Sphinx Search and rank based on word position - sphinx

Question

Is it possible, using Sphinx Search, to have the weight of a result to be determined on the position of words in a list?
For example, if you have rows with a column containing the following text:
Row #1: "dog, bird, horse, cat"
Row #2: "dog, bird, cat"
and then perform a OR search using "dog | cat" I would like row #2 to rank higher than #1 because both "dog" and "cat" were found, but #2 has these two closer together than #1.
Hope this makes sense.
Thanks
Michael

Answer 1

You can do this by using field level ranking. Use "SPH_RANK_EXPR" as your ranker and look at the field level factor "min_hit_pos" to tell which word matched first.
All the information can be found at http://sphinxsearch.com/docs/manual-2.0.4.html#weighting
If you look closely at the SPH_RANK_SPH04 ranking algorithm below, it includes min_hit_pos, but only gives credit to rows where the matched word is the first word.
sum((4*lcs+2*(min_hit_pos==1)+exact_hit)*user_weight)*1000+bm25
What you can do is use the same algorithm but change "2*(min_hit_pos==1)" to be something like this:-
(101-IF(min_hit_pos<100,min_hit_pos,100))
A row will get an extra 100 weight if matched on the first word, 99 if matched on the second word and so on until the 100th word, after which no more weight is given.
You can play around with the values and include a multiplier to see if the results are any better.
Hope that helps. Let me know if you have any questions.

Answer 2

Have you tried SPH_RANK_PROXIMITY ranking mode?
Otherwise could be more explicit and do a query like - with SPH_RANK_WORDCOUNT
"dog cat"/1 | "dog cat"~10 | "dog cat"~8 | "dog cat"~6 | "dog cat"~4 | "dog cat"~3 | "dog cat"~2 | "dog cat"~1
or similar.