: Interesting, omitTf=true eventhough it would give strict enforcement, : wouldnt it affect the relevancy? Like, I am wondering if the ordering : amongst the three word matches would be not as good as it would be when we : have omitNorms=true&omitTf=true. Do you have an idea?
It will *absolutely* affect the ranking ... that's the entire point. if the complaint is "docA containig only two of the clauses scores higher then docB matching all 3 clauses" the reason for that is (usually) because tf/idf scoring for docA is a *REALLY* good match for those two clauses (ie: they occur many, many times) where as docB might match all three but it may only match each of them once. you can't garuntee a strict ordering based on number of clauses that match unless you eliminate term freq and norms from the equation. That said, i realize now that i forgot to finish my previous message with the "However..." comment... However... if you still want the tf/idf and length norm to be a factor, but you just want to change the "penalty" of not matching all terms to be much higher (which doesn't garuntee a strict ordering, but biases things so much it's unlikely to ever be a factor) you could also play arround with a a custom implemntation of the coord factor in the similarity... http://lucene.apache.org/core/3_6_0/api/all/org/apache/lucene/search/Similarity.html#coord%28int,%20int%29 : > : I want to have a strick enforcement that In case of a 3 word search, : > those : > : results that match all 3 term should be presented ahead of those that : > match : > : 2 terms when I set mm=2. : > : : > : I have seen quite some cases where, those results that match 2 out of 3 : > : words appear ahead of those matching all 3 words. : > : > which can happen because of tf/idf and length normalization. : > : > if you disable all of those things for hte fields you : > search on (omitNorms=true omitTf=true) you should see a strict ordering : > based on the number of matching clauses. -Hoss
