Analyzing data with visual methods helps you gain better insight about complexity. Whether you dig in a leaked database, investigate intermingled interactions of an ecosystem, manage your networked organization, or curate a large archive, you start making sense of a complex issue by mapping its actors and relations. As a fundamentally human thought process, mapping helps us navigate particular links among the actors while seeing the patterns in the bigger picture, and gain insight throughout this journey. We’ve been tailoring interfaces and processes to make the analysis experience as intuitive as possible!
Centrality and clustering are the core network analysis metrics.
You start analyzing a network map by examining its centrality and clustering metrics. The network organizes itself by a physics-based simulation between neighbor nodes, pulling and pushing each other like springs. This layout organization process reveals the central and peripheral actors, indirect links, organic clusters, bridging nodes and outliers that you wouldn’t see otherwise.
While browsing a network map, you visually recognize the most connected nodes from its lines in and out, which establish specific connections between parts of the image, while discounting other ones. Font and circle size indicate the relative importance of each node. You notice the clusters of tightly interconnected nodes. The bridging nodes between two or more clusters become distinctly visible. However, when a network map gets larger, the high level of detail overwhelms our senses. To be able to precisely examine and compare such qualities, you need more quantitative views of the data contained in the graph.
The current graph interface on Graph Commons provides a continuous experience of switching from the particular (a specific node and its immediate relations) to the general (seeing the bigger network) and back again. We think this cycle helps you create a useful frame of reference in your mind to digest the complexity. In order to support this qualitative experience with quantitative methods, we provide a feature that we simply call "Analysis".
When you open a graph, you simply want to understand what is it about. A few things you can do:
- Check out the graph title on the top bar
- View the Info panel and check the types of nodes and relationships among them
Info Panel displays a graph's summary: title, authors, dates, and the types, color, and quantity of nodes and relationships (mouse over relationship types to view directions and the node types it connects).
- Mouse over some of the nodes and view their immediate connections
- Click on nodes to view the details in their cards
Node Card displays a node's name, type, description, custom properties, and relationships. Also editing and sharing options.
Now that you have a summary, go on and mouse over some of the items on the graph to dive deeper into the data network you are viewing. As you randomly navigate a graph, you might already find yourself discovering some insights. Check out some tips for free-discovery here.
After the initial exploration of a graph, you want to dig deeper to discover the patterns within the network. You can start asking these questions to your graph:
- Who are the central actors or influencers in this network?
- What are the main focus areas or clusters?
- Who are the bridging actors between separate clusters?
- What are the similarities, differences and short paths in the network?
Analysis bar with Clustering, Centrality, and Similarity modules
A common analysis task on networks is to identify clusters based on connectivity patterns. The idea is to simply find groups that have more connections to one another than they do to outsiders. In the study of networks, we use clustering to recognize communities within large groups of connections.
Typically, a force-directed layout algorithm organizes a network map, makes patterns visually comprehensible, but it cannot identify and mark the clusters. Furthermore, in large network maps, the high level of detail overwhelms our senses. To be able to precisely examine its patterns, we need quantitative views of the data contained in the network. While there are a variety of data clustering methods in machine learning, the Louvain Modularity algorithm works well particularly for large data-networks. It detects tightly knit groups characterized by a relatively high density of ties.
Clustering analysis module with option to select resolution
Clusters are labelled and highlighted
- Use the Clustering function in the Analysis bar to identify clusters in your graph. In the clustering analysis dialogue, select the resolution of clustering algorithm. If you want less and generalized clusters, then decrease the resolution towards left, if your want more and specialized clusters then increase the resolution towards right.
- When clusters are detected, they are automatically labeled based on the most connected node in the cluster. We strongly recommend you to edit clusters and rename yourself to highlight what they really mean. You will see the clusters have percentages and the number of nodes in that cluster, that indicates the precise size of the cluster in the overall network.
- To highlight a particular cluster and hide others, simply mouse over the cluster labels on the graph. Also, you can double click on a cluster to hide others. Finally, you can move the positions of cluster labels for more readability.
- Sometimes you graph may have many isolated islands, you can hide them for more readability using the adjustments in the clustering module.
- Within the clusters you will see few nodes making connections to other clusters besides their neighboring nodes, whose connections are only local, within their immediate cluster. Those are bridging nodes between two or more clusters. Sometimes they are boundry spanners from a periphery of a cluster to another cluster. Bridges have critical roles: they are the ones who carry the information from one group to another; they are in a position to combine variety of knowledge and ideas found in multiple groups; they have more likelihood of being a single point of failure that if a bridge disappears, those formerly connected communities would disconnect.
{% hint style="info" %} Rerun the clustering process multiple times with different resolutions to find the right clustering in your graph (remove and add again to rerun it). {% endhint %}
Sharing cluster-view of your network map often generates the most feedback, because your team or community may not be familiar with the particular data points you have, but when they see the cluster labels, it makes sense and they start engaging with your work.
In a graph some nodes may have relatively more important positions than others. In some situations, important nodes may be defined as being central to the network when they have many connections, or as having a bridging role between two communities. Bridging nodes can be important because their removal may break the network into parts or they become too powerful as they are the broker of the information flow between the communities.
Centrality analysis dialogue options
Centrality metrics on the analysis bar
On the Analysis bar, you can list the most central nodes by either:
- a numeric property (e.g., Companies by valuation)
- a network centrality metric (e.g., Investors by number of companies, or betweenness centrality)
These lists provide a summary of the most important nodes in a graph. By clicking on a node in the list, you see where it is located in the network along with its highlighted connections.
Two nodes in a network are considered "similar" if they share many common neighbors. When you open the similarity analysis dialogue, you will see options for selecting node types and their particular relationship types to generate a similarity analysis.
Similarity analysis dialogue options
Similarity metrics on the analysis bar
Similarity module displays list of nodes that are similar in two metrics: intersection and similarity. Intersection, is how many shared connections these two nodes have. Similarity is a Jaccard similarity coefficient that is taking the ratio of link intersection (shared edges) over union (including not shared edges). You can sort them by intersection or similarity, and click on them highlight on the graph. You will notice similar nodes are positioned next to each other on the graph, thanks to our layout algorithm, but without similary module, they are not so easy to spot on large graphs.