REVIEW 4 major objections 5 minor 23 references
Analyzing Time Attributes in Temporal Event Sequences
T0 review · 4 major / 5 minor · reviewed 2026-08-14 · deepseek-v4-flash
Pith's one-line read The paper claims that a single overview, built from an event-box encoding of duration and time of occurrence, lets analysts identify trends and outliers in event sequences that standard visualizations cannot show.
desk verdict A well-designed event-box encoding for duration and time of occurrence in event sequences, but the paper's central claim that it reveals what 'otherwise could not be obtained' is not supported by any validation. read the letter →
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
The reading
What carries the argument
The load-bearing mechanism is the event box, a boxplot-inspired glyph that aggregates one event type's occurrences: height is proportional to record count, width to the duration range, and interior colored sub-boxes mark the quartiles of the duration distribution. Each data point sits inside the box at coordinates (duration, time of occurrence), so the figure simultaneously shows the distribution of how long events last and when they happen. A standard boxplot outlier rule using 1.5 times the interquartile range marks unusual durations as points outside the colored bands. What this element does is carry the paper's main promise: it replaces the traditional average-duration bar with a full distribution, so individual outliers and time-of-day trends become visible in situ rather than requiring a separate view.
What would settle it
A controlled reading test would settle the claim: take the same outpatient event dataset, ask experienced analysts to find duration outliers and time-of-occurrence trends using the proposed overview versus a conventional sequence overview, and check whether the proposed method yields more correct findings or faster answers. A statistical check would also do: compute the raw distributions behind the case-study findings, for example late-arrival durations by hour of day, and test whether the claimed decreasing trend and morning concentration are significant in the source data.
Extended reading notes
Core claim
The central discovery is an event-level visual encoding that aggregates all occurrences of an event type into a single 'event box'. The box's width is proportional to the full range of observed durations, its interior is split into quartile bands, and each occurrence is a point placed horizontally by its duration and vertically by its time of occurrence. The encoding can be collapsed to a point event or expanded through several levels of detail, and time-of-occurrence can be re-scaled from hours to days, weeks or months, with color adding another time scale. Embedded in the overview and combined with user-driven alignment by multiple events and similarity-based sorting of unique sequences, this encoding lets a user compare duration and timing trends both within and across sequences. The paper reports that, on a real-world hospital outpatient dataset, the approach led to four classes of findings that, in the authors' judgment, could not be obtained with traditional visualizations.
Load-bearing premise
The central claim that these findings could not be obtained with traditional visualizations depends on an untested assumption: that a user looking at the event-box layout will read the plotted points and quartile bands correctly and that the patterns highlighted in the case study are genuine properties of the data rather than artifacts of the chosen display.
Editorial extensions
If this is right
- A user can inspect the distribution of event durations and times of occurrence without leaving the sequence overview, reducing the need for separate secondary views.
- Duration outliers become visible as points outside the quartile bands, so analysts can identify unusual individual cases rather than relying on averages.
- Aligning sequences by multiple user-selected events and sorting by similarity exposes how the presence or absence of an early event relates to the duration of later events.
- The same overview can be filtered by specific days of the week and broken down by day, revealing temporal patterns such as shorter consultations on certain days or a morning concentration of visits.
- The four-class taxonomy of findings (main sequential patterns, impact of an event on later durations, duration trends over time, and temporal distribution of time attributes) is offered as a generalizable scheme for analyzing event data in other domains.
Reading between the lines
- If the event-box encoding is perceptually sound, the same visual grammar could be applied to any interval-valued event attribute beyond time, such as cost or resource usage, turning the overview into a general multivariate sequence display.
- The authors' most ambitious claim, that findings could not be obtained with traditional visualizations, is an empirical statement; a direct comparison with existing sequence tools would convert it from a claim into a measured result.
- Because the method sorts unique sequences by edit distance and then aligns by multiple events, it effectively builds an approximate multiple sequence alignment; borrowing richer alignment cost models might yield better grouping of clinically similar pathways.
- The claimed taxonomy classes could be turned into automated pattern-detection routines, such as statistical tests for duration shifts between sequence groups, making the visual exploration semi-automatic.
