
Vignettes
Guides
Case studies
cooccure builds co-occurrence networks, in which two
items are linked when they appear in the same transaction: the keywords
of one article, the genres of one film, the actors of one cast, the
codes of one interview passage, or the states of one learning session.
The network shows which items tend to go together and how a field, a
corpus or a set of sessions is organized around them. The weight of a
link starts as the number of shared transactions. Because frequent items
co-occur often by chance alone, the weight can be normalized for item
frequency by one of eight similarity measures, such as association
strength or the Jaccard index (van Eck & Waltman, 2009), and each
transaction’s contribution can be counted fractionally
(Perianes-Rodriguez et al., 2016). One network can be built per group,
for example per decade, and the networks compared.
All of this is done by one function, cooccurrence(). It
reads delimited columns, long (item, transaction) tables, binary and
one-hot tables, wide sequences, lists of vectors and event logs; sliding
windows turn ordered sequences into transactions. It returns a tidy edge
list with a weight and a raw count per pair, which converts to the
network classes of igraph, tidygraph, cograph and Nestimate.
Let \(B\) be the transaction-by-item incidence matrix, with \(B_{ti} = 1\) when item \(i\) occurs in transaction \(t\). The co-occurrence counts are \(C = B^\top B\), with the diagonal set to zero, and \(f_i\) is the number of transactions containing item \(i\). Similarity measures rescale \(C_{ij}\) by the frequencies (van Eck & Waltman, 2009):
similarity |
Weight of the edge \(i\)–\(j\) |
|---|---|
"none" |
\(C_{ij}\) |
"jaccard" |
\(C_{ij} / (f_i + f_j - C_{ij})\) |
"cosine" |
\(C_{ij} / \sqrt{f_i f_j}\) |
"inclusion" |
\(C_{ij} / \min(f_i, f_j)\) |
"association" |
\(C_{ij} / (f_i f_j)\) |
"dice" |
\(2 C_{ij} / (f_i + f_j)\) |
"equivalence" |
\(C_{ij}^2 / (f_i f_j)\) |
"relative" |
\(C_{ij} / \sum_k C_{ik}\) (directed) |
With fractional counting each transaction of \(n\) items contributes \(1/(n-1)\) to each of its pairs instead of 1
(Perianes-Rodriguez et al., 2016); with attention counting a pair
contributes \(\exp(-|p_i - p_j| /
\lambda)\), decaying with the distance between the two positions
in an ordered transaction. Every measure except "relative"
gives an undirected network; "relative" divides each row by
its sum, so its edges have a direction and the edge list holds one row
per direction.
install.packages("cooccure")
# development version
install.packages("remotes")
remotes::install_github("mohsaqr/cooccure")movies holds 1,000 highly rated films with their genres
in one comma-separated column. Genres are linked when they are assigned
to the same film, and Jaccard similarity discounts the most common
genres.
library(cooccure)
genres <- cooccurrence(movies, field = "genres", sep = ",",
similarity = "jaccard", min_occur = 20)
print(genres, n = 6)
#> # cooccurrence: 17 nodes, 102 edges | density: 0.7500 | mean degree: 12.0000 | 998 transactions
#> # similarity: jaccard | counting: full
#> # possible edges: 136 | isolates: 0
#> # top nodes: Drama(deg=16,str=1.153), Comedy(deg=16,str=0.8949), Romance(deg=15,str=0.5377), Action(deg=14,str=1.009), Adventure(deg=14,str=0.8788)
#> from to weight count
#> Adventure Animation 0.285 39
#> Action Crime 0.237 59
#> Comedy Drama 0.211 159
#> Action Adventure 0.208 44
#> Biography Documentary 0.204 44
#> Drama Romance 0.198 131
#> # ... 96 more edgesThe weight matrix can be drawn as a heatmap.
plot(genres)
Jaccard weights of the genre network.
split_by builds one network per decade, and the group
table compares them.
by_decade <- cooccurrence(movies, field = "genres", sep = ",",
split_by = "decade", similarity = "jaccard",
min_occur = 5)
as.data.frame(by_decade, what = "groups")
#> group n_nodes n_edges possible_edges density weight_mean count_sum
#> 1 1970s 12 34 66 0.515 0.1334 136
#> 2 1980s 14 43 91 0.473 0.0977 164
#> 3 1990s 13 39 78 0.500 0.1126 187
#> 4 2000s 16 66 120 0.550 0.0769 335
#> 5 2010s 17 85 136 0.625 0.0666 581
#> 6 2020s 13 39 78 0.500 0.1004 184actors is a long table with one row per actor and film;
actors in the same film are linked, giving a collaboration network. One
decade of a split network converts to igraph with
group =.
cast <- cooccurrence(actors, field = "actor", by = "tconst",
split_by = "decade", min_occur = 2)
summary(as_igraph(cast, group = "2010s"))
#> IGRAPH ab8d353 UNW- 145 756 --
#> + attr: name (v/c), weight (e/n), count (e/n)cooccure is written by Mohammed Saqr, Sonsoles López-Pernas and Kamila
Misiejuk. Mohammed Saqr maintains the package. Questions and bug reports
can be submitted through the repository issue
tracker. The package citation is available through
citation("cooccure").
The bundled movies, actors and
actor_genres data are derived from the IMDb non-commercial
datasets (https://data.imdb.com/non-commercial-datasets/).
Perianes-Rodriguez, A., Waltman, L., & van Eck, N. J. (2016). Constructing bibliometric networks: A comparison between full and fractional counting. Journal of Informetrics, 10(4), 1178–1195. https://doi.org/10.1016/j.joi.2016.10.006
Saqr, M., López-Pernas, S., Conde, M. Á., & Hernández-García, Á. (2024). Social network analysis: A primer, a guide and a tutorial in R. In M. Saqr & S. López-Pernas (Eds.), Learning analytics methods and tutorials: A practical guide using R. Springer. https://doi.org/10.1007/978-3-031-54464-4_15
van Eck, N. J., & Waltman, L. (2009). How to normalize cooccurrence data? An analysis of some well-known similarity measures. Journal of the American Society for Information Science and Technology, 60(8), 1635–1651. https://doi.org/10.1002/asi.21075