NEWS

SemNeT 2.1.1

UPDATE: 'pf' gains 'method' for the distance between responses ('"chebyshev"', the default and the distance before 2.1.0, '"euclidean"', '"manhattan"', '"cosine"', '"hamming"', '"jaccard"', or '"paulsen"', the response-order distance of 2.1.0); the distance is a modeling choice, and only '"paulsen"' needs response order (the others also accept a binary matrix)
ADD: 'pf' 'q' and 'r' arguments (Pathfinder parameters; defaults 'NULL', meaning n - 1, and 'Inf' are the previous behavior, computed with MST-Pathfinder); other settings use the general Pathfinder definition, and the documentation suggests an 'r' for each distance
FIX: 'cn' returned the wrong nodes whenever the network had isolated words: the largest component was found on a network without its isolates but indexed into the full one, so the returned nodes (and their edges) were not the largest component
FIX: 'convert2igraph' treated a node whose signed edges cancel (e.g., +1 and -1) as unconnected; it now requires a symmetric network, keeps the matrix form with a single edge, and documents that it drops nodes without edges
FIX: 'forward_flow' with 'type = "free"' now treats '""', '" "', and '"NA"' as missing (as the fluency format and the documentation already did), returns participants in their input order, and stops on missing or infinite 'embeddings' (it previously set them to 0)
FIX: 'bootstrap_SemNeT' node bootstrap of raw responses with 'cn', 'nrw', or 'pf' dropped response positions instead of words; raw responses are converted to ordered responses (words) first
FIX: 'permutation_SemNeT' stopped when a permuted sample had a response every participant gave and 'sim' was '"cor"' or '"phi"'; such responses are now dropped from both samples
FIX: 'random_network_test' left undefined ('NaN') random-network measures in the mean, which turned its p-value into the minimum possible; they are now left out with a warning. Each input network now has its own random networks (they previously shared one set of seeds), 'type' ('"arithmetic"' or '"harmonic"' ASPL) is added, and the random-network mean and standard deviation are no longer rounded to 4 digits
FIX: ‘random_walk' no longer lets a draw fall past a node’s last neighbor because of rounding, and a single level of steps no longer gives a row named '"Steps"'
FIX: 'response_analysis' no longer rounds p-values to 3 decimals (they are kept to 3 significant digits)
FIX: 'bad_response' ignores whitespace around responses ('" "', '" "', '" NA"')
FIX: Louvain aggregation ('q') counted the self-loops of an aggregated community twice
UPDATE: 'cc' requires a symmetric network, and 'type = "global"' returns 'NaN' for a network without connected triples ('na.rm' then only affects the per-node attribute)
UPDATE: 'cn' requires a whole-number 'window' smaller than the mean number of responses (Goni et al., 2011), and warns and returns an empty network when no pair co-occurs significantly; the documentation states that only the largest component is returned
UPDATE: 'finalize', 'bootstrap_SemNeT', and 'permutation_SemNeT' require 'minimum_cases' of at least 1; messages no longer suggest that 'finalize' removes responses everyone gave
UPDATE: 'seed' must be a whole number (a fraction below 1 silently meant random results)
UPDATE: 'compare_networks' restores the graphical parameters and, for 'placement = "match"', requires the same node names
UPDATE: legacy 'CN' accepts 'cutoff'
UPDATE: ‘plot.bootstrap_SemNeT' draws a bootstrap confidence interval of each group’s value below its boxplot ('ci', 'ci_level', 'ci_type'; not drawn for node bootstraps)
ADD: ‘plot.permutation_SemNeT' plots each measure’s permutation null distribution as a raincloud plot with the observed difference, the central range of the null distribution below the boxplot ('null_range', 'level'), and the p-value in each panel title
FIX: the Shiny app's ‘cores' input had no effect because the background task’s worker reported one available core; ‘cores' is now capped at the machine’s cores instead
ADD: the Shiny app's Permutation tab has a "Generate Permutation Plot" button ('plot.permutation_SemNeT')
UPDATE: the Shiny app's Export tab only shows the download buttons of results that have finished
UPDATE: the Shiny app's Permutation tab tests all measures (ASPL, CC, and Q) by default; the measures are checkboxes
ADD: the Shiny app's Load Data tab has a "Random Seed" input (0 for none), passed to every analysis that uses randomness (network measures' modularity, random networks, bootstrap, permutation, and random walks) for reproducible results
UPDATE: the Shiny app's Bootstrap and Permutation tabs show the same tables as ‘print.bootstrap_test_SemNeT' and 'print.permutation_SemNeT' (header, run settings, network sizes, comparison, table, and notes); the Spreading Activation tab shows each node’s activation at the selected time step
ADD: ‘cc' 'type': '"average"' (default; the average of the nodes’ local clustering coefficients, as before) or '"global"' (3 x triangles / connected triples; 'igraph::transitivity(type = "global")'); the weighted (and signed) global coefficient is the ratio of the summed triangle intensities (Onnela et al., 2005) to the number of connected triples; 'semantic_network_measures' and 'random_network_test' pass it on as 'cc_type'
REMOVE: 'tmfg' 'depend' argument (an asymmetric matrix is an error; the documented rule was never what the code did); the legacy 'TMFG' ignores 'depend' with a warning

