Paper
G. Andrienko, N. Andrienko, M. Resinas, S. van den Elzen, and B. Weber, “TracePrism: Extending Semantic Trace Topics to Iterative, Encoding-Aware Trace Partitioning.” Submitted to IEEE Transactions on Visualization and Computer Graphics (TVCG) for possible publication. This site accompanies that paper and is referenced from its System section as the source for the implementation and the three case-study datasets.
About
Real event logs are rarely homogeneous: different subsets of cases follow genuinely different sub-processes, and a single global model, clustering, or classifier fit to the whole population averages these differences away. TracePrism reframes process discovery as iterative partitioning: at each step, the analyst picks a subset of unexplained traces and an encoding for it, guided by evidence-scored suggestions; searches for a candidate group coherent in both structure and outcome; interprets it through an automatic attribute fingerprint; and commits it to a revisable knowledge base or reopens it under a different subset or encoding. This page hosts the live system and the full, reproducible record of applying it to three public event logs from distinct domains.
Technical appendix
Implementation details and reproducibility material that don't fit the paper's page budget: complete token-encoding parameter definitions, a full tab-by-tab interface walkthrough, algorithm defaults and seeds (NMF, k-means, UMAP), and an artifact/archival checklist.
Live demo
Runs entirely client-side — nothing is uploaded anywhere. Loading the largest dataset (Road Traffic Fines, ~150,000 traces) can take a few seconds on first launch.
The system above is a research demonstrator built to accompany the paper below, not a maintained product — provided as-is for reviewers and readers to reproduce and explore the case studies.
Case studies
For each log below: the full narrated case study, the raw timestamped action log the narration is drawn from, the exported knowledge-base report, and the underlying group and per-trace CSV exports — the complete provenance trail behind every number reported in the paper.
Road Traffic Fines
-
Case study write-upFull narrated walkthrough, screenshots included6.1 MB
-
Analysis logTimestamped record of every action267 KB
-
Knowledge base reportEvery committed group's card458 KB
-
Knowledge base report — with temporal dynamicspost-hoc analysisSame report, extended after the session with per-group end-date heatmaps and pooled outcome-vs-month timelines — a follow-up temporal analysis run on the committed groups after the iterative topic-modeling session ended, not part of the session itself1.1 MB
-
knowledge_groups.csvOne row per group2.7 KB
-
knowledge_traces.csvOne row per trace, group membership10.0 MB
Sepsis Cases
-
Case study write-upFull narrated walkthrough, screenshots included12.6 MB
-
Analysis logTimestamped record of every action209 KB
-
Knowledge base reportEvery committed group's card333 KB
-
knowledge_groups.csvOne row per group2.4 KB
-
knowledge_traces.csvOne row per trace, group membership58 KB
BPI Challenge 2013
-
Case study write-upFull narrated walkthrough, screenshots included14.4 MB
-
Analysis logTimestamped record of every action130 KB
-
Knowledge base reportEvery committed group's card212 KB
-
knowledge_groups.csvOne row per group1.2 KB
-
knowledge_traces.csvOne row per trace, group membership368 KB
"Explained" is the share of each log's traces covered by an active, committed knowledge-base entry at the end of that session; see the paper's Discussion for how the three logs compare and why the hardest-to-capture outcome differs across them.