Spreading activation

Abstract

Centroids are comfortable instruments to represent queries and whole texts by single descriptive terms. They can be used to determine the similarity of textual contents and to (hierarchically) cluster sets of documents. However, their computation strictly following the concept's definition may use a plenty of time and hinder any practical application. A more demonstrative view on the meaning and topological interpretation of the definition leads to the derivation of a graph-based algorithm using the well-known spreading activation technique, which is described in this contribution. The experimental results obtained using co-occurrence graphs of varying sizes underline the high performance of this method which is -last but not least- brought about by its clear local working principle.