Source-linked AI summary
Survey on Visual Analysis of Event Sequence Data
Yi Guo, Shunan Guo, Zhuochen Jin, Smiti Kaul, David Gotz, Nan Cao
TL;DR
Event-sequence data are heterogeneous, and the proliferation of specialized visual-analysis methods makes the state of the art difficult to discern. This survey reviews and organizes the literature using a design space and task taxonomy, identifying research challenges and future opportunities.
Problem
Event sequences vary in structure and analytical goals, and the proliferation of specialized methods makes the state of the art difficult to discern.
Source: §1 · p. 1Method
The survey characterizes visual analytics literature through a design space and analytical tasks, organizing methods and identifying under-addressed areas.
Source: §1 · p. 1Results
The survey connects prior studies in a taxonomy of design elements and five analytical task categories, and identifies remaining challenges and future research directions.
Source: §6 · p. 16, §7 · pp. 17–18Takeaways & Limitations
The taxonomy offers practitioners an overview of alternative approaches and helps them find design components suited to their analytical tasks.
Source: §7 · p. 18Scaling across sequence summaries and low-level event details remains unresolved, since existing aggregation and filtering can hinder detailed inspection.
Source: §6 · p. 17
Abstract
from arXiv · showhide
Event sequence data record series of discrete events in the time order of occurrence. They are commonly observed in a variety of applications ranging from electronic health records to network logs, with the characteristics of large-scale, high-dimensional, and heterogeneous. This high complexity of event sequence data makes it difficult for analysts to manually explore and find patterns, resulting in ever-increasing needs for computational and perceptual aids from visual analytics techniques to extract and communicate insights from event sequence datasets. In this paper, we review the state-of-the-art visual analytics approaches, characterize them with our proposed design space, and categorize them based on analytical tasks and applications. From our review of relevant literature, we have also identified several remaining research challenges and future research opportunities.
1 INTRODUCTION
Event sequence data vary widely in their structure and analytical goals, and the resulting range of specialized methods makes the state of the art difficult to discern. This survey organizes prior work into a design space and task taxonomy, identifying common approaches and under-addressed challenges.
1 INTRODUCTION
- Event sequences are used across fields including computer security, advertising, and healthcare, where observations are recorded over time for devices, users, or patients.The authors connect their ubiquity to ease of capture and interest in gaining insights about real-world systems.Source: §1 · p. 1
- Event sequences vary in dimensionality, temporal density, event attributes, and whether events are sequential or parallel, while analysts pursue different goals across individual sequences or groups.Tasks may focus on common patterns, rare outliers, prediction, or predictive factors for intervention.Source: §1 · p. 1
- The proliferation of visual analysis methods with context-specific capabilities makes the state of the art difficult to discern, and some open problems remain unaddressed.A single method may address multiple challenges, while other methods target specific challenges.Source: §1 · p. 1
- The survey characterizes the literature using design-space dimensions and analysis tasks to organize methods and identify common approaches to targeted problems.It also identifies areas with little prior work that remain challenges for future research.Source: §1 · p. 1
- The review is presented as the first, to the authors’ knowledge, comprehensive survey of event-sequence visual analytics, intended to connect prior work and identify insufficiently addressed challenges.The authors also developed a web-based browser for exploring the taxonomy and reviewed techniques.Source: §1 · p. 1
2 RELATED SURVEYS AND METHODOLOGY
This survey takes a broader view of event-sequence visual analytics than reviews focused on particular visualization forms, tasks, or applications. It builds a taxonomy through a design space and a literature-selection process that yields 133 covered papers.
2.1 Related Surveys
- Unlike surveys focused on particular visualization forms, tasks, or applications, this work aims to cover visual analytics approaches for event-sequence data across application domains.The related work includes surveys of temporal data and application areas such as social media, smart manufacturing, and anomalous user behavior.Source: §2.1 · p. 2
2.2 Survey Methodology
- The design space follows data, model, and visualization components of the visual analytics pipeline, excludes subjective knowledge inference, and adds interaction as a dimension.Its four dimensions are data scales, analysis techniques, visual representations, and interactions.Source: §2.2 · p. 2
- Paper selection combined reference-driven tracing with broad searches of visualization and data-mining venues, followed by task- and application-specific searches.Collected papers were labeled by design dimensions, analytical tasks, and applications; the resulting nine tasks were organized into five categories and seven applications into three categories.Source: §2.2 · p. 2
- The survey covers 133 papers, including 100 representative and up-to-date event-sequence visualization and visual-analytics studies.The selection process identified 153 event-sequence analysis publications and 144 visualization or visual-analysis publications before refining the selection.Source: §2.2 · p. 2
3 TAXONOMY
The survey organizes event-sequence visual analytics through a design space and a set of analytical tasks, then uses that taxonomy to characterize methods and guide domain-specific exploration. Its framework spans data scales, analysis techniques, visual representations, interactions, and five task categories.
