Ranking results using multiple nested ranking

A method that provides improved rankings for items returned for a given query comprising:

• employing a processor executing computer executable instructions embodied on a computer readable storage medium to perform the following acts:
• retrieving an initial ranked ordered set of items, wherein each item is assigned a relevance value;
• and
• iteratively re-ranking, conducted via at least one trained ranking algorithm, one or more subsets of the set of items, wherein each subset comprises a plurality of ranked items from the previous iteration that are not below a threshold relevance value, the number of items in the subset decreases at each iteration, wherein the threshold relevance value is automatically selected by artificial intelligence based upon the number of items in the set of items, wherein each iteration employs a ranking algorithm that is slower to rank items than a ranking algorithm employed to rank items in a previous iteration;
• and
• wherein the at least one trained ranking algorithm includes a learning algorithm that is given a set of pairs of items [xi,xj] with probability Pij that item xiis to be ranked higher than item xjand a ranking model f, where f(xi)>f(xj) indicates that the ranking model ranks xihigher than xj,
• P
• _
• ij
• ≡
• e
• o
• _
• ij
• 1
• +
• e
• o
• _
• ij
• ,
• where ranking model output ōij≡f(xi)−(xj).

A ranking system that provides improved rankings for items returned for a given query comprising:

• a processor;
• a memory communicatively coupled to the processor, the memory having stored therein computer-executable instructions to implement the system, including:
• a means for retrieving an initial ordered set of items for the given query;
• and
• a means for iteratively re-ranking, conducted via at least one trained ranking algorithm, one or more subsets of the set of items to facilitate obtaining the most relevant ordering of the items, wherein each subset comprises a plurality of higher ranked items from the previous iteration, the number of items in a subset decreases at each iteration, wherein each iteration employs a ranking algorithm that is slower to rank items than a ranking algorithm employed to rank items in a previous iteration, wherein the at least one trained ranking algorithm includes a learning algorithm that is given a set of pairs of items [xi, xj] with probability Pijthat item xiis to be ranked higher than item xjand a ranking model f, where f(xi)>f(xj) indicates that the ranking model ranks xihigher than xj,
• P
• _
• ij
• ≡
• e
• o
• _
• ij
• 1
• +
• e
• o
• _
• ij
• ,
• where ranking model output ōij≡f(xi)−(xj).