How to implement custom search result ranking?

Question

My lucene index contains documents with the field "itemName". This field is boosted with a boost factor between 0 and 1. When i create a BooleanQuery i'd like that the results are ranked by the count of matched clauses and the boostfactor, so the formula looks like:

score = (count_of_matching_clauses / count_of_total_clauses + boost_factor) / 2

The score would always be a float between 0 and 1. 1 in case all clauses match and the boost factor is 1.

For example, if the field value of "itemName" for three documents with no boost factor are:

document1: "java is an island"
document2: "the secret of monkey island"
document3: "java island adventures"

and the BooleanQuery would look like:

TermQuery query1 = new TermQuery(new Term("name","java"));
TermQuery query2 = new TermQuery(new Term("name","island"));

BooleanQuery query = new BooleanQuery();
query.add(query1, BooleanClause.Occur.SHOULD);
query.add(query2, BooleanClause.Occur.SHOULD);

than document1 would be retrieved with a score of (2/2 +0)/2 = 0.5 because: count_of_matching_clauses = 2 and count_of_total_clauses = 2

document2 would be retrieved with a score of (1/2+0)/2 = 0.25 because: count_of_matching_clauses = 1 and count_of_total_clauses = 2

than document3 would be retrieved with a score of (2/2 +0)/2 = 0.5 because: count_of_matching_clauses = 2 and count_of_total_clauses = 2

How to implement this ranking mechnism in lucene? How can i tell lucene to use my custom ranking class for ranking the results?

Answer 1

You can implement your own scoring algorithm by extending Similarity class and passing it during search. In the Javadoc of this class (follow the link), you can read the details of the scoring algorithm. Some more text on scoring can be found here. An exceptional aid to understand scoring is to actually see the explanation for the scoring as returned by Searcher.explain()

BTW, the scoring you wish to implement is the default scoring. The order of results will be as desired, though actual scores can be different than 0.5 or 0.25.