Linkage

There are a number of pre-defined measures that are implemented on events, utilising different attributes as the basis of an identifier graph, such as the profiling_device_identifier, device_signature, ip_network_prefix.

These metrics are computed periodically (over the course of several hours) and updated to a node in the identity graph. The value associated with an event is that of the last re-calculation and is therefore may not include certain recent interactions.

Linkage Metrics Currently Implemented

Node Strength

E.g. identity[ACCOUNT].login.score.linkage_node_strength
This is a one-hop metric, i.e., it only considers its nearest neighbours in a network. It indicates a strength of the node utilizing number of immediate connections and its correlations with those connections. This is different than node-degree which just counts the number of immediate connections.

H-index

E.g. identity[ACCOUNT].login.score.linkage_h_index
This is a 2-hop metric, i.e. it also looks into the properties of neighbours of all of its immediate neighbours. It finds out a largest integer k such that its k neighbours has at least k neighbours.

Lobby Index

E.g. identity[ACCOUNT].login.score.linkage_lobby_index
This is also a 2-hop metric. It is a natural extension of H-index in the sense that it also incorporates the strength of edges/connections among all the nodes in the network.

PageRank

E.g. identity[ACCOUNT].login.score.linkage_page_rank
This is a metric for strongly connected sub-graph of a network. First we obtain all the nodes connected to a given node for all levels. Then the algorithm computes a probability of arriving to that node starting from any node in a network using random walk. This algorithm has been implemented in Networkx.