CORPUS MAPS

After performing a lexical extraction on the corpus, we used CorText Manager ([doc]) to produce corpus maps applying the tool's clustering and layout algorithms.

Several maps were produced, with either approx. 150 nodes or approx. 250 nodes. The maps are accessible from the Corpus maps menu.

In terms of the lexical items exploited to create the maps, two general types of maps were created:

• The maps that have the word terms in their name (e.g. 150 terms) are based on DBpedia concept mentions, identified thanks to the procedure described here.
• The maps that have the word multiword in their name (e.g. 250 multiword) are based on keyphrases, which were extracted as described here.

Regarding layout and navigation functions, the different types of maps created are outlined in following:

• Static maps
  The complete corpus is considered at once, leaving aside the temporal dimension (i.e. each document's date)
• Tagged clusters maps: Each cluster has been tagged with two terms that represent the content of the cluster.
  CorText does select these terms automatically, but for some clusters the terms have been edited manually, based on the terms suggested by CorText.
  The map is based on a gexf file produced by CorText, edited with Gephi and post-processed with Inkscape to add the cluster tags.
• Navigable maps: The nodes and layout in the maps are the same as in the tagged clusters ones. But they have been produced with two tools that create a searchable and zoomable map:
  • Gephi's Sigma JS exporter plugin.
  • The TINA project's TinawebJS tool, which is also deployed as the MapExplorer view in CorText V2.


• Dynamic maps
These maps are "dynamic" in the sense that they represent how terms found in the corpus evolve over time.
The visualizations have been created with CorText Manager's temporal analysis functions.
The visualization format is a Sankey diagram or river chart, representing the evolution of clusters of terms in the corpus. The tool obtains two representative terms for each cluster at each period. These terms are used to label the nodes of the chart.
Besides, the tool assesses whether a cluster of terms is stable, splits into other clusters or merges with other clusters, based on the terms shared across different clusters across time.
How to read the maps
• Hovering over a node displays the list of terms for that node.
• Hovering over an arc connecting two nodes displays the list of terms shared by both nodes.


• Heatmaps
The heatmaps represent which clusters are focused on in a given subset of the corpus' documents. These subsets can be created according to criteria like document type, author, geographical origin, etc.
Our heatmaps have been created by dividing the corpus into decades. Each heatmap shows the clusters (or areas in the corpus map) that are salient in each decade.
The difference between heatmaps and the dynamic maps also available on this site is that the dynamic maps trace the evolution of a lexical cluster: whether the cluster is stable across time, whether it is split into others, or whether its terms merge with those in other clusters.
How to read the maps
• The Basemap tab represents the clusters in the corpus regardless of decade.
• The remaining tabs display the heatmap for the decade starting with the year mentioned on the tab.
Differences between the networks in the heatmaps and those in maps elsewhere on this site ([a], [b]) are due to a slightly different set of terms chosen to create the networks with. The same clusters are present in all types of networks.
The heatmaps are created by CorText Manager, but cluster names were added with Inkscape to CorText's PDF export.
The fact that nodes in the heatmaps are represented with triangles, whereas in all other maps they are represented with circles, does not have any special meaning.
The way the heatmaps are displayed on the site is inspired by this work.