sense2vec.spacy.io
AI
word embeddings
SHOW HN2016
225points
79comments
#112of 2,264 in 2016
Top 5%of 2016 by points
How HN took it
MixedNo clear lean
11 positive, 19 neutral or asking, 3 mixed, 22 critical, out of 55 comments about the project. Thread score -0.24.
Execution: 5 for, 19 againstIdea: 5 for, 1 againstPerformance: 0 for, 2 against
The thread, comment by comment
Each row is a top-level comment and its replies, in HN's order. Select a cell to open that comment on HN.
positive neutral or mixed critical about another commenter or something else the maker
What HN said
What people liked
Good results for anime, bands, and game jargon
From 3 comments
Interesting side effect: it has learned to give somewhat decent anime recommendations. At the very least it recognizes some basic genres to an extent (for example, TTGL is close…
PeCaN on HNThe band names are particularly cool. It is sorta like a recommendation engine. If you like Pink floyd, you might like Led Zepplin.
andrewtbham on HNPretty good at picking up the Smash Bros Jargon: https://sense2vec.spacy.io/?dash_dance|NOUN
jldugger on HN
Impressive semantic results overall
From 2 comments
It's impressive how good it is. But there is still room for improvement. Searching "Haskell" leads to Clojure and C++. Yes, these are both Programming Languages, but out of all I…
1ris on HNSearched for "Best Iron Maiden album" and it was smart enough not to pick one. Then I tried "Seinfeld" and it returned Seinfeld as the 10th result behind some other sitcoms,…
finishingmove on HN
Objections
Search with Enter key does not work
From 3 comments
is it just me or does it not search when you hit enter?
Palomides on HNIt's not just you, I need to click the search icon.
justinsaccount on HNNothing happens for me where I press enter either, I have to click on the search icon (firefox on linux).
ssalenik on HN
NSFW and some phrases return no results
From 3 comments
NSFW doesn't work: https://sense2vec.spacy.io/?NSFW%7CNOUN
thekingshorses on HNSearched for something NSFW and didn't get any results :(
rocky1138 on HNtried "netflix and chill", didn't find anything? That to me is the phrase of the year that is more than the sum of its parts.
cowsandmilk on HN
Some results are odd, e.g. Haskell near Clojure and C++
From 2 comments
It's impressive how good it is. But there is still room for improvement. Searching "Haskell" leads to Clojure and C++. Yes, these are both Programming Languages, but out of all I…
1ris on HNThe first name to show up when I search 'evil man' is 'Mr Rogers'...what the hell, reddit. (first name, not first word, had to scroll down a bit, past words like 'Nazi soldier'…
gradi3nt on HN
Requests
Downloadable trained vectors
From 1 comment
Would it be possible for you to make the trained vectors available for download and independent analysis without having to rescrape and retrain the model?
j2kun on HN
Phrase and multi-word query support
From 2 comments
tried "netflix and chill", didn't find anything? That to me is the phrase of the year that is more than the sum of its parts.
cowsandmilk on HNI tried "rick and morty" and got the same issue. My guess is it has something to do with the term "and."
sgarman on HN
An API
From 1 comment
This is awesome! I would pay for an API.
Kiro on HN
Questions
Can vector algebra queries be supported
From 1 comment
Now can't you experiment with algebra operations on those vectors, like "queen minus female plus male = king"?
MasterScrat on HN
Can meanings be separated by subreddit
From 2 comments
Word vectors are really intriguing to me, and I have a few questions: If you also trained a model on a different news aggregator's comments, would it be possible to "match up"…
rockmeamedee on HNLet's say we want to know how usage differs in one subreddit vs another. If you just train two entirely separate models, you end up with two vector spaces. The meanings within…
syllogism on HN
Alternatives named
WordNet
From 2 comments
This is very cool. I have played with the word2vec download a lot and I have been sorta amazed. If you had a lot of money... like google... it would be cool to train word vectors…
andrewtbham on HNI don't think wordnet (and similar) will ever be obsolete. From my point of view, they are a "curated" subset of the semantic that you can obtain with word2vec. The latter can…
ecesena on HN
NLTK
From 1 comment
Great to see another use case of spacy...I've been wanting to give it a go ever since OP pitched it as an alternative to NLTK.
danso on HN