SemNeT 2.1.0 (2026-10-14)

FIX: group tests in 'bootstrap_test_SemNeT' and 'test.bootSemNeT' inflated statistics by 'sqrt(iter)' (Tu, Chodorow, Brooks, Ezrina, & Valian, under review); recompute any results
UPDATE: 'bootstrap_test_SemNeT' now returns bootstrap intervals of within-iteration group contrasts ('ci_level', 'adjust', 'ci_type', 'power', 'legacy' arguments)
UPDATE: 'bootstrap_test_SemNeT' node bootstraps are summarized descriptively (no interval or p-value)
UPDATE: 'bootstrap_test_SemNeT' legacy parametric tests need 'legacy = TRUE' and always warn
UPDATE: 'bootstrap_SemNeT' '$summary' standard error is no longer divided by 'sqrt(iter)'; observed (full-data) statistics are stored in '$observed'
ADD: 'bootstrap_test_SemNeT' Wald 'z' and 'p_value' for normal intervals (the default)
ADD: 'bootstrap_test_SemNeT' 'bias_se' (|bias| / 'se'; above about 0.25 the bootstrap is off-centre from the observed difference), printed as 'bias/se'
ADD: 'bootstrap_test_SemNeT' and 'permutation_SemNeT' print a note under their summary tables that they should always be run together
FIX: 'cn' 'alpha' had no effect
FIX: 'cn' counted co-occurrences on uncleaned responses, so placeholders ('" "', '"NA"') emptied the network
FIX: 'nrw' counted every adjacent pair twice per participant; pairs are now counted once (counts are numbers of participants)
FIX: 'pf' dropped zero-distance pairs, leaving identical responses unlinked
FIX: 'binary2responses' ignored response order in an ordered numeric matrix
FIX: 'permutation_SemNeT' computed 'q' on the weighted network; all measures are now unweighted
FIX: 'aspl' per-node averages use distances to the other nodes; weighted 'aspl' rejects negative weights
FIX: 'similarity' method '"euclid"' returned a distance, which 'tmfg' treated as a similarity; it now returns '1 / (1 + d)'
FIX: 'tmfg' stops on undefined similarities and on asymmetric input without 'depend = TRUE'
FIX: 'equate' rejects raw responses and duplicated response names
FIX: 'forward_flow' treats '""', '" "', and '"NA"' as missing and requires 'min_response' of at least 2
FIX: 'q' with 'signed = TRUE' weighted the negative term by N / (P - N) instead of N / (P + N) (Gomez et al., 2009); also fixed in {EGAnet}
FIX: 'cohens_d' pools standard deviations by degrees of freedom (agrees with the equal-variance t-test) and handles groups of one
FIX: ‘random_walk' 5-unique-node statistic depended on the other network’s walk; isolated start nodes and zero variation handled
FIX: 'random_network_test' summed weights for edge counts; networks are binarized, p-values are Monte Carlo, and a degree-preserving null is added
UPDATE: 'cn' gains 'cutoff' for 'enrich = TRUE' (default '0.50', was a hardcoded '0.35') and warns that there is no established value
