An Efficient Algorithm for Computing k-Average-Regret Minimizing Sets in Databases

Abstract

Returning a small set of data points instead of the whole dataset to a user is a major task of a database system which has been studied extensively in recent years. In this paper, we study k-average-regret query, a recently proposed query, which uses “average regret ratio" as a metric to measure users’ satisfaction to avoid the biases towards a few dissatisfied users that the best-known k-regret query suffers from. The main challenge of executing a k-average-regret query is the low efficiency of existing algorithms. Fortunately, as the average regret function exhibits the properties of supermodularity and monotonictity, the computational complexity of k-average-regret query can be significantly reduced exploiting lazy evaluations, thus leading to our accelerated algorithm which we called Lazy-Greedy. Experiments on both synthetic and real datasets confirm the efficiency and quality of output of our proposed algorithm.