SVM Based Learning System for Information Extraction

Abstract. We present an SVM-based learning algorithm for information extraction, including experiments on the influence of different algorithm settings. Our approach needs fewer SVM classifiers to be trained than other recently proposed SVM-based systems. Another distinctive feature is the use of a variant of the SVM, the SVM with uneven margins, which is particularly helpful for mixed-initiative (adaptive) information extraction. We also compare our system to other state of the art systems (including rule learning and statistical learning algorithms) on three IE benchmark datasets: CoNLL-2003, the CMU seminars corpus, and the software jobs corpus. The experimental results showed that our system had a compatible performance. It outperformed a recent SVM system, achieved the highest scores on eight out of 17 categories on the jobs corpus, and was second best on the remaining nine.

• Type: Conference
• Year: 2004
• Where: DSMML
• Authors: Yaoyong Li, Kalina Bontcheva, Hamish Cunningham
• Keywords: DSMML 2004, Fewer Svm Classifiers, Jobs Corpus, Learning Algorithms, Machine Learning