How to build a conceptual search engine?

I would like to build an internal search engine (I have a very large collection of thousands of XML files) that is able to map queries to concepts. For example, if I search for "big cats", I would want highly ranked results to return documents with "large cats" as well. But I may also be interested in having it return "huge animals", albeit at a much lower relevancy score.

I'm currently reading through the Natural Language Processing in Python book, and it seems WordNet has some word mappings that might prove useful, though I'm not sure how to integrate that into a search engine. Could I use Lucene to do this? How?

From further research, it seems "latent semantic analysis" is relevant to what I'm looking for but I'm not sure how to implement it.

Any advice on how to开发者_Python百科 get this done?

I'm not sure how to integrate that into a search engine. Could I use Lucene to do this? How?

Step 1. Stop.

Step 2. Get something to work.

Step 3. By then, you'll understand more about Python and Lucene and other tools and ways you might integrate them.

Don't start by trying to solve integration problems. Software can always be integrated. That's what an Operating System does. It integrates software. Sometimes you want "tighter" integration, but that's never the first problem to solve.

The first problem to solve is to get your search or concept thing or whatever it is to work as a dumb-old command-line application. Or pair of applications knit together by passing files around or knit together with OS pipes or something.

Later, you can try and figure out how to make the user experience seamless.

But don't start with integration and don't stall because of integration questions. Set integration aside and get something to work.

This is an incredibly hard problem and it can't be solved in a way that would always produce adequate results. I'd suggest to stick to some very simple principles instead so that the results are at least predictable. I think you need 2 things: some basic morphology engine plus a dictionary of synonyms.

Whenever a search query arrives, for each word you

1. Look for a literal match
2. "Normalize/canonicalze" the word using the morphology engine, i.e. make it singular, first form, etc and look for matches
3. Look for synonyms of the word

Then repeat for all combinations of the input words, i.e. "big cats", "big cat", "huge cats" huge cat" etc.

In fact, you need to store your index data in canonical form, too (singluar, first form etc) along with the literal form.

As for concepts, such as cats are also animals - this is where it gets tricky. It never really worked, because otherwise Google would have been returning conceptual matches already, but it's not doing that.

First, I agree with most of the advice here about starting slow, and first building bits and pieces of this grand plan, devising a minimal first product and continuing from there. Second, if you want to use some Wordnet functionality with Lucene, there is a contrib package for interfacing Lucene queries with Wordnet. I have no idea whether it was ported over to pylucene. Good luck and be careful out there.

First , write a piece of python code that will return you pineapple , orange , papaya when you input apple. By focusing on "is" relation of semantic network. Then continue with has a relationship and so on.

I think at the end , you might get a fairly sufficient piece of code for a school project.