Signed reviews
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper proposes a visual analytics methodology for temporal event sequences, centered on a 'Sequential and Time Patterns overview' that embeds a boxplot-inspired 'event box' encoding of event duration (horizontal) and time of occurrence (vertical). The pipeline selects a set of frequent unique sequences, sorts them by similarity using complete-link hierarchical clustering with Levenshtein edit distance, aligns them by user-selected events, and renders each event as an interactive event box with multiple levels of detail. The authors demonstrate the approach on one year of Rheumatology outpatient clinic event logs and propose a four-class taxonomy of findings (main sequential patterns, impact of an event on subsequent event durations, duration trends with respect to time of occurrence, and temporal distribution of time attributes). The central claim is that the proposed visualization allows derivation of findings 'that otherwise could not be obtained using traditional visualizations' (Abstract; Sec. 6).
Significance. If substantiated, the event-box encoding and the integrated sequential/time overview would be a useful addition to event-sequence visualization, particularly for healthcare process data where waiting times and durations matter. The paper's strengths are its use of a real-world dataset, the explicit taxonomy of findings, and the interactive design with multiple alignment and levels of detail. However, the central value proposition—that the method yields insights unobtainable with existing tools—rests on two untested assumptions: that the visual encoding is readable by analysts, and that the reported findings are genuine properties of the data rather than artifacts of the display or of the unstated frequency threshold. The paper provides no user study, no comparison with existing tools (EventFlow, Outflow, LifeLines2, Eventpad, or even simple histograms/boxplots), and no statistical verification of the case-study findings. The contribution is therefore plausible but not established at its current level of evidence.
major comments (4)
- [Abstract; Sec. 6] The central claim that findings can be derived 'otherwise could not be obtained using traditional visualizations' is not validated. The paper provides no comparison against existing systems such as EventFlow [16], Outflow [22], LifeLines2 [21], or Eventpad [4], and no user study demonstrating that analysts can correctly decode the event-box layout. Because this claim is the headline contribution, it needs at least a baseline comparison or a controlled usability study to be supported.
- [Sec. 4.1; Sec. 5.2] The frequency threshold used to remove 'unique sequences with relatively low frequency' (step 1 in Sec. 4.1) is never specified, and no sensitivity analysis is reported. The case-study findings in Sec. 5.2 are derived from a display that depends on this threshold, so it is unclear whether the observed patterns (e.g., Finding 2 on consultation duration without Height and Weight, or Finding 3 on late-arrival duration decreasing through the day) are robust or artifacts of the chosen cutoff. The threshold value and a sensitivity analysis should be reported.
- [Sec. 5.2] The reported findings are asserted from visual inspection without statistical verification. For example, Finding 3 states that 'the duration of the late arrival event decreases as the day goes by,' but no correlation, regression, or group comparison is provided to confirm this trend or its significance. Similarly, Finding 6 ('consultation times are significantly shorter on Tuesdays') uses the word 'significantly' without a statistical test. Each quantitative claim in the taxonomy should be backed by an appropriate statistical check or explicitly labeled as an exploratory observation.
- [Sec. 3.1; Fig. 2a] The event-box encoding assumes that the vertical placement of data points according to time of occurrence and their horizontal placement according to duration can be read jointly to identify trends and outliers. However, no evidence is presented that this encoding is perceptually effective—e.g., whether users can distinguish quartile ranges from outlier points, or can compare time-of-occurrence distributions across event boxes. Given that the entire methodology rests on this encoding, a perceptual or empirical evaluation is needed.
minor comments (5)
- [Sec. 6] There is a typo in the phrase 'Sequential and Time patters overview' near the end of the first paragraph of Sec. 6; it should read 'Sequential and Time Patterns overview'.
- [References] Reference [2] lists the edition as '3nd edition'; this should be corrected to '3rd edition.'
- [Sec. 4.1] The clustering step is described as 'complete-link agglomerative hierarchical clustering [9] and the Levenshtein edit distance [12],' but details are missing: how the edit distance is computed for sequences of differing lengths, whether costs are unweighted, and how the dendrogram is cut to produce the displayed ordering. A short description or pseudocode would improve reproducibility.
