1. Read a document, published in
Answering top-K queries with multidimen1_selections: the ranking cube Approach
In a top-K query, two measurements reflect the performance: a selection condition & a ranking function.
The selection condition dimension may be very high, and the ranking function is not necessarily linear.
This document proposes a model called ranking cube and defines a rank-ware measure, which does not solve the curse of dimension and proposes ranking flagments.
Ii. Cube Structure
Build a ranking cube based on selection dimension. The measure of each cell should be rank-ware.
The most naive method is to put all the related tuple into the cell, so there are two shortcomings: waste of space, not rank-ware.
To save no space, you can only store tuple IDs.
To solve the rank-ware problem, two criteria are defined: geometry-based partition & Block-level data access.
So boring. Don't write it first