UPDATE: 'pf' now uses the response-order distance of Paulsen et al. (1996) and needs response order (raw or ordered responses, not 0/1 data)
UPDATE: 'semantic_network_measures' defaults to 'weighted = FALSE' (as documented) and gains 'na.rm', 'signed', and 'type' (passed to 'cc', 'q', and 'aspl')
UPDATE: 'cc' gains 'na.rm' and no longer rounds per-node values; weighted 'cc' divides weights by the largest absolute weight
UPDATE: 'similarity' stops on missing or infinite values (and non-0/1 data for '"angular"', '"faith"', '"jaccard"', '"rr"'); 'finalize' stops on missing values and non-whole 'minimum_cases'
UPDATE: 'q' requires a symmetric network and returns 'NaN' for a network without edges
UPDATE: 'seed = 0' means random results, like 'NULL'
UPDATE: 'response_analysis' returns 'vocabulary' (Jaccard overlap and rarefied richness) and ignores never-given responses in binary input
UPDATE: ‘random_walk' no longer reports p-values (they only reflect 'iter'); differences, Monte Carlo intervals, and Cohen’s d are given instead
UPDATE: 'bootstrap_SemNeT' node bootstrap drops the same nodes from every group in an iteration
UPDATE: 'bootstrap_SemNeT' retries draws only for correlation similarities ('"cor"', '"phi"'); plot shows the full-data value
UPDATE: 'permutation_SemNeT' tests several measures at once and returns 'diagnostics'; printed table columns follow 'bootstrap_test_SemNeT' ('null_range' and 'null_sd' in place of 'CI' and 'se')
UPDATE: 'permutation_SemNeT' gains two-sided options '"two_tailed_centered"' and '"two_tailed_doubled"', tolerates floating-point ties, and aligns samples of different widths or vocabularies
UPDATE: Shiny Bootstrap tab shows intervals instead of t-test/ANOVA/ANCOVA; labels and captions updated
ADD: 'aspl' 'type' argument ('"arithmetic"', the default and previous behavior, or '"harmonic"')
ADD: 'seed' for 'bootstrap_SemNeT' and 'permutation_SemNeT'; each resample gets its own seed, so results do not depend on 'cores' or 'set.seed'
ADD: 'random_network_test' 'seed' argument
ADD: ‘bootstrap_test_SemNeT' prints and stores the groups’ full-sample values ('observed_1', 'observed_2' in each 'table'), like 'permutation_SemNeT'
UPDATE: 'q' seeds now line up between 'bootstrap_SemNeT' and 'permutation_SemNeT': 'permutation_SemNeT' scores the observed data (first permutation) with 'seed' itself, as 'bootstrap_SemNeT' does for its full-data networks, so their empirical 'q' differences match (they previously used different Louvain seeds, which changes 'q' slightly); 'bootstrap_SemNeT' gives each iteration one community-detection seed shared by all groups
ADD: 'print' methods for 'bootstrap_test_SemNeT' and 'permutation_SemNeT'

