Domain Term Extraction Based on Conditional Random Fields Combined with Active Learning Strategy

摘要

Chinese domain term extraction is an important task in Chinese information processing, which has been used in lexicography, ontology construction and so on. This paper presents a Chinese automobile term extraction system based on CRFs (Conditional Random Fields) and active learning. Seven kinds of features are selected in the CRFs model and the experimental result shows that the precision, recall and F-score are 84.61%, 80.50% and 82.50% respectively with 5-fold cross-validation. As the supervised machine learning method needs large-scale labeled corpus and it is expensive to label corpus manually, we integrate active learning into the CRFs model. The active learning method uses the uncertainty-based sampling strategy and the experimental results show that when the percentage of the training corpus on the whole unlabeled corpus becomes 80%, the F-score is 82.49%, which is almost the same as that (82.50%) with the percentage of 100%.