Community Research and Development Information Service - CORDIS

The analysis of networks

An EU team has developed new network analysis methods in support of e-commerce. New, scalable probabilistic models apply primarily to product recommender systems, but also to networking in several broad fields.
The analysis of networks
Networks affect much of modern life and in fact all life. Networks are greatly studied in many contexts, but often using rather limited Bayesian parametric methods.

The EU-funded BNPNET (Bayesian nonparametric methods for networks and recommender systems) project developed and applied new Bayesian nonparametric methods to the topic. Specifically, the work addressed e-commerce, where most buyers purchase only a few of many available items. The scalable new models were applied to the probabilistic modelling of large networks and e-commerce recommender systems. However, the work also applies to all aspects of networking, including biological and social sciences and information technology.

Researchers developed new classes of statistical models for networks. The models illustrate the power-law aspect of real networks, whereby the majority of buyers purchase just a few items. The models also include interpretable parameters, providing an improved understanding of network structure. The BNPNET models have well-understood large-scale properties. Also, algorithms for learning the model parameters are also available.

The main result involved the associated class of models producing 'sparse' graphs. The term means that the number of network connections is smaller than the number of possible connections, as in real situations. Testing proved that the new models were able to learn and evaluate the level of sparsity in a wide range of real-world network contexts. The models were also able to make more accurate predictions for such graphs.

BNPNET developed new, scalable networking models. These were developed to make e-commerce product recommender applications more effective.

Related information


Networks, e-commerce, recommender systems, Bayesian, parametric methods, BNPNET
Follow us on: RSS Facebook Twitter YouTube Managed by the EU Publications Office Top