Skip to contents

Dynamic-state indicators and classification for flow trajectories

Source: R/sniff_trajectory_dynamics.R
sniff_trajectory_dynamics.Rd

Characterizes each sniff_trajectory_braid() trajectory along the evolutionary states of the method (emergence, convergence, divergence, dormancy), grounding the four emergence pillars (novelty, growth, community, persistence) in two complementary lenses tied to the flow model.

Usage

sniff_trajectory_dynamics(
  flow,
  thresholds = NULL,
  k = 1,
  growth_window = 3,
  novelty_window = 5,
  winsorize = 3,
  cct = NULL,
  entropy = NULL,
  hubs = NULL
)

Arguments

flow

A sniff_trajectory_braid() object.

thresholds

A fixed_state_thresholds()-shaped list. NULL (default) derives the cut points from the data via data_state_thresholds() with the k below, so the classification adapts to the dataset; pass fixed_state_thresholds() to force the fixed constants instead. The thresholds actually used are attached to the result as attr(result, "state_thresholds"), which the plots read back.

k

Robust deviations for the data-driven thresholds when thresholds is NULL (default 1); see data_state_thresholds(). Ignored when an explicit thresholds list is supplied.

growth_window

Curve points over which growth_rate is fitted (default 3).

novelty_window

Years counted as recent for novelty (default 5).

winsorize

Cap each pillar's robust z-score at c(-winsorize, +winsorize) before summing into emergence_index, so no single extreme pillar (e.g. a young core 18 MADs out on growth) dominates the ranking. Default 3 (near the Iglewicz–Hoaglin modified-z outlier boundary of 3.5); NULL leaves the robust z-scores uncapped.

cct

Optional sniff_trajectory_cct() output (one row per traj_id), left-joined to add the cct list-column (per-year renewal-pace series); NULL (default) omits it.

entropy

Optional sniff_trajectory_entropy() output (one row per traj_id), left-joined to add the keyword_entropy list-column (per-year keyword-diversity series); NULL (default) omits it.

hubs

Optional sniff_trajectory_hubs() output (one row per traj_id), left-joined to add the hub-role columns; NULL (default) omits them.

Value

A tibble, one row per trajectory, sorted by descending emergence_index (absorbed last): traj_id, type, group, start, end, age, size, growth_rate, doubling_time, novelty, recruitment, emergence_density, attraction_inflow, dest_entropy, dormant_share, phase, fate, state, emergence_index, emergence_index_intensive (the same index without the extensive recruitment pillar), reach_ratio, plus any cct / keyword_entropy (per-year list-columns) and hubs columns.

Details

Forward lens (all trajectories). growth_rate and doubling_time (an exponential fit of the recent size curve), novelty (share of documents that arrived recently), recruitment (net documents gained over the lifespan, the power of attraction), emergence_density (staying power: the share of the lineage's year-steps still growing at the emergence rate, after Carley et al. 2017), and age/size (persistence and community).

Backward lens (absorbed trajectories only). Where the terminal cohort goes, via sniff_trajectory_destination(): dest_entropy (normalized Shannon entropy of the destination split) and dormant_share (share dropping out).

Inflow lens (central trajectories only). attraction_inflow, the documents of the final community delivered by absorbed tributaries (the consolidating side of convergence), from sniff_trajectory_self_sufficiency().

Reach lens (all trajectories). reach_ratio, the share of the future field a lineage's origin seeds – its birth node's forward temporal reach in the DAG over all nodes born later (after the temporal-reachability of Marino & Silva 2023). Structural downstream influence, distinct from size and, via the ratio, from age.

Classification uses two axes: a life-cycle phase for living (central) trajectories – emergence / maturity / dormancy(stall) by growth and novelty – and a terminal fate for declining (absorbed) ones – convergence / divergence / dormancy(death) by destination. state is phase for centrals and fate for absorbed; phase and fate are kept separate so the two senses of dormancy (a stalled living core vs a dead cohort) stay distinguishable. The emergence_index (a within-central combination of growth, novelty and recruitment) is computed for living trajectories only, so an absorbed lineage that grew before dying is never ranked as emerging. Because recruitment is an extensive (size-carrying) pillar, the full index partly tracks community size; emergence_index_intensive drops it (growth and novelty only) for a rate-based ranking that does not reward a large core for its volume alone.

See also

sniff_trajectory_braid(), sniff_trajectory_destination(), sniff_trajectory_self_sufficiency(), sniff_trajectory_cct(), sniff_trajectory_entropy(), sniff_trajectory_hubs(), fixed_state_thresholds()

Other trajectory analysis: sniff_trajectory_cct(), sniff_trajectory_community(), sniff_trajectory_confluence(), sniff_trajectory_contribution(), sniff_trajectory_destination(), sniff_trajectory_emergence_owners(), sniff_trajectory_entropy(), sniff_trajectory_formation(), sniff_trajectory_hubs(), sniff_trajectory_self_sufficiency()