Network Analysis

NETWORK ANALYSIS RESOURCES

Network analysis allows you to explore and visualize relationships in your data. For humanists, this most often means social networks of some kind, whether musicians and patrons in medieval Venice or retweets on Twitter.

GETTING STARTED

Desktop opensource solutions (intermediate/advanced)
Gephi, Cytoscape

Web-based solutions (collaborative, classroom-geared)
Net.Create (ask for more info)

Programming solutions (advanced)
Python (with package networkx)

R (with package igraph)

Networks (or 'graphs') are comprised of nodes, which are also referred to as 'vertices,' and edges, which are also referred to as 'links,' 'connections,' or 'ties.' Nodes represent individuals, groups, or concepts, while edges represent the relationships between nodes that structure the network.

Edges can be directed or undirected. Friendships, collaborations, and other reciprocal interactions are undirected. Social media follows, citations, or journeys to a particular place are directed.
Edges are also weighted by their intensity or frequency. A greater edge weight signals more robust or frequent interactions.
A network as a whole can be unimodal--comprised of one class of nodes--or bi/multimodal, comprised of two or more classes of nodes. For example, a bimodal network is comprised of students and clubs, while a unimodal network is only comprised of students.

An ego network is a network graph focused around a single node (the ego), the nodes to which it is connected (the alters), and the relationships between those alters. Ego networks are useful for reconstructing the networks of specific individuals or a precise segment of a population.

Certain other attributes of a node or network may be of interest depending on your research question(s):

Centrality refers to the importance of a specific node to the overall network. There are four common measures of centrality:
- Degree centrality refers to the number of links a node has to other nodes.
  - In a directed network, you can measure the in-degree centrality (how many edges are directed to the node in question) and the out-degree centrality (how many edges are directed from the node in question) separately.
  - Degree distribution measures the number of nodes of a specific degree in a network.
- Closeness Centrality refers to how close a node is to all other nodes in a network.
- Betweenness Centrality refers to how often a node serves as the shortest point of connection (a bridge) for otherwise disconnected parts of the network.
- Eigenvector centrality refers to how connected a node is to other individually central nodes.
Density measures how interconnected a network is. It refers to the ratio of actual (or realized) edges in a network to all possible edges.
Average path length and network diameter measure how easy it is to move from one side of the network to the other.
- Average path length refers to the shortest average path between all possible pairs of nodes in a network. A network with a low average path length exhibits few degrees of separation between the majority of nodes contained within it.
- Network diameter refers to the longest of all shortest paths between two nodes. A network with a high network diameter includes nodes that are only connected to one another by many degrees of separation.
Clustering refers to the extent to which a node's connections are connected to each other. One example of a node with significant clustering is a member of a clique, or a node whose immediate neighbors are all connected to one another.
- Cliquishness is measured by the local or individual clustering coefficient, which is the ratio of actual connections between a node's neighbors to the total possible connections.
  - The global clustering coefficient is the average of all local coefficients in the network. A network with a high global clustering coefficient is very close-knit.
  - A small world network is characterized by a high clustering coefficient and short average path length. This means that the network is comprised of multiple closely knit cliques, where any one node has fairly immediate access to any other node.
- You can also measure clustering by the ratio of transitive triples (or triangles) in the network, trios of nodes where each node is connected to each of the other two.
Communities in a network are groupings of nodes that have higher connectivity between one another than would be expected at random for the network overall. In Python and R, specialized libraries can be used to detect communities in a network:
- igraph for R.
- cdlib for Python.

Musicians in Venice
http://musiciansinvenice.com/dissertation/

Networks of Performance and Patronage: Russian Artists in American Dance, Vaudeville, and Opera, 1909-1947
http://networksalarusse.com/network/

Florentine Marriages: A Use Case in R
https://cran.r-project.org/web/packages/netrankr/vignettes/use_case.html

The Making of a Pope
https://arxiv.org/html/2505.17635v1#S6

PROCESS OVERVIEW

Network analysis almost always begins with collecting and structuring your data in a format that one of the software tools can read. At IDAH, we mainly support the open source software Gephi. Getting your data ready for this and other software could like something like this:

Identify what kind of relationships you have. In your data, decide what type or types of entities you will be connecting. (The entities are referred to as nodes and the connections edges.) Are there different types of nodes in your data (i.e. individuals and institutions)? Is the connection a simple or two-way one (undirected), or does one node initiate the connection (directed)? How will you label these nodes and what other information might you want to include about either the nodes or the edges?

Structure your data in spreadsheet form. A software like Gephi will ultimately want two related spreadsheets, one with a list of nodes and one with a list edges. But if you are digging through source material to compile this information, often it is easier to build just the edges sheet. At a minimum this will mean filling in a "source" column and a "target" column (those terms are used for any kind of network graph, but they have a special meaning when the graph is a directed one). If there is a large amount of data you need to add for each node, then you may want to build these sheets simultaneously. Ultimately, Gephi will want numbered IDs for each node and ID relationships in the edge table. This is easy enough to generate in Excel after you've added your data, so long as you've been careful to make sure that each node name is unique.

Load your data into a network analysis software and analyze. See tutorials below for how to do this in Gephi and for recommendations on other tool options.

TUTORIALS AND OTHER RESOURCES

Gephi needs two spreadsheets: nodes and edges. Nodes refers to the individual entities in your network—the points. Edges are the relationships between the nodes.

The nodes sheet should contain:

Label	ID	NodeType (optional)	Other (optional)
Sarah Smith	1	scholar
Plos One	2	journal

The edges sheet should contain:

Source*	SourceLabel	Target	TargetLabel	Other (optional)
1	Sarah Smith	2	Plos One

*If there is directionality in the relationships, this should be reflected in where entities appear in the "Source" or "Target" columns. If there is no directionality (Gephi calls this "undirected"), then it does not matter in which column an entity is placed, only that there is a relationship.