SemNeT 2.0.0 (2026-09-12)

FIX: 'equate' compared vocabulary size rather than shared words
FIX: duplicated group names passed via '...' silently hid results
FIX: weighted 'CC' returned 'NaN' for negative weights
FIX: 'ASPL' zeroed unreachable pairs, understating ASPL in disconnected networks
FIX: 'TMFG' with 4 nodes iterated backwards and fewer than 4 nodes gave silent 'NA's
FIX: 'NRW' rejected valid ordered response matrices
FIX: 'test.bootSemNeT' errored when all three measures were requested
FIX: 'permSemNeT' dropped 'NRW' 'threshold' ('thresold' typo), 'CN' 'enrich', and passed 'threshold' as 'window'
FIX: several '1:n' ranges ran backwards on empty or size-1 inputs
FIX: ‘q' is reproducible via 'seed' (it depended on R’s global RNG through 'igraph::cluster_louvain')
FIX: 'forward_flow' 'type = "free"' mismatched or crashed on numeric participant IDs (GitHub Issue #10)
FIX: Shiny Spreading Activation all-zero output and BRM upload failure (GitHub Issue #11)
FIX: Shiny app called unexported functions ('responses2binary', 'read_uploaded_file') that failed under 'library(SemNeT)'
FIX: Shiny background analyses ran sequentially and froze the app; they now run in the background with a live progress bar
FIX: Shiny progress bar jumped around with 'cores > 1', and 'cores' above 1 failed inside background workers
UPDATE: full rebuild of the package's internals and the Shiny app; every pre-2.0.0 exported name still works
UPDATE: 'TMFG', 'ASPL', 'CC', 'CN', 'NRW', 'PF', and 'Q' are now lowercase ('tmfg', 'aspl', ...); the old names remain as wrappers
UPDATE: 'bootSemNeT', 'test.bootSemNeT', 'randwalk', 'randnet.test', 'sim.fluency', 'response.analysis', 'semnetmeas', and 'compare_nets' are now 'bootstrap_SemNeT', 'bootstrap_test_SemNeT', 'random_walk', 'random_network_test', 'simulate_fluency', 'response_analysis', 'semantic_network_measures', and 'compare_networks'; the old names remain as wrappers
UPDATE: 'bootSemNeT', 'randwalk', and 'randnet.test' default to sequential processing instead of auto-detecting cores
UPDATE: 'finalize' arguments renamed from 'rmat' and 'minCase' to 'data' and 'minimum_cases'
UPDATE: Shiny maximum ‘cores' is the host’s core count minus one
UPDATE: 'bootstrap_SemNeT' and 'permutation_SemNeT' no longer hold every resampled network in memory ('keep_networks' argument); 'random_network_test' reduces each random network immediately
UPDATE: dropped 'pbapply', 'dplyr', 'plyr', 'magrittr', 'scales', and 'philentropy'; added 'future', 'future.apply', and 'progressr'; 'ggplot2' is now 'Suggests'
UPDATE: compiled C code (Louvain, signed modularity, xoshiro256++) vendored from {EGAnet}; 'NeedsCompilation' is now 'yes'
UPDATE: one file per exported function, a consistent 'bootstrap_SemNeT' result schema, vectorized loops, and removal of dead code
ADD: 'permutation_SemNeT' (the old 'permSemNeT' was internal and undocumented)
ADD: 'forward_flow' is back, using user-supplied word embeddings (no network access)
ADD: 'responses2binary' and 'read_uploaded_file' are exported
ADD: 'find_textcleaner_objects' and a Load Data picker in the Shiny app for {SemNetCleaner} objects

SemNeT 1.4.4 (2023-08-12)

FIX: minor fix to get back on CRAN
TODO: major refactor for the entire package to increase future reliability

SemNeT 1.4.3 (2021-09-04)

FIX: switching to different network estimations in Shiny app

SemNeT 1.4.2 (2021-06-22)

FIX: bug in reset of Shiny application
FIX: name ordering of 'bootSemNeT' plots
FIX: p-value bug in random networks tests
ADD: t-tests to 'test.bootSemNeT' and Shiny app

SemNeT 1.4.1 (2020-12-04)

FIX: unequal nodes and edges in 'test.bootSemNeT' (GitHub Issue #3)
FIX: 'PF' remove zeros bug (when there were no zero distances, then an empty matrix was given)
UPDATE: tooltips added to network estimation methods in Shiny
UPDATE: associated Shiny packages messages suppressed on Shiny startup
UPDATE: message for Shiny results output
UPDATE: 'test.bootSemNeT' no longer uses number of nodes in ANCOVA
UPDATE: added minimum number ('type = "num"') and proportion ('type = "prop"') of co-occurrences argument to 'NRW'
UPDATE: Shiny with new 'NRW' arguments
UPDATE: Shiny now provides effect sizes for partial eta squared in Bootstrap Network Analyses tab

SemNeT 1.4.0 (2020-09-26)

FIX: ‘randwalk' bug with ’steps' and 'reps' argument
FIX: Shiny 'Clear Results' for new analyses
FIX: output for animation plots in Shiny
UPDATE: output names for 'randwalk'
UPDATE: Shiny interface is condensed and more step-by-step progressive
UPDATE: Shiny includes random walk analysis
UPDATE: Shiny includes spreading activation analysis (thanks to Cythina Siew)
UPDATE: progress bar for some Shiny analyses
UPDATE: S3method for animation plots
UPDATE: citations added to Shiny analyses
UPDATE: ANOVAs controlling for the number of nodes and edges is now the statistical test in 'test.bootSemNeT'
ADD: animation plot for spreading activation analysis in Shiny
ADD: permutation test to SemNeT ('permSemNeT) and Shiny
ADD: functions to convert data into other softwares format: SNAFU ('convert2snafu') and igraph ('convert2igraph')

