Interactive Set Discovery

TL;DR

This paper introduces an interactive approach for set discovery that minimizes user interactions by efficiently guiding the search for target sets using optimized decision trees and pruning strategies, validated through extensive experiments.

Contribution

It presents a novel algorithm for interactive set discovery that finds optimal or near-optimal decision trees with significant efficiency improvements over previous methods.

Findings

Pruning reduces search time by 2-5 orders of magnitude.

The algorithms effectively identify target sets with minimal user interactions.

Experimental results demonstrate superior efficiency and accuracy compared to prior approaches.

Abstract

We study the problem of set discovery where given a few example tuples of a desired set, we want to find the set in a collection of sets. A challenge is that the example tuples may not uniquely identify a set, and a large number of candidate sets may be returned. Our focus is on interactive exploration to set discovery where additional example tuples from the candidate sets are shown and the user either accepts or rejects them as members of the target set. The goal is to find the target set with the least number of user interactions. The problem can be cast as an optimization problem where we want to find a decision tree that can guide the search to the target set with the least number of questions to be answered by the user. We propose a general algorithm that is capable of reaching an optimal solution and two variations of it that strike a balance between the quality of a solution and the running time. We also propose a novel pruning strategy that safely reduces the search space without introducing false negatives. We evaluate the efficiency and the effectiveness of our algorithms through an extensive experimental study using both real and synthetic datasets and comparing them to previous approaches in the literature. We show that our pruning strategy reduces the running time of the search algorithms by 2-5 orders of magnitude.