3 TAXONOMY
- The taxonomy organizes visual analytics techniques by primary analytical task and characterizes methods through design-space dimensions, while applications provide separate domain-specific guidance.Applications are excluded from the taxonomy because most reviewed techniques address general event-sequence analysis rather than a specific application.Source: §3 · pp. 2–3
3.1 Design Space
- The design space identifies frequently used data scales, analysis techniques, visual representations, and interactions for event-sequence analysis systems.The survey uses these dimensions to structure its characterization of existing methods.Source: §3.1 · p. 3, §3 · p. 3
3.1.1 Dimension 1: Data Scales
- Data scales range from individual events and subsequences to complete sequences and collections of sequences, matching analysis from local detail to group patterns.Collections support summarization and comparison, while complete sequences capture an entity’s full event record.Source: §3.1.1 · p. 3
3.1.2 Dimension 2: Analysis Techniques
- Analysis techniques include pattern discovery, sequence inference, and sequence modeling, with inference drawing conclusions from observed data under conditions that may not hold for unobserved data.Pattern discovery covers frequent-pattern mining and similarity analysis; inference includes point processes and graphical models, while modeling includes methods such as RNNs.Source: §3.1.2 · p. 3
3.1.3 Dimension 3: Visual Representations
- The survey groups visual representations into chart-, timeline-, hierarchy-, Sankey-, and matrix-based designs for displaying event features, sequences, structures, transitions, and frequencies.Sankey-based designs can use node-link graphs or width-encoded links for flow proportions.Source: §3.1.3 · p. 4
3.1.4 Dimension 4: Interactions
- Alignment, scaling, and emphasis support comparison around selected events or times, inspection at different granularities, and highlighting, ranking, or arranging visual elements.Alignment can reveal precursor, co-occurring, and aftereffect events; scaling includes both visualization-level zoom and data-level abstraction or elaboration.Source: §3.1.4 · p. 5
- Interaction techniques include aggregation, filtering and querying, editing, and segmentation to adjust or focus sequence data during exploration.Aggregation merges sequences; filters select data by event, time, attributes, or patterns; editing changes events, order, or duration; segmentation focuses on shorter sequence portions.Source: §3.1.4 · pp. 4–5
3.2 Visual Analysis Tasks
- The task taxonomy covers summarization, prediction and recommendation, anomaly detection, comparison, and causal analysis; visual-analytics work on causality remains very limited.Summarization seeks major patterns and groupings, while comparison examines similarities and differences across event sequences.Source: §3.2 · pp. 5–6
4 VISUAL ANALYSIS TECHNIQUES
The survey organizes visual-analysis techniques by task, spanning event-sequence summarization, prediction and recommendation, anomaly detection, comparison, and causality analysis.
4.1 Visual Summarization
- Summarization spans aggregated timelines, Sankey and hierarchy views, query- and mining-based pattern extraction, progression analysis, and clustering; explicit displays can grow messy, and frequent patterns may lack meaning.Source: §4.1 · pp. 6–7
4.2 Visual Event Prediction & Recommendation
- Prediction uses historical sequences to forecast events or outcomes, while recommendations suggest goal-directed actions; visualizations also expose model mechanisms because deep-learning interpretability is a challenge.Source: §4.2 · pp. 8–9
4.3 Visual Anomaly Detection
- Anomaly methods target anomalous events, frequent patterns, or sequences, using normal-progression comparisons, stage and group inspection, and temporal or glyph-based summaries.Source: §4.3 · pp. 9–10
4.4 Visual Comparison
- Comparison techniques address individual sequences, sequential patterns, and sequence collections, using juxtaposition, superposition, or explicit encodings to inspect differences.Source: §4.4 · pp. 10–11
4.5 Visual Causality Analysis
- Causality visualizations use graphs, animation, and causal semantics, but event-sequence analysis remains sparse and must handle temporal delays and durations, event hierarchies, and collection complexity.Source: §4.5 · p. 11
5 APPLICATIONS
The survey reviews event-sequence visual analytics in health informatics, internet applications, and Industry 4.0, organizing examples around their applications and analytical goals. Across these domains, techniques support cohort and outcome analysis, behavior exploration, system monitoring, anomaly analysis, and manufacturing-plan optimization.
