Representing the Data as a Network

A network, analogous to a mathematical graph, is a tool often used in data science to represent complex interconected data. In the broadest sense a network consists of a set of nodes and edges, where often the nodes represent entities within some system and edges correspond to relations between these entities. Two examples are given below, where the left/right show undirected/directed networks.

Networks are used to represent data in a diverse range of settings, examples include:

• Sociology - social networks and physical interaction networks;
• Biology - protein-protein interaction networks;
• Neuroscience/Psychology - brain connectiviy networks, known as connectomes;
• Operations Research - transportation networks or networks of qeues.

In a similar vein, we are going to make use of a network representation to analyse Elsevier's online usage data. In this case nodes will correspond to platforms/products and edges will represent movements between products. If we then look to compare users based on this data, we will be left with a network for each user. This leaves us with multiple network data, a situation becoming more prevelant in the wider literature thanks in most part to improvements in data collection technologies.

Given this network representation, our research will be focussed on the development and implementation of network models which are applicable to the problem at hand. Certain features of the data provide new challenges which traditional network models often fail to meet, hence novel approaches must be developed.

Project Challenges and Aims

As alluded to above, this project throws up certain challenges which will require novel approaches. These include:

• Volume of data, which itself brings forth two challenges,
  • Multiple network data, since usage data is available for individual users;
  • Computation, particularly when conducting inference for formulated models.
• Temporal aspect of the data - often usage data comes with timestamps, which can be informative of network dynamics.

In the face of these challenges the project has the following initial aims:

• Develop a statistical model for multiple network data, applicable to the Elsevier's usage data. This model should:
• Respect features of the data;
• Facilitate informative inference;
• Be as flexible as possible.
• Develop effective inference procedures for the proposed model, noting that with users in the millions these procedures would ideally be scaleable.