Abstract:

The conditional probability models gain great developments these years. The conditional models gradually took place of generative models in sequence labeling

problems. It covers a wide range of applications, such as image recognition, natural language processing, intrusion detection and other issues. Conditional Random Fields is representative of conditional models and becomes one of the most popular models, for it not only overcomes the shortcomings of generative models but also defeats the label bias problem of Maximum Entropy Model. That’s why it’s very popular. But when CRFs is used for specific applications, it’s found that the results may not achieve the best. So in every specific application some improvements are made except for the CRFs model itself. The research includes military commands segmentation, military named entity recognition, ntrusion detection, etc. All these specifications are made on the basis of the CRFs model and the system performances are greatly improved.

Key words: conditional random fields, sequence labeling, Chinese word segmentation, named entity, intrusion detection, layered framework