Re: [port-peer-review] reviews
Phillipe, (01)
thank you for writing provocative reviews! I finished reading the
first one (LaBarre/Dent) and immediately felt compelled to respond:
It's a pity that I didn't finish a PORT contribution of my own this
year because it would have addressed some crucial points. I am
currently working on the distinction between what I call associative
and formal structures. Associative structures are fuzzy, probabilistic,
bottom-up, complex, depend on a local context etc. Formal structures
are precise, defined, top-down, use symbolic logic etc. WebKB is a
formal structure. Latent Semantic Indexing is associative. In short,
I believe that both need to be combined in knowledge representation
because, for example, human reasoning combines both. (This is explained
in more detail in my ICCS paper.) Latent Semantic Indexing is
certainly not perfect (and is definitely not "semantic") but it is
one of the better statistical tools. I am not sure how it compares
to other methods such as Kohonen maps. But because of the complexity
of the information on the WWW and the "fuzzy" components of human
reasoning, I don't think there is a way to build successful knowledge
representation tools, without incorporating some associative methods.
How will the users of WebKB map their own intuitive, somewhat
associative conceptual structures to the formal structures of the
system, without an associative interface? I guess that's where
topic maps come in. I understand LaBarre & Dent's paper as a similar
attempt at exploring the combination of associative
and formal methods. They chose faceted knowledge organization as
their representative formal method. But that is just one formal method
and could be replaced with FCA or WebKB or something else. Similarly
latent semantic indexing could be replaced with other kinds of
associative methods. The important point is that their work explores
a combination of associative and formal methods, which is a topic
that is often ignored. I see that as an important contribution to PORT. (02)
Uta (03)