Ontology Generation from Social Networks

Analysis of social network data is gaining popularity with the increased availability of real-world data including data publicly available over the Internet such as publications and data resulting from interactions via social networking platforms and e-communication tools. In this chapter we present an approach to constructing light-weight ontologies from social network data. In relation to the more traditional (semi-)automatic ontology learning techniques we reuse the approach typically used in learning ontologies from text (see Grobelnik and Mladenić(2006) for details) where the candidate instances and classes for the ontology are lexical items described by a set of attributes. We replace lexical items with nodes in the social network and attributes by descriptions of the node context in the graph. Similar techniques can then be applied for ontology learning either from text or from social networks.

To prove our claims we perform experiment on a real life dataset taken from a mid-size organization (700–800 people). The dataset represents log files from organizational spam filter software giving us the set of e-mail transactions for a period of 19 months resulting in 2.7 million successful e-mail transactions used here for analysis and ontology learning.

The main contribution of this work is an architecture consisting of five major steps that enable transformation of the data from a set of e-mail transactions inside an organization to an ontology representing the structure of the organization.