Mining and Modeling Complex Relational Data with Unnamed Actors

3 different network layouts for the same data

Mining and Modeling Complex Relational Data with Unnamed Actors

Through text mining to extract named and unnamed entities and social network visualization to illustrate their relationships, we can represent unnamed women’s presence in the socio-political network despite their absence in the archival record.

By examining quantitative measures of the social network, we learn more about women’s positions within the structure of Ottoman-Algerian society.

Through an analysis of the individual lives, relationships and the underlying structures that make up the Ottoman-Algerian network in Constantine between 1567 and 1837, I argue that Algerian women were essential intercultural mediators and conduits to power.


Mining Entities and Relationships

Three strategies

  1. Simple named entity recognition (NER) using Stanford CoreNLP with a French language package.
  2. Referenced individuals: those who were referenced in the text but unnamed (e.g. ‘the daughter of Ahmed’). Done manually for this project, as this remains an unsolved computational challenges. See my current work on this task.
  3. Inferenced individuals: if there was a parent-child relationship, we added the second parent. Also done manually.

Significance: This is not merely a numbers game. The more people we can accurately represent in the graph, the better sense we have for the structure of relationships and the relative positions of men and women of various ethnicities in the society under consideration. It is this structure that we can meaningfully explore with network analysis.

GenderNamedReferenced (% = Named + Referenced)Inference
(% is cumulative)
Row Totals
Column Totals1193825182
Raw numbers and proportions of men and women in the three social network graphs
Network Analysis

Methods to assess individual actor importance in a social network graph:

  • Degree centrality
  • Betweenness centrality: Ulrik Brandes, “A Faster Algorithm for Betweenness Centrality.” The Journal of Mathematical Sociology 25, no. 2 (June 1, 2001): 163–77.
  • Harmonic Closeness Centrality: Used Brandes’ algorithm for this measure, but the results were inconclusive due to the high number of triangles in the network.
Ongoing research focuses on the following methods
  • Eigenvector centrality
  • Transitivity
  • Cutpoints

R-script of recent work and analysis using the methods above.

Key Findings

Kinship connections can be meaningfully investigated using quantitative network metrics. Betweenness centrality scores are particularly informative because they highlight the individuals who served as essential social bridges between people, family units, and socio-political cliques. Technically, betweenness centrality is a measure of the number of shortest paths that travel through a node.[i] A path is a sequence of edges in the graph in which all nodes and edges are distinct. The length of a path is the number of edges on it.

Portion of Ottoman-Constantine social network with nodes sized by betweenness centrality and the top individuals labeled.

Therefore, those with the highest rankings forged and maintained links between Ottoman officials and local families, bolstering Ottoman sovereignty in this frontier province. Imperial officers depended on these connections both to govern effectively and for their own safety and security while in office. Of the top individuals, ranked by betweenness centrality, 12 of the 26 shown (46 percent) are women, a proportion higher than the general proportion of women to men in the graph (38 percent). Of these twelve women, half are unnamed but explicitly referenced in the documents.

Individuals ranked by betweenness centrality as calculated using Ulrik Brandes’ Betweenness Centrality algorithm and graphed with Tableau.

Data and Code

Data: Sanders, Ashley, and Veronica Dean. “Social Network of Ottoman Constantine, Algeria,
1567-1837,” Open Science Framework, May 16, 2023.

Code: Ongoing research in R

Leave a Reply

Your email address will not be published. Required fields are marked *

Social Media Auto Publish Powered By :