Re: LDA Topic Model and Topic Maps

Hi Antonisamy,

> Is there a way to derive Topic
> Maps<http://en.wikipedia.org/wiki/Topic_Maps> from
> an LDA Topic
> model<http://en.wikipedia.org/wiki/Latent_Dirichlet_allocation>
> ?

As Steve N. and Patrick have already said: Yes, it is.

Maybe you're interested in some hands-on experience: I used LSA [1]
for a project, but the problems should be similar to LDA. I used LSA
to create a term-document-matrix from a corpus. The result may be
similar to the following example:

     {
       "cat": [doc-1, doc-2, doc-3],
       "dog": [doc-2, doc-3, doc-4, doc-5]
     }

To map the result into a topic map, you have to translate the strings
"cat" and "dog" into subject proxies (my documents are represented by
subject proxies already).

The Topic Maps Data Model [2] enforces IRIs to identify subjects
(other Topic Maps models like TMRM [3] don't enforce that, but I think
the problem of translating a string into something more meaningful
would be the same).

In the particular project I was only interested in persons and
countries, so I used Wikipedia or DBpedia to find the subject IRIs for
my terms which works to some extent. A lot of strings don't have an
obvious subject IRI, so I've had to map some strings to subject
proxies by hand.

I think the essence of the problem is mapping a term (a string) which
was found by an algorithm to a subject (proxy) as already pointed out
by Steve N.

Limiting your terms to a specific domain (as I did) may provide usable
results, though.

[1]
<https://secure.wikimedia.org/wikipedia/en/wiki/Latent_semantic_analysis>
[2] <http://www.isotopicmaps.org/sam/sam-model/>
[3] <http://www.isotopicmaps.org/tmrm/>


Best regards,
Lars
-- 
Semagia 
<http://www.semagia.com/>

<http://www.topicmaps.de/mailinglist/> German Topic Maps mailing list
<http://tinytim.sourceforge.net/> Open Source Topic Maps engine
<http://mappa.semagia.com/> Mappa - Python Topic Maps engine