Towards a Framework for Learning from Networked Data

摘要

Over the past decades, one has seen databases of ever increasing size and complexity. While the increasing size is easy to measure in bytes, kilobytes or terabytes, the increase in complexity is more difficult to quantify, however, it has a very deep effect on the theory we use to reason about the data. While in earlier days many researchers reasoned in terms of sets of similarly structured and independent objects, today we are facing large networks of data where everything is connected directly or indirectly to everything else. Examples include social networks, traffic networks, biological networks, administrative networks and economic networks. These developments have spurred a renewed interest in data storage and knowledge extraction (answers to queries, patterns, models, ...). Three key underlying challenges are the representation of the data and knowledge, managing the computational cost of the problems which we need to solve and the statistical challenge related to the complexity of the data. In this contribution, I will survey these challenges from a data mining point of view. I will argue that in order to address the current challenges it is valuable to gain a better understanding of fundamental statistical and algorithmic properties of large data networks and to integrate ideas from the many fields of research that are concerned with such networks.