You can easily create the two spreadsheets that Gephi requires from a single spreadsheet of relationships. At minimum, the spreadsheet you start with needs to contain one column for one entity and another column for the entity it relates to. In this scenario, each row is one instance of a connection (i.e. a scholar publishes an article in a journal). Using Excel, your next steps could look something like this:

Create a new sheet called "nodes." Rename the original sheet "edges."
From "edges" copy both columns of the connected entities paste them into the single column "Label" in "nodes," then run "Remove Duplicates" to make sure each entity appears only once. (Be sure to copy over any additional information you need for each node, such as node type, dates, etc.)
Create IDs for each of the nodes in the "ID" column.
In "edges" rename existing columns and create new ones to correspond to the above formatting.
Use VLOOKUP to insert the node ID for each node when it appears in "SourceLabel" or "TargetLabel" in "edges."
Once all the IDs are imported, do a copy and "paste as value" for both "Source" and "Target."
If all of the data and IDs seem to be present in both sheets, then you are ready to load these sheets into Gephi. You can export each sheet as a separate CSV file, or load the Excel file as is and select the appropriate sheets in Gephi.

Open Gephi and select "New Project."
In the top left, select "Data Laboratory"
Select "Import Spreadsheet" and select your nodes CSV or Excel file.
Make sure the sheet is being imported appropriately (as either nodes or edges), then click "Next."
Click "Finish."
Change "Graph Type" to the direction type (most often "Undirected").
Select "Append to existing workspace," then click "OK."
Repeat steps 3–7 for your edges file.
In top left, selected "Overview" to view and analyze your network.

Once your network data is loaded into Gephi, here is an overview of the most common things you might want to do with it. All of these options are visible in the "Overview" environment, which is represented by a globe at the top left of your display. Note: the windows to the right and left of the graph can be (re)opened from the 'Window' tab in the title bar.

Appearance (top left, under the menu and title bars)

Top left of the window: Select to style either nodes or edges
Top right of the window: Select for color, size (nodes only), label color, label size
Unique: Set a hardcoded color or size
Partition: Set based on an attribute within your data. Partitioning based on modularity allows you to visualize communities in your data.
Ranking: Set based on a number range. You may be interested in varying node size or color by degree or a centrality measure (see Statistics section below).

Layout (bottom left)

Gephi implements multiple layout algorithms, which set the shape of the graph. Algorithms like Force Atlas, Yifan Hu, and Fruchterman Reingold emphasize complementarities within the network. OpenOrd instead emphasizes divisions in the network.
We recommend that you begin by running Force Atlas, and then run Yifan Hu Proportional.
From there, additional layout algorithms can be used to increase the graph's legibility:
- Contraction and Expansion contract or expand the graph (by an adjustable scale factor).
- LabelAdjust ensures that labels are readable and not overlapping.
- Noverlap prevents nodes from overlapping.
If you want to start from scratch, Random Layout resets the graph.

Left-Side-of-Graph-Screen Controls

The tools here allow for selection and editing. You can hover over each icon to see its name and function. However, a few tools are worth highlighting here:
- 'Drag' (represented by a hand) is used to reposition specific nodes.
- 'Edit' (a cursor and question mark) is used to edit the properties of specific nodes (size, color, etc.) and to view attributes like centrality.
- 'Shortest path' (a triangle) highlights the most efficient path between two nodes of interest.
- 'Heat map' (several blobs) is used to show distance from a node of interest.
- 'Center on graph' (a magnifying glass) recenters the graph within the window.

Bottom-of-Graph-Screen Controls

Selection and editing tools are also located at the bottom of the graph window.
- The bold "T" toggles node labels on and off, while the light "T" toggles edge labels on and off.
- The two sliders adjust the edge weight scale and font size scale--changing the thickness of the edges and size of the labels, respectively.
Gephi also allows researchs to export their graphs as static images. The camera at the bottom left of the graph window saves a screenshot of the current graph as a PNG file.
- Note: To obtain a high-resolution image, use the dropdown menu to configure the screenshot settings. Reset the image size to 3000 x 2000 and antialiasing to 16x (the maximum).
- For more information about exporting static images or interactive web apges, see Visualizing a Network Dataset Using Gephi: https://mdl.library.utoronto.ca/technology/tutorials/visualizing-network-dataset-using-gephi

Statistics (right)

Gephi can also generate a variety of network statistics, including centrality measures, network diameter, and modularity (see Key Terms and Concepts above for definitions).
- Running a statistic generates a downloadable report, updates node attributes, and update the Data Laboratory (see below).

Filters (right)

Researchers can also use filters to hide specific nodes and edges before exporting the graph as a static image.
- For example, researchers may want to filter by degree range (under Topology) in order to hide nodes that only have one edge.
- Researchers may also want to filter by the ego network (under Topology) of a specific node.
- Finally, researchers can partition the network by selecting a filtering category from Attributes > Partition.
Filters can be used independently or nested. The basic procedure is:
- Drag the desired filter into Queries, select settings, and click "Filter."
- Select "Stop" to turn filter off.

Data Laboratory

You can view the nodes and edges tables by selecting "Data Laboratory" at the top left. You can also right click on a node and select "Select in data laboratory" to view that node's record in the nodes table.

Download Gephi here: https://gephi.org/.

Find tutorials and examples here: https://gephi.org/users/.

Gephi also comes with a few sets of sample data that can be loaded on start up.

Network Analysis