Reliable normalization in resume information extraction
Jonathan Medema · Utrecht University Repository (Utrecht University) · 2008
In this thesis the subject of reliable normalization in Information Extraction is discussed, specifically for the normalization of extracted items out of people's resumees. A meta-classifier approach is presented, which is based on the Memory Based Learning (MBL) implementation of the k-Nearest Neigbor Algorithm. To investigate whether this approach is a practical solution to reliable normalization a literature study is done on the subject and various experiments on domain-specific data sets were conducted. It is shown that the meta-classifier approach in combination with the MBL algorithm is one of the fastest implementations in comparison to other reliability measures and classifier algorithms. The meta-classifier is able to reach an F2-score of 86.1% on randomly created test sets.