Contents lists available at ScienceDirect Computers, Environment and Urban Systems journal homepage: www.elsevier.com/locate/ceus A brute force method for spatially-enhanced multivariate facet analysis Anthony C. Robinson a, ⁎ , Sterling D. Quinn b a GeoVISTA Center, Department of Geography, The Pennsylvania State University, 302 Walker Building, University Park, PA 16802, USA b Department of Geography, Central Washington University, 301 Dean Hall, Ellensburg, WA 98926, USA ARTICLE INFO Keywords: Spatial analysis Geovisual analytics Faceted search Computational methods ABSTRACT Faceted search is a common approach for helping users query multivariate data. While the method is found widely in contemporary tools, so far there has been little exploration of its potential to incorporate a spatial perspective. In this article we extend multivariate faceted search through the application of a brute force computational process to reveal facet combinations that have spatially-interesting results. We explore the po- tential utility of spatially-enhanced facet combinations in case study analyses of multivariate spatial data from learners in a massive open online course and multivariate spatial data from restaurant inspections. Spatially- enhanced facet combinations improve on ordinary faceted search by helping analysts understand which com- binations have significant spatial footprints. We also show how this method can be integrated into a geovisual analytics system through a simple user interface. Finally, we draw on our case study analyses to highlight important challenges and opportunities for future research. 1. Introduction Approaches for exploring high-dimensional spatial data often invite an analyst to forage through geographic features and their attributes with the goal of constructing insights and knowledge about particular phenomena (Andrienko et al., 2007; Pirolli & Card, 2005; Sacha et al., 2014). This usually involves manipulating user interface (UI) compo- nents such as range sliders and dropdown menus to perform dynamic queries (Shneiderman, 1994). In contemporary website design these combinations of query parameters are often symbolized on the interface using a technique called faceted search. As a user makes selections to narrow items of interest across different categories, each categorical choice is represented as a facet that can be toggled on or off (Fig. 1). Faceted search alone is a simple and powerful way for users to de- velop and refine queries for simple tasks like finding products to buy, but in a multivariate analysis context with spatial data where the goal is to identify and evaluate interesting patterns, users are likely to miss inter- esting combinations as the method does not support an exhaustive ap- proach. For example, in a census data analysis context a user may string together a query that retrieves all counties that have population above a certain median age, an average income above the poverty level, at least one college or university, and more than two hospitals. In this case, the faceted search method by design leads analysts down increasingly narrow paths to an outcome, and it relies on users to self-recognize when they have narrowed criteria too far to make a meaningful discovery. In the present work we draw upon prior work to develop so-called multivariate faceted search, which does not assume the independent ca- tegorical hierarchy that is found in traditional faceted search (Ben- Yitzhak et al., 2008). Multivariate faceted search supports exploratory analysis of variables that may be correlated with one another. The ap- proach we present here attempts to help users avoid the problem of narrowing queries into empty sets by using computational methods to automatically extract patterns, and then employing interface cues to direct user attention toward those patterns. In this paper we describe a new approach that combines the simplicity of faceted queries with computational analysis to identify facet combinations that have sig- nificant spatial clusters. Specifically, we outline a method for batch computing faceted queries and feeding the most fruitful results into a LISA (local indicators of spatial association) analysis to determine which facet combinations result in interesting spatial patterns. Based on our review of the literature, we have found no prior work proposing similar approaches for spatially-enhancing facet analysis. The use of faceted search is widespread in spatial analysis applications, and our work pro- vides a novel approach that allows analysts to conduct multivariate ex- ploration based not only on attribute combinations, but also based on whether or not that pattern has any spatial significance. In case studies examining student engagement with a massive open online course (MOOC), and high-dimensional data on restaurant food safety violations in New York City, we show how our method can be used to help analysts explore facet combinations that have significant spatial patterns. In addition to extending the basic method of faceted search to in- corporate the spatial dimension explicitly, we believe this approach has https://doi.org/10.1016/j.compenvurbsys.2017.12.003 Received 14 June 2017; Received in revised form 18 December 2017; Accepted 24 December 2017 ⁎ Corresponding author. E-mail addresses: arobinson@psu.edu (A.C. Robinson), sterling.quinn@cwu.edu (S.D. Quinn). Computers, Environment and Urban Systems xxx (xxxx) xxx–xxx 0198-9715/ © 2017 Elsevier Ltd. All rights reserved. Please cite this article as: Robinson, A.C., Computers, Environment and Urban Systems (2017), https://doi.org/10.1016/j.compenvurbsys.2017.12.003