Link prediction models

All type of measurements and information collections are suffering from incompleteness due to lack of resources or they are error prone due to finite accuracy. It is especially true for social interactions, where the large number of different relations and the restricted possibilities to observe and track them result in an incomplete dataset. Therefore, applications can utilize methods that are able to predict the existence of unrecovered relations or provide a ranking score for missing links recommending an importance order for targeted investigations or give hints for tracking complex information spreading at LEAs.
The D3.2 deliverable, “Link prediction models” aims to implement link prediction models for Social Network Analysis. The deliverable is the result of Task 3.2 in the Social Network Analysis (WP3) work package. The SNA work package delivers a solution for processing and analysing networked data coming from data pre-processing components. The output of D3.2 will be an input of visualization (delivered in D5.4). The predicted links will be labelled with a reliability score, which can be utilized in Machine Learning (delivered in D2.2).
The link prediction solution adopts network theoretic similarity and distance measures for the special purposes of the RED-Alert solution. Based on the special targeted measures, missing links and nodes are predicted by the module. Furthermore, some features e.g. weights, labels, directionality of the links are updated as well.
The report explains the key outcomes that are delivered by this task, the software solutions, installation instructions and some scaling estimations of computing resources.

D3 2 Link prediction models-FINAL