Reproducing the illustrative example

This vignette reproduces the illustrative example of Delias et al. (2023), “Improving the non-compensatory trace-clustering decision process”. The log ships with the package as illustrative_log.

The scenario

A service desk handles two customer tiers, "GOLD" and "NORMAL" (the paper calls the latter “Blue”), each with its own process flow, and records each customer’s satisfaction. There are 25 fictitious customers (Table 1 of the paper), numbered in Table 1’s order; the last two (cases 24 and 25) are deliberate outliers whose flow does not fit either tier.

traces <- as_traces(illustrative_log, "case_id", "activity", "timestamp")
table(illustrative_log$status[!duplicated(illustrative_log$case_id)],
      illustrative_log$satisfaction[!duplicated(illustrative_log$case_id)])
#>         
#>          High Low
#>   GOLD      9   2
#>   NORMAL   12   2

The criteria (Table 2)

The paper defines four criteria with the following parameters. “Status” and “Satisfaction” are binary criteria; a veto of -1 in the paper means “no veto”, which we express as veto = NULL.

Criterion	Direction	Similarity	Indifference	Veto	Weight
Activities	similarity	0.8	0.7	0.4	0.2
Transitions (edit distance)	dissimilarity	2	3	6	0.2
Status	similarity	1	0	–	0.3
Satisfaction	similarity	1	0	–	0.3

criteria <- list(
  crit_activity_profile(weight = 0.2, indifference = 0.7, similarity = 0.8,
                        veto = 0.4),
  crit_edit_distance(weight = 0.2, similarity = 2, indifference = 3, veto = 6),
  crit_nominal("status", weight = 0.3),
  crit_nominal("satisfaction", weight = 0.3)
)
sim <- outrank_similarity(traces, criteria)
sim
#> <outrank_sim>: 25 cases, 4 criteria
#>   S in [0.000, 1.000], symmetric = TRUE

Cases that share the same trace, tier and satisfaction are fully similar under every criterion, so their credibility is exactly 1:

sim$S[c("1", "2", "3"), c("1", "2", "3")]
#>   1 2 3
#> 1 1 1 1
#> 2 1 1 1
#> 3 1 1 1

Run 1: clustering without domain knowledge

The paper targets four clusters (tier x satisfaction). This is the baseline “Run 1”, with no constraints or trimming.

clust <- cluster_traces(sim, k = 4, seed = 42)
split(names(clust$memberships), clust$memberships)
#> $`1`
#> [1] "11" "12" "13" "14" "15" "24"
#> 
#> $`2`
#> [1] "16" "17" "18" "19" "20" "21" "22" "23"
#> 
#> $`3`
#> [1] "6"  "7"  "8"  "9"  "10"
#> 
#> $`4`
#> [1] "1"  "2"  "3"  "4"  "5"  "25"

The clusters recover the main structure of the process: cases with an identical profile always land together, the two satisfaction levels are largely separated, and the short-path Gold customers form their own group.

A note on exact reproduction

The paper reports its memberships as a shaded figure (Table 3) rather than as a table, so they are not machine-readable here. simOutrank also standardises on the corrected Ng-Jordan-Weiss row normalisation (see ?cluster_traces), which can differ from the normalisation used to produce the original figure. We therefore reproduce the pipeline and its structural conclusions rather than a bit-for-bit membership vector; the credibility matrix S itself is pinned as a regression fixture in the package’s tests.

Runs 2 and 3 of the paper – separating the tiers with a cannot-link constraint and trimming the outliers – are covered in the robustness vignette.

Reproducing the illustrative example

The scenario

The criteria (Table 2)

Run 1: clustering without domain knowledge

A note on exact reproduction

Reference