- [Fig. 1] The figure is dense and the numbered findings (1–8) are referenced in the text but are not all easy to locate in the three panels. Adding a legend or enlarging the relevant regions would improve readability.
- [Sec. 5.2, Class II] The explanation of Class II findings would benefit from an explicit statement that the causality is only hypothesized: the current wording ('this finding should require further investigation') is appropriate, but the earlier 'could be related to' phrasing could be mistaken for an established result.
Circularity Check
No significant circularity: the visualization pipeline is self-contained and no prediction reduces to a fitted input.
full rationale
The paper proposes a visual encoding for duration and time of occurrence in event sequences, built from standard boxplot quartiles and Tukey outlier rules, then applies it in a case study. There is no derivation chain in which an output quantity is defined in terms of the quantity it claims to predict, and no parameter is fitted to a subset of data and then renamed as a prediction. The case-study findings are qualitative interpretations made by the authors from their own visualization, and the claim that these findings 'otherwise could not be obtained using traditional visualizations' is empirically unsupported because no baseline comparison or statistical validation is provided. Unsupported empirical claims are a correctness or evaluation gap, not circularity. The paper also does not rely on load-bearing self-citations: its references are to external prior work, and the methodological components (hierarchical clustering, Levenshtein distance, boxplots, multiple alignment) are standard and independently grounded. The only stated limitation, scalability in Sec. 6, does not reveal a circular step. Accordingly, the circularity score is 0.
Assumptions & free parameters
free parameters (1)
- Unique sequence frequency threshold =
not specified
assumptions (4)
- standard math Boxplot quartile definitions and the Tukey outlier rule (k=1.5) are valid summaries for event duration data.
- domain assumption Hospital staff entries in the workflow tracking system accurately reflect real event start and end times.
- ad hoc to paper Complete-link hierarchical clustering with Levenshtein edit distance produces a meaningful similarity ordering for event sequences.
- ad hoc to paper Removing low-frequency unique sequences keeps the patterns of interest while removing noise.
Cite this review
Pith. "Pith review of Analyzing Time Attributes in Temporal Event Sequences." pith.science (2026). https://pith.science/paper/BRZFI53V
@misc{pith2026190800903,
author = {Pith},
title = {Pith review of: Analyzing Time Attributes in Temporal Event Sequences},
year = {2026},
howpublished = {\url{https://pith.science/paper/BRZFI53V}},
note = {Machine review of arXiv:1908.00903}
}
read the original abstract
Event data is present in a variety of domains such as electronic health records, daily living activities and web clickstream records. Current visualization methods to explore event data focus on discovering sequential patterns but present limitations when studying time attributes in event sequences. Time attributes are especially important when studying waiting times or lengths of visit in patient flow analysis. We propose a visual analytics methodology that allows the identification of trends and outliers in respect of duration and time of occurrence in event sequences. The proposed method presents event data using a single Sequential and Time Patterns overview. User-driven alignment by multiple events, sorting by sequence similarity and a novel visual encoding of events allows the comparison of time trends across and within sequences. The proposed visualization allows the derivation of findings that otherwise could not be obtained using traditional visualizations. The proposed methodology has been applied to a real-world dataset provided by Sheffield Teaching Hospitals NHS Foundation Trust, for which four classes of conclusions were derived.
