Comparing Similarity Perception in Time Series Visualizations

Abstract

A common challenge faced by many domain experts working with time series data is how to identify and compare similar patterns. This operation is fundamental in high-level tasks, such as detecting recurring phenomena or creating clusters of similar temporal sequences. While automatic measures exist to compute time series similarity, human intervention is often required to visually inspect these automatically generated results. The visualization literature has examined similarity perception and its relation to automatic similarity measures for line charts, but has not yet considered if alternative visual representations, such as horizon graphs and colorfields, alter this perception. Motivated by how neuroscientists evaluate epileptiform patterns, we conducted two experiments that study how these three visualization techniques affect similarity perception in EEG signals. We seek to understand if the time series results returned from automatic similarity measures are perceived in a similar manner, irrespective of the visualization technique; and if what people perceive as similar with each visualization aligns with different automatic measures and their similarity constraints. Our findings indicate that horizon graphs align with similarity measures that allow local variations in temporal position or speed (i.e., dynamic time warping) more than the two other techniques. On the other hand, horizon graphs do not align with measures that are insensitive to amplitude and y-offset scaling (i.e., measures based on z-normalization), but the inverse seems to be the case for line charts and colorfields. Overall, our work indicates that the choice of visualization affects what temporal patterns we consider as similar, i.e., the notion of similarity in time series is not visualization independent.

Visual‐Interactive Preprocessing of Multivariate Time Series Data

“…Small multiples visualizations juxtapose individual dimensions to a list of charts with a shared x‐axis (time). The value domains are either represented with position encodings (line charts, area charts, bar charts, symmetric area charts) [EMJ10, AMST11, CLKS19, LYK∗12, CLKS19] or encoded with color, opacity, shape, or stroke size (heat map approaches) [GTPB19, CLKS19], or both position and color in case of the horizon graph [Rei08]. The vertical display space often limits the number of dimensions that can be shown simultaneously [EMJ10].…”

TimeCluster: dimension reduction applied to temporal data for visual analytics

“…Different approaches have been developed to extract useful information from raw time-series data including datamining. In many situations, however, automated techniques do not achieve satisfactory results, so experts rely on visual analytics tools to perform their tasks [17]. Visual analytics [23] combines the strengths of machine capabilities with human capabilities to facilitate exploration, analysis, understanding, and providing insights.…”

“…They have focused on elementary visual tasks that evaluate estimation, such as, point comparison and discrimination tasks, or estimation of averages. Thus, the results say very little about how the users assess the similarity of two or more time-series when utilizing various time-series visualizations [17]. Such tasks usually involve the notion of similarity between time-series which is sometimes inefficient [31].…”

Report on the First and Second Interdisciplinary Time Series Analysis Workshop (ITISA)

“…Scalable visualization solutions are direly needed, especially in support of progressive analytics [9,24]. At the same time, even some basic problems, such as the interplay between visual perception and similarity measures [8], deserves to be studied in more detail. Evidently, in order to be used in practice, all the above components should be combined in general, easy-to-use by non-experts, time series management systems [6,11], a task that is by itself a challenge.…”

“…For example, Granger causality 7. Neuroscientists for instance are interested in correlations among signals recorded by sensors that are spatially close to one another 8. Based on metadata, time intervals, value thresholds 9.…”