victorHeard about this: https://github.com/numenta/nupic in a podcast, it sounded very interesting. If I got it correctly, you could train (unsupervised) two models on two different corpus and resulting "semantic fingerprints" would still be comparable. I'm wondering what replicating some of the word embedding art with HTM would lead to.
This podcast btw: https://www.oreilly.com/ideas/natural-language-analysis-using-hierarchical-temporal-memory and it didn't mention numenta/nupic -- what I meant was that I found out about HTM through there, and then googled HTM and found an OSS project in Python.
Michael_Spencer@vhf With hashtags dead even on the birdsite, have to admit I'm enjoying using them here at on LinkedIn (via their new "trends" feature).
:two_hearts:Euphoria@Michael_Spencer @vhf Hastags are dead? Man, how can I be so behind on some things? Ah well. I came to them late and now I really enjoy inventing them, so I guess I'll keep it up.
Hmm.. now where did I put the window with your LinkedIn piece that I opened when you tooted it whenever the heck that was? Looking forward to reading it!
@Euphoria @vhf If you lose the URL, just check my bio and go to articles. Yes apparently machine learning now can recognize keywords everywhere without the need for the old "#". I still like them innately as signals for targeting.
#s are generally considered bad for Facebook, old for Twitter, but I still find them cute!
Well done, Michael! So well written and you really said something new and fresh that I hadn't seen anyone else say about Mastodon yet. I will be sharing it wherever possible.
@Michael_Spencer You're welcome! Thanks for taking the time to write yours and for sharing it.
I, personally, have a fondness for cheesy, but we also need crackers with our cheese.
@Michael_Spencer On the birdsite I think they are either geographic or global. Here they are instance-local in the same sense as the federated timeline, which means they should be much less noisy and much more tailored to an instance's average interest.
So yeah, agreed, they seem to be, on mastodon, a new and exciting variation on the hashtag theme.
@vhf Great insights vhf, thanks for the info. I didn't know here they were only instance-local, that's a bit odd.
@Michael_Spencer Everything is a bit odd when you compare federated to centralized (or even distributed).
Having them as "global" hashtags would require each instance to maintain a full index "hashtag->toot". So, all hashtags, all hashtagged toots. Would take a bunch of space and quite some CPU over time. It would also require all hashtagged toot to be broadcasted to (every) other instance, so network-intensive as well.
That's not how a federation operates. :)
@vhf Btw, do you know how this instance is comparing to others instances in terms of size of users? I attempted to dive into a dense one but I no longer know if that's the case.
@Michael_Spencer The instance you're on, you mean? Looks like reasonably sized: https://instances.mastodon.xyz/
@vhf Thanks for the link, looks like I might have to migrate to the Cloud. :laughing: