Mysql query of inverted index data

问题:

I have thousand of pages in website which I parsed and stored it as Inverted Index viz

document

• docid (PK,FK)
• url
• charactercount
• wordcount

Charactercount and wordcount helps me determine long document from short which I may use later.

word

• wordid (PK,FK)
• word
• doc_freq
• inverse_doc_freq

For inverse_doc_freq calculation I use fictional high number (100000000) to prevent total document recalculation.

loc

• wordid
• docid
• word_freq
• weight

(wordid & docid combined unique)

The weight is a score calculated on simple basis like word in title + word in url + word frquency etc.

I am having problem framing my sql query for search words. For 3 word search I am doing like

1. Break query into each word
2. Check inverse_doc_freq for each word and remove low idf word (removal of stop word)
3. stem the remaining word (assume still 3 words remain)
4. Query for each word

It is at stage 4 that I am getting stuck ! My SQL query is like

SELECT d.docid,url,inverse_doc_freq,word_freq,weight from document d,word w,loc l WHERE d.docid=l.docid AND w.wordid=l.wordid AND (word='word1' OR word='word2' OR word='word3') ORDER BY weight DESC

The returned documents are not correct though. Trust I might have to Search thrice to find documents for each word and then try to find the common documents, but how ? Is it possible to use only 1 MySQL query for it ? Also is it possible to use TF-IDF and how ?

回答1:

You need to aggregate at the document level.

select d.docid, d.url, sum(weight) as weight
from document d join
     loc l
     on d.docid = l.docid join
     word w
     on w.wordid = l.wordid
where w.word in ('word1', 'word2', 'word3')
group by d.docid
order by weight DESC;