Figures
Reference graph
Works this paper leans on
- [16]
-
[22]
K. Wongsuphasawat and D. Gotz. Exploring flow, factors, and outcomes of temporal event sequences with the outflow visualiza- tion. IEEE Transactions on Visualization and Computer Graphics , 18(12):2659–2668, 2012
work page 2012
-
[21]
T. D. Wang, C. Plaisant, B. Shneiderman, N. Spring, D. Roseman, G. Marchand, V . Mukherjee, and M. Smith. Temporal summaries: Sup- porting temporal categorical searching, aggregation and comparison. IEEE transactions on visualization and computer graphics, 15(6):1049– 1056, 2009
work page 2009
-
[4]
B. C. Cappers and J. J. van Wijk. Exploring multivariate event se- quences using rules, aggregations, and selections. IEEE Transactions on Visualization & Computer Graphics, (1):1–1, 2018
work page 2018
-
[1]
M. Bardsley, A. Steventon, and G. Fothergill. Untapped potential: Investing in health and care data analytics. The Health Foundation, 2019
work page 2019
-
[2]
V . Barnett and T. Lewis. Outliers in statistical data. John Wiley & Sons Ltd., 3nd edition edition, 1994
work page 1994
-
[3]
R. J. C. Bose and W. van der Aalst. Trace alignment in process mining: opportunities for process diagnostics. In International Conference on Business Process Management, pages 227–242. Springer, 2010
work page 2010
-
[5]
Y . Chen, P. Xu, and L. Ren. Sequence synopsis: Optimize visual summary of temporal event data. IEEE transactions on visualization and computer graphics, 24(1):45–55, 2018
work page 2018
Show all 23 references
-
[6]
Frigge, D
M. Frigge, D. C. Hoaglin, and B. Iglewicz. Some implementations of the boxplot. The American Statistician, 43(1):50–54, 1989
1989
-
[7]
Gotz and H
D. Gotz and H. Stavropoulos. Decisionflow: Visual analytics for high-dimensional temporal event sequence data. IEEE transactions on visualization and computer graphics, 20(12):1783–1792, 2014
2014
-
[8]
S. Guo, K. Xu, R. Zhao, D. Gotz, H. Zha, and N. Cao. Eventthread: Visual summarization and stage analysis of event sequence data. IEEE transactions on visualization and computer graphics , 24(1):56–65, 2018
2018
-
[9]
J. A. Hartigan. Clustering algorithms. 1975
1975
-
[10]
B. C. Kwon, J. Verma, and A. Perer. Peekquence: Visual analytics for event sequence data. In ACM SIGKDD 2016 Workshop on Interactive Data Exploration and Analytics, volume 1, 2016
2016
-
[11]
H. Lam, D. Russell, D. Tang, and T. Munzner. Session viewer: Visual exploratory analysis of web session logs. 2007
2007
-
[12]
V . I. Levenshtein. Binary codes capable of correcting deletions, in- sertions, and reversals. In Soviet physics doklady, volume 10, pages 707–710, 1966
1966
-
[13]
Z. Liu, B. Kerr, M. Dontcheva, J. Grover, M. Hoffman, and A. Wilson. Coreflow: Extracting and visualizing branching patterns from event sequences. In Computer Graphics Forum, volume 36, pages 527–538. Wiley Online Library, 2017
2017
-
[14]
Z. Liu, Y . Wang, M. Dontcheva, M. Hoffman, S. Walker, and A. Wilson. Patterns and sequences: Interactive exploration of clickstreams to understand common visitor paths. IEEE Transactions on Visualization and Computer Graphics, 23(1):321–330, 2017
2017
-
[15]
M. H. Loorak, C. Perin, N. Kamal, M. Hill, and S. Carpendale. Times- pan: Using visualization to explore temporal multi-dimensional data of stroke patients. IEEE transactions on visualization and computer graphics, 22(1):409–418, 2016
2016
-
[17]
Monroe, K
M. Monroe, K. Wongsuphasawat, C. Plaisant, B. Shneiderman, J. Mill- stein, and S. Gold. Exploring point and interval event patterns: Display methods and interactive visual query.University of Maryland Technical Report, 2012
2012
-
[18]
Perer and F
A. Perer and F. Wang. Frequence: interactive mining and visualization of temporal frequent event sequences. In Proceedings of the 19th international conference on Intelligent User Interfaces, pages 153–162. ACM, 2014
2014
-
[19]
J. W. Tukey. Exploratory Data Analysis. Addison-Wesley, 1977
1977
-
[20]
Vrotsou, J
K. Vrotsou, J. Johansson, and M. Cooper. Activitree: Interactive visual exploration of sequences in event-based data using graph sim- ilarity. IEEE Transactions on Visualization and Computer Graphics, 15(6):945–952, 2009
2009
-
[23]
J. Zhao, Z. Liu, M. Dontcheva, A. Hertzmann, and A. Wilson. Ma- trixwave: Visual comparison of event sequence data. In Proceedings of the 33rd Annual ACM Conference on Human Factors in Computing Systems, pages 259–268. ACM, 2015. 5
2015
Reviewed August 14, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.