The Challenges of Specifying Intervals and Absences in Temporal Queries: A Graphical Language Approach Megan Monroe Department of CS & HCIL, University of Maryland madeyjay@umd.edu Rongjian Lan Department of CS & HCIL, University of Maryland rjlan@cs.umd.edu Juan Morales del Olmo HCIL & Universidad Polit´ ecnica de Madrid juanmoralesdelolmo@gmail.com Ben Shneiderman Department of CS & HCIL, University of Maryland ben@cs.umd.edu Catherine Plaisant Department of CS & HCIL, University of Maryland plaisant@cs.umd.edu Jeff Millstein Oracle jeff.millstein@oracle.com ABSTRACT In our burgeoning world of ubiquitous sensors and afford- able data storage, records of timestamped events are being produced across nearly every domain of personal and profes- sional computing. The resulting data surge has created an overarching need to search these records for meaningful pat- terns of events. This paper reports on a two-part user study, as well as a series of early tests and interviews with clinical researchers, that informed the development of two temporal query interfaces: a basic, menu-based interface and an ad- vanced, graphic-based interface. While the scope of temporal query is very broad, this work focuses on two particularly complex and critical facets of temporal event sequences: in- tervals (events with both a start time and an end time), and the absence of an event. We describe how users encounter a common set of difficulties when specifying such queries, and propose solutions to help overcome them. Finally, we re- port on two case studies with epidemiologists at the US Army Pharmacovigilance Center, illustrating how both query inter- faces were used to study patterns of drug use. Author Keywords Query languages; temporal query; event sequences; query interfaces; electronic health records. ACM Classification Keywords H.5.m. Information Interfaces and Presentation (e.g. HCI) INTRODUCTION In 1992, a small class of elementary school students (includ- ing one of the authors of this paper) was given a simple writ- ing assignment: describe how to make a peanut butter and Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. To copy otherwise, or republish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee. CHI 2013, April 27–May 2, 2013, Paris, France. Copyright 2013 ACM 978-1-4503-1899-0/13/04...$15.00. jelly sandwich. After a short period of fervent scribbling, pa- pers were handed in to a teacher, who sat at the front of the room, surrounded by every tool and ingredient that one might need to construct the aforementioned confection. Then, one by one, the teacher acted out each set of instructions with the strictest adherence, and with total disregard for disaster or absurdity. Amidst the cackling laughter of the delighted stu- dents, peanut butter was spread onto walls, jelly onto desks, and bread was pressed into the carpet. The lesson: complex- ity can mask itself within a seemingly simple concept. Nearly 20 years later, we found ourselves in the same position as we watched users specify temporal queries: they would be- gin with an idea that was so simple and purposeful, and end up with metaphorical peanut butter smeared onto metaphori- cal walls. The difficulty with questions involving temporal event se- quences is not necessarily understanding the underlying com- plexity, but articulating that understanding into meaningful queries. Users assume that the simplicity with which they perceive these events, will translate just as easily to the un- derlying search application. For queries involving patterns of point events (events that occur at a single point in time), this assumption typically holds true. However, such simple events do not adequately cover the complete range of both medical and real world phenomena. Our main partners, epidemiolo- gists at the US Army Pharmacovigilance Center (PVC), are primarily responsible for conducting drug related studies in- volving prescription administration and medication interac- tion. Their data and their inquiries, are inherently interval- based. For example, they might need to know when two med- ications are being taken at the same time. Additionally, they frequently explore questions in which the absence of an event is the critical point of interest. For example, they might be looking for patients who did not experience a symptom after receiving a medication. When these types of questions arise, user strategies for speci- fying queries tend to remain tethered to the simplistic logic of point events, despite the increase in the underlying complex- ity. The result is queries that fail to match the users’ intention. Session: Data Navigation CHI 2013: Changing Perspectives, Paris, France 2349