thisplusthat
AI
word vector search
SHOW HN2013
148points
64comments
#133of 1,739 in 2013
Top 8%of 2013 by points
How HN took it
CuriousMostly questions and neutral comments
11 positive, 31 neutral or asking, 1 mixed, 14 critical, out of 57 comments about the project. Thread score -0.03.
Performance: 5 for, 10 againstIdea: 4 for, 1 againstExecution: 0 for, 3 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
Interesting, fun results with some queries
From 3 comments
Interesting. Currently it generates garbage for lot of queries but, some stuff is kinda fun. Forrest Gump - comedy + romance gives pulp fiction (!), as good as it gets (match)…
grishma on HNLisp + JVM = Clojure I'm sold. This is really cool! (Though it's worth noting that a Google search with the same terms returns the exact same result...)
nemo1618 on HNPretty impressive for my first try. iPad - cool -> Windows Phone
epaga on HN
Concept is cool and relevant to related work
From 3 comments
Lisp + JVM = Clojure I'm sold. This is really cool! (Though it's worth noting that a Google search with the same terms returns the exact same result...)
nemo1618 on HNFantastic work and is relevant to something we are working on in this space. Thanks. On a lighter note I tried "sarah palin + sexy" and I got John Mccain, Hillary Clinton and…
somberi on HNInteresting. I've played around with words as vectors (with values) and the cosign similarity algorithm (http://en.wikipedia.org/wiki/Cosine_similarity). This is very cool stuff.…
emehrkay on HN
Objections
Results are often nonsensical or uninspiring
From 3 comments
Hmm. It's a lovely idea but I find the results uninspiring. (Which isn't surprising; it's a very difficult problem. But I was hoping to be amazed.) Here are the examples I tried…
gjm11 on HNI was really excited by this writeup, so I tried it. Four test queries returned nothing that seemed useful or even relevant: fluid dynamics + electromagnetism : expected…
leephillips on HNI had the same experience. I haven't really gotten any convincing results except for the example queries verbatim. Some queries I tried: medicine - science + superstition…
tveita on HN
Limited training data hurts accuracy
From 1 comment
It's not perfect; the real limiting factor is the volume of text. For the research paper behind word2vec, to get accurate associations for common words (King, man, woman, etc.)…
juxtaposicion on HN
Server crashes on non-ASCII input
From 2 comments
Bug report: using some non-ascii characters crashes the server (for example é or É).
cocoflunchy on HNThis is neat, and I found a few queries that added interesting results. However, I tried ``` Slavoj Žižek - Jacques Lacan - Hegel ``` which yielded an internal server error,…
jboynyc on HN
Requests
Access to the raw vector database
From 2 comments
I'd love access to the raw vector database if you feel like sending it my way.
doctoboggan on HNHey doctoboggan, awesome questions! >> What is the dimensionality of each word vector and what does a words position in this space "mean"? What is this dimensionality determined…
juxtaposicion on HN
Faster nearest-neighbour search
From 2 comments
Hey doctoboggan, awesome questions! >> What is the dimensionality of each word vector and what does a words position in this space "mean"? What is this dimensionality determined…
juxtaposicion on HNThanks! The number crunching is indeed very intense. The full body of vectors has a million rows and thousand columns, which fills up all of the 10GB of available memory. When…
juxtaposicion on HN
Questions
How the vectors and dimensions work
From 2 comments
Hey juxtaposicion, fascinating work. I have many questions so I am just going to shoot them rapid fire. What is the dimensionality of each word vector and what does a words…
doctoboggan on HNHey doctoboggan, awesome questions! >> What is the dimensionality of each word vector and what does a words position in this space "mean"? What is this dimensionality determined…
juxtaposicion on HN
How to handle changing content
From 2 comments
Interesting concept, but how will it work with more dynamic content? You can train the model on a fairly static corpus such as Wikipedia, but what if you content changes with a…
donretag on HNHi, creator here (Chris Moody). Great question. The underlying algorithm, word2vec, (https://code.google.com/p/word2vec/) isn't built for streaming data which means that at the…
juxtaposicion on HN
Alternatives named
Google search
From 1 comment
Lisp + JVM = Clojure I'm sold. This is really cool! (Though it's worth noting that a Google search with the same terms returns the exact same result...)
nemo1618 on HN
Latent semantic indexing
From 2 comments
Does this relate to Latent Semantic Indexing? http://en.wikipedia.org/wiki/Latent_semantic_indexing
toolslive on HNKind of. LSI is a dimensionality-reduction method, which means it takes really sparse "bag of words" vectors and compresses them in a way which approximates the original…
dhammack on HN