Improving discriminative sequential learning with rare--but--important associations

Discriminative sequential learning models like Conditional Random Fields (CRFs) have achieved significant success in several areas such as natural language processing, information extraction, and computational biology. Their key advantage is the ability to capture various non?independent and overlapping features of inputs. However, there are several unexpected pitfalls influencing negatively on model's performance that mainly come from the imbalance among classes/labels, the irregular phenomena, and the ambiguity potentially existing in the training data. This paper presents a data?driven approach that can deal with such hard?to?predict data instances by discovering and emphasizing rare?but?important associations of statistics hidden in the training data. Mined associations are then incorporated into these models in a couple of ways to boost difficult examples. The experimental results of English phrase chunking and named entity recognition using CRFs show a significant improveme...

• Type: Conference
• Year: 2005
• Where: KDD
• Authors: Xuan Hieu Phan, Minh Le Nguyen, Tu Bao Ho, Susumu Horiguchi
• Keywords: Data Mining, Discriminative Sequential Learning, English Phrase Chunking, KDD 2005, Sequential Learning Models