Re: LDA Topic Model and Topic Maps

Antony,

If I understand your question, and it's possible I that don't, I would 
answer it this way:

First of all, it's necessary to understand that every topic in a topic 
map presumably is a proxy for exactly one subject.  A topic constitutes 
an allegation by its author that the author was thinking about exactly 
one thing, and that that single thing is identifiable, in the author's 
chosen (and explicit) universe(s) of discourse, in one or more ways 
which are explicit in the reifying topic.

A topic map is an instance of a rhetoric in which it is assumed that 
everything that is mentioned as a topic (i.e., that is "reified" in the 
philosophical sense of that term), is being regarded by the author of 
the topic map as a distinct and unique thing.

It is quite possible for the distinct and unique thing that is the 
combination of the classes "cat" and "dog" to be reified as a topic, 
even though there is no name (at least in English) for that combination 
of classes.

I can easily imagine the relationship between LDA and a given topic 
(i.e., a given reifier of a single unique subject) to be one in which 
LDA was used as a tool by the topic's author to arrive at the notion 
that the author then chose to reify as a topic.  It is not necessary 
that there be a pre-existing universe of discourse in which that 
combination of classes already has a name, nor is it necessary that a 
"name" (at least in the usual sense of the term "name") be used to 
uniquely identify it.  It may be very useful, albeit perhaps quite 
tricky, to identify a subject in terms other than a simple "name".  It 
may require the creation (and disclosure) of a new universe of 
discourse.   I can imagine using some expression that is meaningful only 
in terms of LDA to do it, given some universe of discourse in which LDA 
plays some key role.  Whatever a subject-identifying expression used in 
a subject proxy is, it must be independently useful, in some processing 
context, for the purpose of unequivocally answering the question:

    "Given topics (subject proxies) A and B, each of which identifies
    its subject via an expression that is processable in that same
    processing context, do the two proxies reify the same subject, or not?"

I hope that helps.  BTW, Ontopia doesn't support multiple arbitrary 
universes of discourse for subject identification, as discussed above, 
but the basic notion of topic mapping certainly does.  The relationship 
between Ontopia's "Topic Maps Data Model" and topic mapping is not 
unlike the relationship between an early version of Ingres and the 
relational model as originally described by C.J. Date.  There's much 
more in the Topic Maps paradigm than has been realized in any existing 
implementation, or in the "Topic Maps Data Model".  The Topic Maps Data 
Model is extremely useful, and Ontopia really works.  But as Yuri 
Rubinsky famously said about SGML (of which XML is a sub-species): "But 
wait, there's more!"

Steve Newcomb

P. S. I drafted ISO 13250-2000, which includes ISO 10744 by reference.  
I co-edited the latter, too.  If you don't grok the full generality of 
10744, you don't grok the full generality of 13250.  Not that it 
matters, really, but I don't think my authority to explain these 
standards is open to reasonable debate.

On 08/03/2011 03:41 PM, Antonisamy Arokiasamy wrote:
> Hello,
> 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>?
>
> I asked this question on the Topic Models mailing list and was soon 
> pointed to issues with hierarchies, direction, etc. And was also 
> introduced to this group - I am new to this group.
>
> Given that these two concepts dont have a one to one match, how can we 
> generate a user assisted Topic Map or even bootstrap a Topic Map from LDA.
>
> We have a LDA model that classifies documents and I am trying to see 
> how this will map to Topic Maps and Ontologies.
>
> Thanks
> /antony
>
>
>
> _______________________________________________
> topicmapmail mailing list
> topicmapmail-Zo64W7twoUFWk0Htik3J/[email protected]
> http://www.infoloom.com/mailman/listinfo/topicmapmail

_______________________________________________
topicmapmail mailing list
topicmapmail-Zo64W7twoUFWk0Htik3J/[email protected]
http://www.infoloom.com/mailman/listinfo/topicmapmail