Project Member:
Nick Gramsky - ngramsky {at}

General Objective:

To visualize the various kinds of relationships, or ties, via color in link-node diagrams.

Network attributes vary over time. Analyzing and visualizing these temporal changes is quite difficult. Current methods adequately visualize static networks and capture standard metrics such as degree between nodes and reciprocity among others. However the quantification, classification and visualization of how relationships change over time has yet to be mastered. A single image of a network that captures both the structure and benefits of a standard link-node diagram with the added feature of showing the historical evolution of all relationships in at least one dimension would greatly contribute to network analysis. Understand what dyads are strengthening and weakening in respect to reciprocity or general interactions over an extended period of time can help identify a shift in key actors or a change in interaction patterns or within a network. The methods discussed here should transcend across any social network, be it an email, Twitter, co-authorship or any social network.


This project will cover two dimensions of relationships:
- Variance over Time
The level of interactions between nodes within a network can vary over time. In a simplistic setting interactions can remain constant, increase in duration or decrease in duration over time.
- Reciprocation
The level of directional interaction can vary between nodes. Nodes can share who initiates the number of interactions (bi-directional), only one node can initiate all interactions (one-way), or they can fall somewhere in-between.

Utilizing a matrix to help classify relationships and identify where they fall might look like this:




This combination of frequency over time and reciprocating relationships as a classification has not thoroughly been visualized or tested when analyzing social networks yet could lead to novel network analysis.

Consider the following network as motivation:

This network topology has nodes sized according to degree, edges are sized according to number of interactions and arrows show direction. Let's assume this visualization covers a 2-year period. Observing this it is unclear how consistent relationships are. Edges with similar sizes show they have the same number of interactions, yet it is unclear if one relationship is more stable than another. Additionally reciprocity can be accounted for by sizing up the arrow-heads, but classifying all even bi-directional relationships takes time as all edges must be evaulated in order to do so.

Consider the same network with the following color scheme applied (green - consistent bi-direction relationships red - bi-directional relationships with decreasing activity black - one-way relationships with decreasing activity blue - consitent one-way relationships)
It is immediately clear what relationships are classified according to the matrix above. Searching is minimized and analysis can proceed much quicker.

Specific Objective:

This project aims to:
- define how to classify relationships by establishing methods for classifying where the relationship fall in the above matrix
- effectively visualize them
- evaluate the success of doing so

Technical Details:

Gather Data Sets that show social interaction over time. Data sets include Facebook activity, Twitter activity or vast email collections. Data mining programs/scripts will need to be written to categorize the relationships.Polynomial interpolation or other forms of curve fitting may be used to classify the temporal variation of frequency. Methods to classify reciprocation have yet to be determined.

Possible extensions to NodeXL will be considered in order to visualize the network. Possible control panels may include:

The above window shows a potential visualization control for NodeXL. Users can decide if they want to color relationships based on varying levels of reciprocity or general interaction.

User study at the conclusion of the study could validate the effectiveness of the methodology/scheme.

Progress Report



Final Report

Comment (Ben):
This looks like a good idea - with one caveat. I am not *sure* this hasn't been done. So you should start with a literature review and gain confidence about what has and hasn't been done in this area. One related bit you should look at is the work on "Augmented Social Graph" at CMU by Hong and others -

Also, stay in touch with your fellow student John who is working on the one other visualization project for this class (