Relevancy of Lucene search results

Question

Hi, I've following 3 Documents in Lucene index.

1. As MBA you will play an integral role in implementing the strategy of the business and will have the responsibilities of the statutory accounts, compliance, audit including banking relationships, tax, treasury & cash management

2. As M.B.A. you will play an integral role in implementing the strategy of the business and will have the responsibilities of the statutory accounts, compliance, audit including banking relationships, tax, treasury & cash management

3. As Master of Business Administration you will play an integral role in implementing the strategy of the business and will have the responsibilities of the statutory accounts, compliance, audit including banking relationships, tax, treasury & cash management

My search input is :MBA and the query I search execute on Lucene is:

+((description:mba^3.0) (description:m.b.a.) (description:\"master business administration\"))

I get results in following sequence after sorting results by score in descending order:

Document # 3
Document # 2
Document # 1

Shouldn't Record # 1 come on top of search results since I've given it a higher boost and also that document contains the exact word MBA??

What am i missing here?

Thanks.

Answer 1

The matching query string makes up about 10% of the content of Doc#3. but only a tiny fraction of #1 and #2.

You might have to adjust your boosts to reflect the different lengths of the alternative query strings.

Answer 2

If you are using Lucene's StandardAnalyzer, docs #1 and #2 are actually equivalent and both will match the term "mba". It's hard to guess why #3 has the highest score - maybe because it matches multiple terms. You may want to consider handling synonyms like this at index time.

I wouldn't guess the field length would be a big factor, but what you probably want to do is use IndexSearcher.Explain() to get a breakdown of the scoring - that the best way to debug problems like this one.