Re: [infomap-nlp-users] Document vectors

[Menno v. Z. <M.M...@uv...>]

Dear Beate,

* Beate Dorow <do...@IM...> wrote on [2004-05-05 18:13]:
> Are looking for a way of using a document (or combination of
> documents) as query to look for related documents or words ?
> If so, we have an implementation of this feature in a modified version of
> the infomap-nlp code which we can add on to the package.
> It works similar to associate, e.g.
> "associate_doc -d doc1 doc2 ... docN NOT docN+1 ... docN+k"
> will return documents which are similar to doc1 ... docN but are unrelated
> to docN+1 ... docN+k.
> 
> Or are you just looking for a way of retrieving the vector associated with
> a certain document?

I'm actually building a more complex system where I would like to use
the document feature vectors in combination with a lot of other
features and compare these ``extended'' feature vectores (by computing
distances).

What I've done now is changing the neighbors.c file so that when
computing distances, I simply print out the document vectors. Next,
I've changed associate.c to print out the document names (once
computing the distances is done). (Oh, I've also changed some
constants to print out information of all documents.) This gives me a
file like:
<feature vector>
<feature vector>
<feature vector>
...
<document name>
<document name>
<document name>
...

I analyse this file to get the right feature vector of the
corresponding document and I'm done.

There must be an easier way, but I think not many people will be
interested in the raw document vectors (or am I wrong)?

Best regards,

Menno

-------------------------------
- Menno van Zaanen            - They can't stop us,
- mvz...@uv...             - we're on a mission from God!
- http://ilk.uvt.nl/~mvzaanen -          -The Blues Brothers 
-------------------------------