SemNeT 1.3.0 (2020-06-09)

FIX: output of two-way ANOVA in 'test.bootSemNeT'
FIX: 'groups' argument in 'test.bootSemNeT'
UPDATE: added integration of SemNetCleaner output into 'SemNeTShiny'
UPDATE: 'compare.nets' has changed to 'compare_nets'
UPDATE: 'cosine' was added to internal functions to remove archived package from dependencies
ADD: Several semantic network estimation methods (see 'CN', 'NRW', and 'PF') -- these are included in the Shiny app
MOVE: 'equate' and 'finalize' moved from SemNetCleaner to SemNeT
REMOVE: foreach and doParallel dependencies were removed (all parallelization is through the pbapply package)
REMOVE: SemNetCleaner dependency was removed to allow SemNeT to operate independently -- all necessary functions from SemNetCleaner have been moved to internal functions
CHANGE: 'partboot' is now 'bootSemNeT'
CHANGE: 'plot.partboot' is now 'plot.bootSemNeT'
CHANGE: 'partboot.test' is now 'test.bootSemNeT'

SemNeT 1.2.0 (2020-03-22)

FIX: 'groups' argument in 'org.plot'
UPDATE: moved several functions to internal to clean up package
ADD: Shiny app
ADD: binary responses matrices from animals verbal fluency data associated with the 'SemNetCleaner' package (open.animals)

SemNeT 1.1.3 (2020-02-07)

added extended ANOVA capabilities to 'partboot.test' (e.g., two-way ANOVA)

SemNeT 1.1.2 (2019-10-28)

fixed bug for weighted modularity in 'semnetmeas'
updated citation (PsyArXiv)
changed y-axis in plots for 'partboot' to read as integers (as added black dot for mean in the scatterplots under the raincloud)
added vignette for analyzing semantic networks
added data for plots in vignette

SemNeT 1.1.1 (2019-09-27)

updated citation

SemNeT 1.1.0

bug fixes
changed 'semnetmeas' to allow calculation of any combination of measures
'partboot', 'partboot.test', 'plot.partboot', 'randnet.test', and 'compare.nets' now handle groups greater than 2

SemNeT 1.0.0 (2019-08-05)

serious, and I mean serious, revamp of functions (all for the better)
moved most functions over to 'SemNetCleaner' and now only consists of methods and measures for semantic network analysis
added a 'sim.fluency' function for toy examples in CRAN checks
added two datasets 'one.result' and 'two.result' for toy examples in CRAN checks
'partboot.plot' changed to 'plot.partboot' and added to S3 methods
'partboot.test' now outputs a list for each network measure

SemNeT 0.9.9

animals.database: a database for animals was added
animals.misnomer: common misnomers for animals was added
textcleaner: added database functionality, which makes the cleaning process more accurate, efficient, and reproducible
several wrapper functions added to aid textcleaning from a database

SemNeT 0.1.2

autoConverge: significantly improved interface
separated all function files into individual files
semnetcleaner (changed to textcleaner): added argument to input participants by row or column; ids now included for output; added output to show what spellcheck changed
destr: improved functionality to allow selection of separating character
randwalk: added a function to simulate random walks on the network to examine spontaenous spreading activation
bin2resp: a function to change cleaned binary response data back into character data for each participant

SemNeT 0.1.1

cosine: added cosine similarity
semnetboot: changed name to partboot and added the cosine similarity
semnetmeas: moved from NetworkToolbox
plot.partboot: a function that generates "raindrop" plots for bootstrapped data
org.plot: an organization function for plot.partboot (not for individual use)
convert2cytoscape: added a function to convert a network matrix into a sparse matrix for Cytoscape visualization
added new examples with corresponding data files

SemNeT 0.1.0

autoConverge: created an automated converge function to eliminate the manual converge process
autoDeStr: created an automated de-string function to eliminate most manual de-stringing
fine-tuned most functions
semnetboot: moved from package NetworkToolbox