University of Konstanz
Graduiertenkolleg / PhD Program
Computer and Information Science

Graduation Talks


Spreading Activation on Bisociative Networks


Kilian Thiel, University Konstanz
Konstanz, Germany

date & place

Wednesday, 03.06.2009, 16:15 h
Room C 252


Spreading activation techniques have been designed to query networks. They allow relevant subgraphs, nodes or edges to be extracted according to a given query. For this purpose specific nodes of a network are activated, this activation is spread iteratively to adjacent nodes until the process is terminated. The basic idea behind spreading activation is that relevant information can be retrieved by considering the relations between concepts, that are either known to be relevant, or speci?ed by a user.
First of all spreading activation methods have been applied to semantic networks, which consist usually of concepts represented by nodes and the relations between them represented by edges. Similar to semantic networks, Bisociative Networks (BisoNets) represent concepts and their relations. In contrast to semantic networks not only semantically meaningful information but also loosely coupled information fragments can be integrated into BisoNets. The underlying data structure of a BisoNet is a k-partite graph, and each partition represents a domain or area of knowledge. Every concept or unit of information is assigned to one partition. The aim of building a BisoNet is to link information of several different knowledge domains, in order to explore this information in a holistic way and to ?nd hopefully useful and previously unknown information. Due to the graph structure of BisoNets, spreading activation techniques offer a promising method to query these networks, allowing speci?c answers, more general concepts, and correlations, or bisociations according to a given query to be found.
A formal analysis of spreading activation methods is the topic of the proposed thesis. The description of the spreading activation framework is given in the following section.