engleski

Reasoning about Large-Scale Social Networks with Probabilistic Logic

Network Matching and Link Prediction are relatively unexplored in the area of Social network analysis, but solving those problems in an efficient way is crucial in many real‐world applications. Network Matching is a generalized problem of node identification. Node identification (matching individuals) is a task of unique identification a person in the analyzed network as a known entity in existing network, based on the known links and additional attributes. In the Link Prediction (social) graph is build or completed by inferencing links based on existing network's structure and node attributes. Both problems in the most realworld applications have to deal with incomplete information and probabilities. In perfect information environment, those problems would be naturally modelled in the predicate logic, hence, the real‐ world problems require methods of probabilistic logic ("ProbLog" framework). Two large‐scale social networks were used to develop and test the model: (1) the sample of the largest social network consisting of 372 volunteers with over 1M links ; (2) huge social network generated from all available Croatian public records with 540.000 individuals and over 100 million links among them. First network was matched with the second using developed model, with completeness of 86, 8% (323 individuals). Results are evaluated against matched volunteers with an error of 5, 3%. Probabalistic logic link prediction model was applied on a second network with promising results.

Probabilistic Logic; Social Networks Analysis

nije evidentirano