On Thu, 07 Oct 2010, Roman Chyla wrote:
> It was trading space for speed, why that is not practical? Especially
> if the values are data points

Not practical, because 2nd order operators can operate on anything, like
keyword.  So you don't manage with storing `citing_author' only.

> could you point me to some info about how this 2nd order lookup is
> done, or where int the code, or discuss it here?

Concerning refersto/citedby operators, see 13910a4.

> if somebody could point out reasons why they are not possible or very
> difficult to implement with the search engine index, that would help
> me to understand much more.

They can surely be implemented with Solr/Lucene.  I have implemented
refersto/citedby for Invenio in something like two hours one evening.
(Plus something like an hour in a few weeks later, to add documentation
and test cases.)  It may be more or less complex for Solr/Lucene, but it
is surely doable.

Let me recall that we have many add-on applications of this kind,
e.g. cite summary, co-cited with, download similarity, etc.  It is not
only one commit.  Porting these to Solr is also theoretically possible.
After all, we could always simply rewrite our algorithms in Java and
change our data storage layer from DB to Solr.  But... why?  What
problem are we trying to solve?  We are in a situation where we have a
practical system to run and maintain, so we should be presumably more
interested in practical musings like: system S1 provides features F1..FN
that satisfy N% of user requirements and bring H% of happiness to their
lifes, at the price of X franks.  If we make extension E', we also
satisfy F', raising N and H levels to N' and H' in such and such a way,
at the price of X' franks.  If we make E'' then... if we switch to
system S2 then... if we switch to S3 then... You know what I mean. :)

I think we'd better start by defining Fi and Ni and Hi with the current
system, before thinking too hard about whether porting some obscure
feature to some other system is theoretically possible and at what
practical cost.

Best regards
--
Tibor Simko

Reply via email to