Network of topics — topic_network • hypergraphs

Builds the topic-by-topic network of a document hypergraph, one node per topic and one undirected weighted edge per pair of related topics. The topics come either from a partition of the documents or from a mixed-membership topic model.

Usage

topic_network(
  hg,
  clusters = NULL,
  topics = NULL,
  threshold = NULL,
  cutoff = 0.01,
  similarity = c("none", "association", "cosine", "jaccard", "inclusion", "equivalence"),
  what = c("edges", "network")
)

Arguments

hg: The document hypergraph the topics were computed on.
clusters: A partition: the tidy table returned by hg_cluster() (columns node, cluster), a named vector of cluster labels, or the name of a column of the hypergraph's document table.
topics: A mixed-membership topic model of hg fitted by hg_topics(). Give clusters or topics.
threshold: Topic model only: the share at which a topic counts as present in a document, one number in (0, 1]. NULL (default) relates topics by the correlation of their shares instead.
cutoff: Topic model without threshold only: the correlation a pair must exceed to be kept (default 0.01, as in stm::topicCorr()).
similarity: "none" (default, the raw count), "association", "cosine", "jaccard", "inclusion" or "equivalence". Count networks only.
what: "edges" (default) for the edge list, "network" for a cograph_network built from it with cograph::as_cograph(), whose node table carries each topic's size.

Value

For what = "edges": a base data.frame, one row per pair of topics with a positive weight, columns source, target, weight, pairs in the topics' natural order. For what = "network": a cograph_network with one node per topic (label, name, size), a topic without any edge included, and one undirected weighted edge per pair; size is the number of documents of a cluster, the documents in which a topic is present, or (without threshold) a topic's expected number of documents. Raises hypergraphs_bad_input for unknown node names, for neither or both of clusters and topics, for a topic model not fitted on hg, for an invalid threshold or cutoff, and for a similarity measure on a correlation network. A topic whose share is the same in every document has no correlation with any other: it is kept without edges, with a hypergraphs_constant_topics warning.

Details

For a partition (clusters), the network is the co-occurrence network of the bibliometric kind: the strength between two topics is the sum, over all words, of the product of the number of documents in each topic that contain the word (full counting, the aggregation bibnets uses for keyword co-occurrence).

For a topic model fitted by hg_topics() (topics), two topics are related when the same documents draw on both. Without threshold, the weight is the correlation of the two topics' shares over the documents, and the pairs with a correlation above cutoff are kept, the "simple" topic correlation of stm::topicCorr() (Roberts et al. 2019). With threshold, a topic counts as present in a document when its share is at least threshold, and the weight of a pair is the number of documents in which both are present (Abuhay et al. 2017; Cassi et al. 2017).

A count network, from a partition or from a thresholded topic model, can be normalised by the similarity measures of bibnets' normalize(), with each topic's total on the diagonal (its documents with the word, or its documents with the topic present). With D_i the diagonal entry of topic i and A_ij the raw count: "association" is A_ij / (D_i D_j), "cosine" is A_ij / sqrt(D_i D_j), "jaccard" is A_ij / (D_i + D_j - A_ij), "inclusion" is A_ij / min(D_i, D_j) and "equivalence" is A_ij^2 / (D_i D_j) (van Eck & Waltman 2009).

The pairwise network of topics is the projection of the topic combinations that group_hypergraph() builds from the same topic model, in which every document binds all the topics it contains at once.

References

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.

Roberts, M. E., Stewart, B. M., & Tingley, D. (2019). stm: An R package for structural topic models. Journal of Statistical Software, 91(2), 1-40. doi:10.18637/jss.v091.i02

Abuhay, T. M., Kovalchuk, S. V., Bochenina, K., Kampis, G., Krzhizhanovskaya, V. V., & Lees, M. H. (2017). Analysis of computational science papers from ICCS 2001-2016 using topic modeling and graph theory. Procedia Computer Science, 108, 7-17. doi:10.1016/j.procs.2017.05.183

Cassi, L., Lahatte, A., Rafols, I., Sautier, P., & de Turckheim, E. (2017). Improving fitness: Mapping research priorities against societal needs on obesity. Journal of Informetrics, 11(4), 1095-1113. doi:10.1016/j.joi.2017.09.010

Examples

hg <- text_hypergraph(c(
  cooking_1 = "simmer the soup with onions and carrots",
  cooking_2 = "this soup recipe needs salt on a cold night",
  space_1 = "the telescope revealed a distant galaxy and stars",
  space_2 = "astronomers aimed the telescope at the stars all night"
), stop_words = c("the", "with", "and", "a", "this", "at", "on", "all"))
topics <- hg_cluster(hg, k = 2, seed = 1)
topic_network(hg, clusters = topics)
#>      source    target weight
#> 1 Cluster 1 Cluster 2      1
topic_network(hg, clusters = topics, similarity = "cosine")
#>      source    target     weight
#> 1 Cluster 1 Cluster 2 0.07715167

corpus <- c(
  a = "soup salt onion soup broth", b = "salt soup broth onion",
  c = "stars sky moon night", d = "sky stars night moon moon",
  e = "soup stars salt sky night broth", f = "onion salt stars broth")
corpus_hg <- text_hypergraph(corpus)
model <- hg_topics(corpus_hg, k = 2, nstart = 2)
topic_network(corpus_hg, topics = model, threshold = 0.2)
#>    source  target weight
#> 1 Topic 1 Topic 2      1