Weighted Reciprocal Rank Fusion (WRRF) method for combining ranked lists.
You can use this endpoint to combine and ranked document items
from different systems or user preferences.
This API allows you to provide a list of ranked items and supporting
weights and scores so that you can get one list of ranked items
that caters to the user preferences.
Each list can have a weight. In addition, each item inside a list can have a score.
The position and score of each item is used to determinie the normalized
rank of the item. Items that come earlier in the list are given more weight,
and normalized after applying the additional score. A smoothing function
is applied to the position scores to ensure that the absolute position of items
does not skew the final ranking.
The final ranking is based on the combined score of each item and
the weight of the list it belongs to.
Item scores and list weights can be combined to formulate a
ranking that reflects the user's preferences.
For example, if one list comes from a search engine and another list
comes from human recommendation, you can assign higher weight to the
human recommendation list. If each item has a score that can be derived
from its attributes, e.g. date, it can be used to boost the
rank of certain types of items higher, despite their position in any list.