|Adam's Lair Forum
|Word Vector Database? - Missing from GitHub.
|Page 1 of 1|
|Author:||A4L [ 2019/05/29, 04:28 ]|
|Post subject:||Word Vector Database? - Missing from GitHub.|
Hi there... a while ago I got pointed to something and I finally have gotten around to trying ot test it but the links are all broken...
I briefly tested the concept at a local game jam where my goal was to start with a classic RPG / adventure game dialog tree, but hide the dialog tree from the player by requiring them to enter free form text when talking to the character. This makes no sense with strict text matching, but what I wondered was whether it could work when the matching was fuzzy enough to hit a positive on the rough meaning of a sentence. Honestly, it didn't really work that well - but it didn't completely not work either And there's a lot of space to still be explored here.
This is it in action:
It's using this internal dialog tree, and as you can see it sometimes manages to find a good match despite input differences, but also sometimes fails doing so even though it was reasonably close.
And here are some word similarities based on the word vectors where I marked the ones with highest similarity scores:
If you're interested to look into this later on, all the source code is online here, though it's not set up in a user-friendly, documented way, and you first need to download 10GB of pre-trained (not mine!) word vectors. Let me know if you need help setting it up at some point.
Basically, the GitHub is still there but the "vector database" is missing.
What I have been doing is reading in hand made synonym lists into a Dictionary<string, string>... Dictionary<synonym, command>
Then I can search for the key.. based on any word they use.. and if it finds that it is a synonym it converts it to a hardcoded command... behold = look, inspect = look (etc etc)... I am also using a "stemmer" to strip words down so behold, beheld, beholding etc should all boil down to the same "key".
This is basically a really crappy version of the vector thing... which I guess measures similarity and outputs a value, this value determines if it is a matching synonym.. . removing the need for me to source large word lists on my own.
Anyway.. if you know of a place I can get the vector data so I can get this code up and running to test, I would be appreciative.
|Author:||Adam [ 2019/05/30, 14:42 ]|
|Post subject:||Re: Word Vector Database? - Missing from GitHub.|
It appears they moved the pretrained vectors around within their repository. Here's the new link, will fix it in the repo as well. Thanks!
|Page 1 of 1||All times are UTC + 1 hour [ DST ]|
|Powered by phpBB® Forum Software © phpBB Group