Extraction of failure graphs from structured and unstructured data

Abstract

Quality analysis in the automotive domain is up to now mainly focused on structured data obtained from repair visits, using for example association rules or decision trees on model families, model years and damage codes. This work will outline a way to extract failure graphs from textual repair orders using taxonomy based concept recognition, significant co-occurrences and graph clustering methods. We will furthermore combine unstructured data with structured data and demonstrate the benefits of this method for root cause analysis in the automotive domain. © 2008 IEEE.