User-driven refinement of imprecise queries

Abstract

Abstract-We propose techniques for exploratory search in large databases. The goal is to provide new functionality that aids users in homing in on the right query conditions to find what they are looking for. Query refinement proceeds interactively by repeatedly consulting the user to manage query conditions. This process is characterized by three key challenges: (1) dealing with incomplete and imprecise user input, (2) keeping user effort low, and (3) guaranteeing interactive system response time. We address the first two challenges with a probability-based framework that guides the user to the most important query conditions. To recover from input errors, we introduce the notion of sensitivity and propose efficient algorithms for identifying the most sensitive user input, i.e., those inputs that had the greatest influence on the query results. For the third challenge, we develop techniques that can deliver estimates of the required probabilities within a given hard realtime limit and are able to adapt automatically as the interactive query refinement proceeds.

Overview of Data Exploration Techniques

“…For example, in [13] the authors infer join queries via labeling relevant objects, while in [58] and [51] the focus is on discovering queries given example output tuples. We also cover recent work on query learning based on example tuples [3] as well as solutions for tuning imprecise queries, where the relevance of query predicates is uncertain to the user [52]. Finally, we cover techniques for recommending SQL queries [21] as well as systems that identify the "best" segmentations of the data space to propose to the user [57].…”

Using preference-enriched faceted search for species identification

“…PFS has been proven useful for recall-oriented information needs, because such needs involve decision-making that can benefit from the gradual interaction and expression of preferences (Papadakos and Tzitzikas, 2015). A relevant work to PFS approach is discussed in Qarabaqi and Riedewald (2014). That work ranks the results based on a probabilistic framework that does not consider explicit users' preferences and assumes a data model that, on contrast to PFS, does not exploit hierarchically organised and/or set-valued attributes.…”