Chinese organization name recognition based on co-training algorithm

Organization name recognition is the most difficult part in named entity recognition, in order to reduce the use of tagged corpus and use a large amount of untagged corpus, we firstly present using semi-supervised machine learning algorithm co-training combining with conditional random fields model and support vector machines on Chinese organization name recognition. Based on the principles of compatible and uncorrelated, we construct different classifiers from different views of conditional random fields model, and also construct different classifiers from two models of conditional random fields model and support vector machines as two views. Then present a heuristic untagged samples selection algorithm. From the experimental results we can see that, under the same F-measure, co-training algorithm simply use about 30% of the tagged data compared to single statistical model; under the same tagged data, co-training algorithm has an F-measure increase about 10% than single statistical model.