they are not likely to be used, particularly the ones that say there is a high processing cost to achieve a modest improvement. Those I ignore and don't ever talk about. This particular research is one of the papers that report a significant advance in the state of the art and an improvement over previous methods.
This part of my statement: "used as a part of a group of algorithms" means that the algorithm could be dropped in. The entire statement itself is an acknowledgement that nobody at Google has said that the algorithm is in use. Then I lay the case for why the algorithm is remarkable, why the algorithm is a perfect match for what Danny Sullivan described.
I dig through a lot of research papers, download them and do a bit of detective work to dig out which of the patents and research paper underlie the algorithms. Google does not tell us what the algorithms are.
As it happened, I had already discovered this paper and had jotted notes about it. When Danny Sullivan announced Neural Matching, the similarities were remarkable to be more than just a coincidence.
Google rarely if ever links to a specific research paper. In this case Google did not. For example, there are research papers for Caffeine, but the research papers have a different name for it, as I recall, and I'm pretty sure, from my recollection, that it is coffee related but not called caffeine.
I had discovered that algorithm previous to the announcement. When I read the announcement and the details, I noticed the extraordinary similarities between the algo and what Danny Sullivan had described. I'm fairly certain that the algo or an improved variant are what's powering Neural Matching.
As you can see, that 32 word boilerplate statement conveys everything I just wrote in over 300 words. It's necessary to do in 32 words because the full version would kill the flow of the article. ;)
Of further interest, I researched that phrase "neural matching" in relation to anything Google's published and, just like RankBrain, there is no matching patent or research paper. It's a brand name name Google slapped over a less "PR Sexy" algorithm name.
Now, lastly, after all of that digging around, reading the PRIOR RESEARCH that underlies this research and so on, I put it all together, omitting details of some of the prior research that this technology is built on (for brevity) it pissed me off to no end to see several plagiarists steal my content and others who stole my conclusions and linked to the same exact document which was never published or heralded on Google's official webmaster blog or by Danny, as if Google had stated "this is Neural Matching."
So all those articles you see on Neural Matching that cite that same paper without citing my original article, they are content spinners and plagiarists. ;)
I did the same thing with the Penguin Algorithm, I discovered that there was a new way to score links, called something like Distance Ranking.
Some people speculated (meaning they were guessing) that it was a trust algorithm.
These link ranking algorithms were a new way to rank links, different than what's been done before, like statistical analysis. So I laid down the case for that. Here's the article https://www.searchenginejournal.com/googles-penguin-algorithm-really-research/185261/
That article is an exhaustive research through these new kinds of link ranking algorithms. I cite several and connect them to what we know about Penguin. Nobody had ever done that before for Penguin. There was speculation, but no citations and no credible research and like I published in that article.
Unfortunately, it's a long article. It's not difficult to read. But it's long, because I felt I really needed to hammer it home and be complete. I try to keep it shorter now.
[edited by: martinibuster at 9:36 pm (utc) on Mar 14, 2019]