Skip to contents

Bundled datasets

Source: vignettes/articles/datasets.Rmd
datasets.Rmd

Bundled datasets

amorem ships five real-world relational-event datasets directly, each as a tidy (time, sender, receiver, ...) event table loadable via data(...). They span three orders of magnitude in size and cover three interaction types — face-to-face contact, phone calls, and email / instant messaging.

Dataset data(...) Events Actors Span Rate (/day) Interaction
Classroom classroom_events 691 20 44 d 15.8 face-to-face
Social Evolution social_evolution_calls 439 54 42 d 10.4 phone calls
Manufacturing radoslaw_email 82,927 167 271 d 305.8 email
CollegeMsg college_msg 59,835 1,899 194 d 308.9 instant messages
Email-EU-Core email_eu_core 12,216 89 803 d 15.2 email

Event-rate trace

The bundled datasets are emphatically non-stationary: term boundaries, weekly cycles, weekends, and bursty episodes all show up in the per-bin rate. This is exactly the regime the time-varying / global-covariate machinery in Estimation is built for.

Event-rate trace
Event-rate trace
  • Classroom: two spikes corresponding to the two recorded sessions; otherwise quiet.
  • Social Evolution: linear ramp-up — calls accumulate as the recording instrumentation rolls out.
  • Manufacturing & Email-EU-Core: visible weekly modulation superimposed on a roughly stationary base rate.
  • CollegeMsg: front-loaded — student onboarding traffic in the first quarter dominates the entire stream.

Per-actor activity distributions

Out-degree (events sent) and in-degree (events received), per actor, smoothed on a log scale:

Activity distributions
Activity distributions
  • Classroom is bell-shaped on a tight log axis — small homogeneous group.
  • Social Evolution and Email-EU-Core show moderately heavy tails.
  • Manufacturing and CollegeMsg are unambiguously heavy-tailed across three orders of magnitude — a small core drives most of the traffic.

These differences matter for inference: the *_count family in the Endogenous catalogue absorbs activity heterogeneity directly, so heavy-tailed datasets like Manufacturing and CollegeMsg make sender / receiver random effects (or compare_models_smooth() non-linear specs) a near- requirement for honest interpretation. See Real-data analysis for the worked example on Classroom (where the AIC ranking flips once sender frailty is added).

Provenance

  • Classroom — networkDynamic CRAN package. McFarland (2001).
  • Social Evolution — goldfish GitHub package. Madan et al. (2011).
  • Manufacturing — Network Repository, ia-radoslaw-email. Michalski et al. (2014).
  • CollegeMsg — SNAP, CollegeMsg.txt.gz. Panzarasa, Opsahl & Carley (2009).
  • Email-EU-Core — SNAP, email-Eu-core-temporal-Dept3. Paranjape, Benson & Leskovec (2017). Self-loops removed.

The fourth dataset analysed by Juozaitienė & Wit (2024) — Enron — is intentionally not bundled because the only publicly archived version is an aggregated daily edge-weight table rather than the event-level slice the paper analyses.

Time conventions

Times are normalised to days since the first event for every dataset except Classroom, which uses minutes (its original time unit). The original Unix-epoch timestamps are preserved as a unix_origin attribute on each event frame where the source provided them:

data(radoslaw_email)
attr(radoslaw_email, "unix_origin")
#> [1] "2010-01-02 14:45:35 UTC"

data(college_msg)
attr(college_msg, "unix_origin")
#> [1] "2004-04-15 03:02:41 UTC"

Auxiliary objects

  • classroom_actors, social_evolution_actors, social_evolution_friendship — per-actor covariates and the friendship-survey event log accompanying their main event tables.
  • dist_matrix — a 56 × 56 matrix of inter-state geographic distances; used by Experiment 2 in Validation experiments.

Citing the datasets

If you publish results from any bundled dataset, please cite the original source in addition to citing amorem:

  • McFarland, D.A. (2001). Student resistance: How the formal and informal organization of classrooms facilitate everyday forms of student defiance. American Journal of Sociology.
  • Madan, A. et al. (2011). Sensing the “health state” of a community. IEEE Pervasive Computing.
  • Michalski, R. et al. (2014). Matching organizational structure and social network extracted from email communication. New Generation Computing 32(3–4), 213–235.
  • Panzarasa, P., Opsahl, T., Carley, K. (2009). Patterns and dynamics of users’ behavior and interaction: Network analysis of an online community. Journal of the American Society for Information Science and Technology 60(5), 911–932.
  • Paranjape, A., Benson, A.R., Leskovec, J. (2017). Motifs in temporal networks. WSDM ’17, 601–610.
  • Juozaitienė R., Wit E.C. (2024). It’s about time: revisiting reciprocity and triadicity in relational event analysis. JRSS-A 188(4), 1246–1262.