Information Extraction via Double Classification

Abstract

Information Extraction is concerned with extracting relevant information from a (collection of) documents. We propose an approach consisting of two classification-based machine learning loops. In a first loop we look for the relevant sentences in a document. In the second loop, we perform a word-level classification. We test the system on the Software Jobs corpus and we do an extensive evaluation in which we discuss the influence of the di#erent parameters. Furthermore we show that the type of evaluation method has an important influence on the results.