Calculating Query Result Overlap

From Icbwiki

Jump to: navigation, search

We often want to know how similar or different two document results are (when searching with the same topic, but with different methods). This page lists some methods to quantify such an overlap.


L1 and L2 are lists of query results. Each list contains items Ii with a document number and rank.

Absolute Overlap at Rank

We define the absolute overlap at rank r as the set of items that occur both in L_1 and L_2 and has rank less or equal than r. We denote this set O(r,L1,L2). We define O(L1,L2) the overlap at the maximum rank visible in the run.

Overlap plots

We can plot the rank in one dimension and the overlap O(r,L1,L2) in the other dimension. The profile should indicate in which rank regions the runs most overlap.

Personal tools