Hello,

I'm curious to know how to make an inverted index.

Thanks.

White papers at
http://www.n3labs.com/pdf/putz91usin...rted-DBMS.html
and
http://staff.science.uva.nl/~christo...clean-w-05.pdf (PDF)

One way to simulate one is with three tables (built one (long time ago) for an Intranet text document search)

document;
id
doc_location (URL or UNC path to file)
content (if document caching needed)

word_list;
id
word

word_pos;
id
word_id (word_list.id)
doc_id (document.id)
keyword (boolean for if the word is to be included as a keyword for the document)
position (location of word in document if ranking/sorting based on position)
tag (this can be used if you are using a ranking algo for HTML based on tag/structure weighting)

So, really what its saying is, its not the table structure that has to change but the data that goes into it?

the tables and how they are structured isn't really important, and the data isn't really a concern. It's all about the data is indexed into the structure

In my example structure, the inverted index would be built as the word_pos table and would be the word_id, doc_id and the position.

A simple (relatively speaking ) explanation;

from http://www.nist.gov/dads/HTML/invertedIndex.html

Note: Suppose we want to search the texts "i love you," "god is love," "love is blind," and "blind justice." (The words of the text are all lower case for simplicity.) If we index by (text, character within the text), the index with location in text is:

blind (3,8);(4,0)
god (2,0)
i (1,0)
is (2,4);(3,5)
justice (4,6)
love (1,2);(2,7);(3,0)
you (1,7)

The word "blind" is in document 3 ("love is blind") starting at character 8, so has an entry (3,8). To find, for instance, documents with both "is" and "love," first look up the words in the index, then find the intersection of the texts in each list. In this case, documents 2 and 3 have both words. We can quickly find documents where the words appear close to each other by comparing the character within the text.

You can see from there how simple it is to query the word_pos table and retrieve all documents pertaining to a word. If you use tag weighting you also has the basis of a simple ranking algorithm.

eg;
you SUM the occurences of that word within any given document = document_rank_weight

then if;
the word was found in the title, add extra rank weight
the word was found in bold, headings, italics etc add extra rank weight for each

then of course if you have an index where the word is used in links that point to a particular document from other locations you can use that to up the document ranking weight as well.

This is of course a very simple ranking method and one that is easily subverted, so don't think you can take on Google with it

They all say how it works but thats not what i'm after, i'm after how to do it.

What programming skills do you have?
What context do you want to use it (webclient, thin client, desktop app)
Are you considering a full inverted index (monolithic structure) or a simulated one using a RDBMS ?

All the information you need is around to build one, it's simply how you interpret it for your needs. There is no "correct" way to build the structure and query system, it will be purely up to how you need it to work.
Bearing in mind it is not a task for a newbie programmer or a semi skilled chancer who thinks they know databases.
get it wrong and it will perform like a "three legged pig"

I have intermediate programming skills although my friend who helps me is an expert, the context is a search engine the main purpose of it, no idea about the 3rd one, maybe a full inverted index?