Interactive Search for One of the Top-k

Abstract

When a large dataset is given, it is not desirable for a user to read all tuples one-by-one in the whole dataset to find satisfied tuples. The traditional top-k query finds the best k tuples (i.e., the top-k tuples) w.r.t. the user's preference. However, in practice, it is difficult for a user to specify his/her preference explicitly. We study how to enhance the top-k query with user interaction. Specifically, we ask a user several questions, each of which consists of two tuples and asks the user to indicate which one s/he prefers. Based on the feedback, the user's preference is learned implicitly and one of the top-k tuples w.r.t. the learned preference is returned. Here, instead of directly following the top-k query to return all the top-k tuples, since it requires heavy user effort during the interaction (e.g., answering many questions), we reduce the output size to strike for a trade-off between the user effort and the output size. To achieve this, we present an algorithm 2D-PI which asks an asymptotically optimal number of questions in a 2-dimensional space, and two algorithms HD-PI and RH with provable performance guarantee in a d-dimensional space (d >= 2), where they focus on the number of questions asked and the execution time, respectively. Experiments were conducted on synthetic and real datasets, showing that our algorithms outperform existing ones by asking fewer questions within less time to return satisfied tuples.