Give readable names to clusters

A cluster identifier ("01", "cl_12") says nothing about its content. name_clusters() gives each cluster a provisional label that helps reading plots and tables, with one of three methods: the words that distinguish the cluster from the others ("tf-idf"), or the metadata of its most central node ("tidygraph_functions") or of its node with the highest value of a given column ("given_column"). These labels are starting points for interpretation, not the result of it: naming a cluster requires reading its documents, the references they share, and what distinguishes them from neighbouring clusters (Ding 2011) .

Usage

name_clusters(
  graphs,
  method = c("tidygraph_functions", "given_column", "tf-idf"),
  name_dynamic_clusters = FALSE,
  cluster_id,
  label_columns,
  label_name = "cluster_label",
  tidygraph_function = NULL,
  order_by = NULL,
  text_columns = NULL,
  nb_terms_label = 3,
  name_merged_clusters = NULL,
  ...
)

Arguments

graphs: A tbl_graph or a list of tbl_graphs.
method: "tf-idf" (chosen when no method is given), "tidygraph_functions" or "given_column" (see the details).
name_dynamic_clusters: Logical. TRUE when cluster_id identifies clusters across a list of graphs (the dynamic clusters of merge_dynamic_clusters()): each cluster then gets one name, computed on all its nodes in all windows. FALSE (default) names the clusters of each graph separately.
cluster_id: Character scalar, the node column with the clusters to name.
label_columns: For "tidygraph_functions" and "given_column", the node column(s) used to build the label of the chosen node, joined by "_" (e.g. c("source_author", "source_year")).
label_name: Character scalar, the name of the new column ("cluster_label" by default).
tidygraph_function: For "tidygraph_functions", a call to a centrality function of tidygraph, such as tidygraph::centrality_pagerank() (see tidygraph::centrality).
order_by: For "given_column", the numeric node column used to choose the node (e.g. "node_size" from build_network() with compute_size = TRUE, i.e. the most cited document of the cluster).
text_columns: For "tf-idf", the node column(s) with text (titles, abstracts, keywords), passed to extract_tfidf().
nb_terms_label: For "tf-idf", the number of terms in each label (3 by default), separated by commas. A term that is part of a term already in the label, or contains one, is skipped ("business" after "business cycle").
name_merged_clusters: Former name of name_dynamic_clusters.
...: For "tf-idf", other arguments of extract_tfidf() (e.g. clean_word_method, n_gram, ngrams_filter).

Value

The same object as graphs, with a new node column label_name, also joined to the edges through their cluster column.

Details

tf-idf. Each cluster is treated as one document made of the texts of its nodes, and each term $t$ (word or n-gram) gets the weight $$\mathrm{tf}_{tc} \times \mathrm{idf}_t = \frac{n_{tc}}{\sum_u n_{uc}} \times \log\frac{C}{\mathrm{df}_t}$$ (Spärck Jones 1972; Salton and Buckley 1988) , with $n_{tc}$ the number of occurrences of $t$ in cluster $c$, $C$ the number of clusters and $\mathrm{df}_t$ the number of clusters where $t$ appears. The label gathers the terms with the highest weight: frequent in the cluster, rare in the others. With name_dynamic_clusters = FALSE, the comparison is made within each graph, so clusters of different windows may get the same label.

Most central node. "tidygraph_functions" computes the centrality given in tidygraph_function on each graph (it is added as a node column) and labels each cluster with the node with the highest value. The centrality is computed on the whole graph, not within the cluster. This method needs a network structure and therefore cannot be used with name_dynamic_clusters = TRUE.

Given column. "given_column" labels each cluster with its node with the highest order_by value: for an article network, typically its most cited article.

Nodes of an unnamed cluster get "no_name".

References

Ding Y (2011). “Community Detection: Topological vs. Topical.” Journal of Informetrics, 5(4), 498–514. doi:10.1016/j.joi.2011.02.006 .

Salton G, Buckley C (1988). “Term-Weighting Approaches in Automatic Text Retrieval.” Information Processing & Management, 24(5), 513–523.

Spärck Jones K (1972). “A Statistical Interpretation of Term Specificity and Its Application in Retrieval.” Journal of Documentation, 28(1), 11–21.

Examples