5.1 Health Informatics
- Health informatics techniques support cohort summarization and comparison, outcome analysis, and prognosis using patient event sequences.Cohort methods summarize patterns or compare groups; outcome methods examine progression paths, while prognosis tools visualize predicted risk and let users edit sequences to explore alternatives.Source: §5.1 · pp. 12–13
- In prognosis analysis, RetainVis and CarePre let clinicians inspect predicted risks under edited patient event sequences or alternative treatments.RetainVis links risk trajectories with medical-event sequences; CarePre estimates influential treatments and allows event edits to explore different predicted risks.Source: §5.1 · pp. 13–14
5.2.1 Social media
- Social-media visual analytics address collective behaviors such as information spreading and mobility, as well as individual users’ interactions and anomalous behavior.Examples include maps and timelines for diffusion, interactive exploration of mobility patterns, and glyphs for suspicious users’ communication, features, and social interactions.Source: §5.2.1 · p. 14
5.2.2 E-commerce
- E-commerce and online-learning techniques visualize frequent browsing paths and user behavior, including interaction peaks and student click patterns.Some methods explore branching paths because frequent patterns are not always meaningful; PeakVizor uses additional views to show peak timing and correlations.Source: §5.2.2 · pp. 14–15
5.2.3 Computer systems
- Computer-system visual analytics combine aggregate timelines, per-node line charts, anomaly views, and multiple visualization modes to inspect performance and system behavior.These approaches support exploration of trends and anomalies in cloud and high-performance computing systems.Source: §5.2.3 · p. 15
5.3 Industry 4.0
- Factory anomaly analysis uses production-line event reports, product traces, and equipment-condition views to inspect recurring errors, performance, and deviations from estimated normal conditions.Wu et al. pair a short-term stacked timeline with a radial overview of longer-term equipment conditions.Source: §5.3 · p. 16, Figure 10 · p. 16
- Manufacturing-plan tools help planners inspect and revise schedules, using Gantt charts for current schedules and juxtaposed charts to compare production plans.LiveGantt supports interactive exploration and rescheduling; PlanningVis compares plans using heat maps, line charts, and bar charts.Source: §5.3 · p. 16, Figure 10 · p. 16
6 CHALLENGES AND OPPORTUNITIES
The survey identifies challenges in data quality, uncertainty, scalability, heterogeneity, multivariate-event visualization, interpretability, and causality analysis. It describes limitations of current approaches and directions for further visual analytics research.
Data quality
- Data-quality problems—including missing events or timestamps, inconsistent event names, and duplicate records—can mislead statistical analyses and inferences.The authors note a need for additional data-processing effort to prevent misleading results.Source: §6 · p. 17
Uncertainty
- Existing uncertainty visualizations focus on probabilistic uncertainty in event prediction, leaving other types such as bounded uncertainty underexplored.Uncertainty may arise from data-quality issues or user-specified data transformations and wrangling.Source: §6 · p. 17
Scalability
- Sequence aggregation and event filtering improve visual scalability but hinder inspection of individual sequences and events.The survey calls for scaling across summaries and low-level details so users can move between overview and detail.Source: §6 · p. 17
Heterogeneity
- Combining heterogeneous event types into one process may hinder discovery of relationships and distinctive patterns across event types.The survey calls for frameworks supporting both integrated analysis and investigation of individual processes.Source: §6 · p. 17
Multivariate event sequence visualization
- Most existing techniques characterize events by type and timestamp, while visualizing many multivariate attributes across differing data formats remains challenging.The reviewed bar-chart approach can be limited for finding associations within an event or across event types.Source: §6 · p. 17
Interpretability
- Complex machine-learning models can make results harder to interpret and reduce user control, while existing visual analytics approaches address particular tasks and models.The survey calls for more generalizable techniques that communicate model interpretations as comprehensible, actionable guidance.Source: §6 · p. 17
Causality Analysis
- Although event-sequence causality analysis has gained data-mining attention, few visual analytics techniques address it.The survey notes additional challenges from temporal structure and high dimensionality.Source: §6 · p. 17
7 CONCLUSION
The survey organizes visual analytics approaches for event sequence data through a taxonomy combining a design space and analytical-task categories. It also identifies remaining challenges and future directions, offering practitioners an overview of alternatives for their problems.
7 CONCLUSION
- The survey connects prior work in a taxonomy that characterizes visual analytics approaches for event sequence data.The taxonomy includes a design space and primary analytical tasks.Source: §7 · pp. 17–18
- Techniques are partitioned into five analytical-task categories, characterized by design elements, and summarized across major application domains.Source: §7 · p. 18
- The survey identifies remaining challenges and future research directions while offering practitioners an organized overview of alternative approaches for their analytical tasks.Source: §7 · p. 18