MAC, Montag, 27.06. 2011, 17:00 Uhr, F380 

Stefan Betzmeir & Ute Schmid: Prototype Extraction and Classification for Trees and an Application to Incident Mining

In a cooperation with SAP we developed an approach to classify incident reports. We were given a set of incident clusters which were constructed by a domain expert. Our algorithm constructs a prototype for each cluster and a new incident is classified by similarity to prototype. Incidents are given as XML files, i.e. as trees. Instead of using the standard approach of feature extraction from trees and applying feature-based methods for classification learning, we propose an algorithm for constructing structural prototypes, i.e. for learning on tree structures. In the talk we will focus on the generic problem of aligning sets of arbitrary trees and present our preliminary results.